By many estimates, quantum computers will need millions of qubits to realize their potential applications in cybersecurity, drug development, and other industries. The problem is, anyone who has wanted to simultaneously control millions of a certain kind of qubits has run into the problem of trying to control millions of laser beams.

That’s exactly the challenge that was faced by scientists working on the MITRE Quantum Moonshot project, which brought together scientists from MITRE, MIT, the University of Colorado at Boulder, and Sandia National Laboratories. The solution they developed came in the form of an image projection technology that they realized could also be the fix for a host of other challenges in augmented reality, biomedical imaging, and elsewhere. The device is a one-square-millimeter photonic chip capable of projecting the Mona Lisa onto an area smaller than the size of two human egg cells.

“When we started, we certainly never would have anticipated that we would be making a technology that might revolutionize imaging,” says Matt Eichenfield, one of the leaders of the Quantum Moonshot project, a collaborative research effort focused on developing a scalable diamond-based quantum computer, and a professor of quantum engineering at the University of Colorado at Boulder. Each second, their chip is capable of projecting 68.6 million individual spots of light—called scannable pixels to differentiate them from physical pixels. That’s more than fifty times the capability of previous technology, such as micro-electromechanical systems (MEMS) micromirror arrays.

“We have now made a scannable pixel that is at the absolute limit of what diffraction allows,” says Henry Wen, a visiting researcher at MIT and a photonics engineer at QuEra Computing.

The chip’s distinguishing feature is an array of tiny micro-scale cantilevers, which curve away from the plane of the chip in response to voltage and act as miniature “ski-jumps” for light. Light is channeled along the length of each cantilever via a waveguide, and exits at its tip. The cantilevers contain a thin layer of aluminum nitride, a piezoelectric which expands or contracts under voltage, thus moving the micromachine up and down and enabling the array to scan beams of light over a two-dimensional area.

Despite the magnitude of the team’s achievement, Eichenfield says that the process of engineering the cantilevers was “pretty smooth.” Each cantilever is composed of a stack of several submicrometer layers of material and curls approximately 90 degrees out of the plane at rest. To achieve such a high curvature, the team took advantage of differences in the contraction and expansion of individual layers caused by physical stresses in the material resulting from the fabrication process. The materials are first deposited flat onto the chip. Then, a layer in the chip below the cantilever is removed, allowing the material stresses to take effect, releasing the cantilever from the chip and allowing it to curl out. The top layer of each cantilever also features a series of silicon dioxide bars running perpendicular to the waveguide, which keep the cantilever from curling along its width while also improving its length-wise curvature.

A micro-cantilever wiggles and waggles to project light in the right place.Matt Saha, Y. Henry Wen, et al.

What was more of a challenge than engineering the chip itself was figuring out the details of actually making the chip project images and videos. Working out the process of synchronizing and timing the cantilevers’ motion and light beams to generate the right colors at the right time was a substantial effort, according to Andy Greenspon, a researcher at MITRE who also worked on the project. Now, the team has successfully projected a variety of videos from a single cantilever, including clips from the movie A Charlie Brown Christmas.

The chip projected a roughly 125-micrometer image of the Mona Lisa.Matt Saha, Y. Henry Wen, et al.

Because the chip can project so many more spots in any given time interval than any previous beam scanners, it could also be used to control many more qubits in quantum computers. The Quantum Moonshot program’s mission is to build a quantum computer that can be scaled to millions of qubits. So clearly, it needs a scalable way of controlling each one, explains Wen. Instead of using one laser per qubit, the team realized that not every qubit needed to be controlled at every given moment. The chip’s ability to move light beams over a two-dimensional area, would allow them to control all of the qubits with many fewer lasers.

Another process that Wen thinks the chip could improve is scanning objects for 3D printing. Today, that typically involves using a single laser to scan over the entire surface of an object. The new chip, however, could potentially employ thousands of laser beams. “I think now you can take a process that would have taken hours and maybe bring it down to minutes,” says Wen.

Wen is also excited to explore the potential of different cantilever shapes. By changing the orientations of the bars perpendicular to the waveguide, the team has been able to make the cantilevers curl into helixes. Wen says that such unusual shapes could be useful in making a lab-on-a-chip for cell biology or drug development. “A lot of this stuff is imaging, scanning a laser across something, either to image it or to stimulate some response. And so we could have one of these ski jumps curl not just up, but actually curl back around, and then move around and scan over a sample,” Wen explains. “If you can imagine a structure that will be useful for you, we should try it.”

Large language models have already transformed software engineering, for better or worse. Now, so-called large physics models are also starting to transform design engineering. These tools are beginning to replace—or at least, amend—the role of full-fledged physics simulation in the automotive and aerospace industries, semiconductor engineering, and more.

Before the advent of computer simulation, a car manufacturer, for example, would create prototypes to test their designs, says Thomas Von Tschammer, managing director at physics-based AI company Neural Concept. “For the past 40 years, we reduced a lot of the need for prototypes by using numerical simulations for aerodynamics, for crash testing, and so on.” Now, Tschammer explains, AI is drastically reducing the need for simulation, the same way simulation reduced the need for physical prototypes.

Growing adoptions of this type of AI was a topic of interest at Nvidia GTC in March. Chris Johnston, senior technical specialist at Jaguar Land Rover, presented how his company is using Neural Concept’s technology. PhysicsX, another physics-based AI company, announced a collaboration with Nvidia to advance open standards for such models, also at GTC.

The AI design engineering workflow

Over the past six months, General Motors (GM) has introduced large physics models into their car design process to speed up the workflow.

Previously, a creative design engineer would develop a 3D model of a new car concept. This model would be sent to aerodynamics specialists, who would run physics simulations to determine the coefficient of drag of the proposed car—an important metric for energy efficiency of the vehicle. This simulation phase would take about two weeks, and the aerodynamics engineer would then report the drag coefficient back to the creative designer, possibly with suggested modifications.

Now, GM has trained an in-house large physics model on those simulation results. The AI takes in a 3D car model and outputs a coefficient of drag in a matter of minutes. “We have experts in the aerodynamics and the creative studio now who can sit together and iterate instantly to make decisions [about] our future products,” says Rene Strauss, director of virtual integration engineering at GM.

For GM and other companies, running inference on an AI model trained on physics simulations, instead of running the simulation itself, can bring immense time savings. “Depending on the kinds of physics [being simulated], or the resolution, it can be anywhere between 10,000 to close to a million times faster,” says Jacomo Corbo, CEO and co-founder of PhysicsX.

How accurate are large physics models?

But, what about accuracy? For GM’s purposes, Strauss says accuracy is not a huge concern at the design stage because finer details are ironed out later in the process. “When it really starts to matter is when we’re getting close to launching a vehicle, and the coefficient of drag is going to be used for our energy calculation, which eventually goes to the certification of our miles per gallon on the sticker.” At that stage, Strauss says, a physical model of the car will be put into a wind tunnel for an exact number.

PhysicsX’s Corbo argues that, with the right data, the AI model accuracy can supersede the accuracy of the simulation it’s trained on. The trick is to incorporate experimental measurements to fine-tune the model. If a physics simulation doesn’t agree exactly with experimental data, it is often difficult to figure out why and tweak the model until they agree. With AI, incorporating a few experimental examples into the training process is a lot more straightforward, and it’s not necessary to understand where exactly the model went wrong.

All in all, by drastically bringing down the time it takes to model the physics, large physics models enable engineers to explore a much greater range of possibilities before a final design is reached.

Training large physics models

There is no one-size-fits-all approach to training large physics models. Depending on the types of data available, and the physics in question, the models may use the transformer architecture that underlies LLMs, a generalized version of convolutional neural networks known as geometric deep learning, or an architecture that can solve partial differential equations called neural operators.

Currently, most companies are training their own models on their simulation data, catering to specific use cases. In GM’s aerodynamics implementation, there are different AI models for different types of cars: think SUVs versus sedans. But PhysicsX’s Corbo says his team is working on building more “foundational” physics models that can be applied across different scenarios.

Both LLMs and robotics have benefitted from scaling laws, which describe how a system improves as the models increase in size or get trained on more data. In AI, models tend to improve quickly, in a non-linear way. Along the way, the models also become more generalizable—extending them to new settings takes less and less fine-tuning to reach the same accuracy. Corbo says his team is now starting to see the same types of scaling laws for large physics models.

“What we’re seeing here is maybe a little bit unsurprising,” Corbo says, “but it’s also pretty incredible. And it’s given us the confidence to make these models bigger, because they perform a whole lot better, and they cover broader domains, and they have these really amazing emergent properties.”

Developing open standards for the data formats used in training, as well as the model architectures, should help develop these more powerful foundational models. That’s the goal of PhysicsX’s collaboration with Nvidia, and of Nvidia’s physicsNeMo open source platform.

“The thing that we’re collaborating on is being able to compose architectures from building blocks,” Corbo says, making it easy for those in both academia and industry to re-use and build upon existing models.

A type of AI called a large physics model is used by an to quickly generate heat flow in a 3D data center server design. Neural Concept

The long-term role of simulations and engineers

While some are working on developing more powerful models, others are pushing to implement what’s already available into existing workflows, which is no easy task. “With any innovation, it’s not a straight line. There’s some steps forward and then some steps back and improvements that we find along the way. But that’s part of the joy of the innovation process and using new tools like this,” GM’s Strauss says.

This technology is still in the early stages, and it’s unclear what the final role of AI tools will be in the engineering workflow. For one, opinions vary on whether AI will replace simulations completely, or just reduce their use.

“We will never fully replace simulations,” Neural Concept’s von Tschammer says. “But the idea is to make a much smarter usage of simulation at the most major phase of developments, and you use AI to speed up the early design stages, where you need to explore a very wide set of options.”

PhysicsX’s Corbo begs to differ. “The whole idea is to take numerical simulation … out of the workflow,” he says, “and to move that to inference.”

Whatever the role of simulation will be, everyone in the field is adamant that human design engineers will continue to be in the driver’s seat, enabled to do their best work by these newfangled tools. (After all, when has AI ever threatened to replace human labor?)

“What we’re seeing is that actually, these tools are empowering the engineers to be much more efficient,” Tschammer says. “Before, these engineers would spend a lot of time on low added value tasks, whereas now these manual tasks from the past can be automated using these AI models, and the engineers can focus on taking the design decisions at the end of the day. We still need engineers more than ever.”

Artificial intelligence harbors an enormous energy appetite. Such constant cravings are evident in the hefty carbon footprint of the data centers behind the AI boom and the steady increase over time of carbon emissions from training frontier AI models.

No wonder big tech companies are warming up to nuclear energy, envisioning a future fueled by reliable, carbon-free sources. But while nuclear-powered data centers might still be years away, some in the research and industry spheres are taking action right now to curb AI’s growing energy demands. They’re tackling training as one of the most energy-intensive phases in a model’s life cycle, focusing their efforts on decentralization.

Decentralization allocates model training across a network of independent nodes rather than relying on one platform or provider. It allows compute to go where the energy is—be it a dormant server sitting in a research lab or a computer in a solar-powered home. Instead of constructing more data centers that require electric grids to scale up their infrastructure and capacity, decentralization harnesses energy from existing sources, avoiding adding more power into the mix.

Hardware in harmony

Training AI models is a huge data center sport, synchronized across clusters of closely connected GPUs. But as hardware improvements struggle to keep up with the swift rise in size of large language models, even massive single data centers are no longer cutting it.

Tech firms are turning to the pooled power of multiple data centers—no matter their location. Nvidia, for instance, launched the Spectrum-XGS Ethernet for scale-across networking, which “can deliver the performance needed for large-scale single job AI training and inference across geographically separated data centers.” Similarly, Cisco introduced its 8223 router designed to “connect geographically dispersed AI clusters.”

Other companies are harvesting idle compute in servers, sparking the emergence of a GPU-as-a-Service business model. Take Akash Network, a peer-to-peer cloud computing marketplace that bills itself as the “Airbnb for data centers.” Those with unused or underused GPUs in offices and smaller data centers register as providers, while those in need of computing power are considered as tenants who can choose among providers and rent their GPUs.

“If you look at [AI] training today, it’s very dependent on the latest and greatest GPUs,” says Akash cofounder and CEO Greg Osuri. “The world is transitioning, fortunately, from only relying on large, high-density GPUs to now considering smaller GPUs.”

Software in sync

In addition to orchestrating the hardware, decentralized AI training also requires algorithmic changes on the software side. This is where federated learning, a form of distributed machine learning, comes in.

It starts with an initial version of a global AI model housed in a trusted entity such as a central server. The server distributes the model to participating organizations, which train it locally on their data and share only the model weights with the trusted entity, explains Lalana Kagal, a principal research scientist at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) who leads the Decentralized Information Group. The trusted entity then aggregates the weights, often by averaging them, integrates them into the global model, and sends the updated model back to the participants. This collaborative training cycle repeats until the model is considered fully trained.

But there are drawbacks to distributing both data and computation. The constant back and forth exchanges of model weights, for instance, result in high communication costs. Fault tolerance is another issue.

“A big thing about AI is that every training step is not fault-tolerant,” Osuri says. “That means if one node goes down, you have to restore the whole batch again.”

To overcome these hurdles, researchers at Google DeepMind developed DiLoCo, a distributed low-communication optimization algorithm. DiLoCo forms what Google DeepMind research scientist Arthur Douillard calls “islands of compute,” where each island consists of a group of chips. Every island holds a different chip type, but chips within an island must be of the same type. Islands are decoupled from each other, and synchronizing knowledge between them happens once in a while. This decoupling means islands can perform training steps independently without communicating as often, and chips can fail without having to interrupt the remaining healthy chips. However, the team’s experiments found diminishing performance after eight islands.

An improved version dubbed Streaming DiLoCo further reduces the bandwidth requirement by synchronizing knowledge “in a streaming fashion across several steps and without stopping for communicating,” says Douillard. The mechanism is akin to watching a video even if it hasn’t been fully downloaded yet. “In Streaming DiLoCo, as you do computational work, the knowledge is being synchronized gradually in the background,” he adds.

AI development platform Prime Intellect implemented a variant of the DiLoCo algorithm as a vital component of its 10-billion-parameter INTELLECT-1 model trained across five countries spanning three continents. Upping the ante, 0G Labs, makers of a decentralized AI operating system, adapted DiLoCo to train a 107-billion-parameter foundation model under a network of segregated clusters with limited bandwidth. Meanwhile, popular open-source deep learning framework PyTorch included DiLoCo in its repository of fault tolerance techniques.

“A lot of engineering has been done by the community to take our DiLoCo paper and integrate it in a system learning over consumer-grade internet,” Douillard says. “I’m very excited to see my research being useful.”

A more energy-efficient way to train AI

With hardware and software enhancements in place, decentralized AI training is primed to help solve AI’s energy problem. This approach offers the option of training models “in a cheaper, more resource-efficient, more energy-efficient way,” says MIT CSAIL’s Kagal.

And while Douillard admits that “training methods like DiLoCo are arguably more complex, they provide an interesting tradeoff of system efficiency.” For instance, you can now use data centers across far apart locations without needing to build ultrafast bandwidth in between. Douillard adds that fault tolerance is baked in because “the blast radius of a chip failing is limited to its island of compute.”

Even better, companies can take advantage of existing underutilized processing capacity rather than continuously building new energy-hungry data centers. Betting big on such an opportunity, Akash created its Starcluster program. One of the program’s aims involves tapping into solar-powered homes and employing the desktops and laptops within them to train AI models. “We want to convert your home into a fully functional data center,” Osuri says.

Osuri acknowledges that participating in Starcluster will not be trivial. Beyond solar panels and devices equipped with consumer-grade GPUs, participants would also need to invest in batteries for backup power and redundant internet to prevent downtime. The Starcluster program is figuring out ways to package all these aspects together and make it easier for homeowners, including collaborating with industry partners to subsidize battery costs.

Backend work is already underway to enable homes to participate as providers in the Akash Network, and the team hopes to reach its target by 2027. The Starcluster program also envisions expanding into other solar-powered locations, such as schools and local community sites.

Decentralized AI training holds much promise to steer AI toward a more environmentally sustainable future. For Osuri, such potential lies in moving AI “to where the energy is instead of moving the energy to where AI is.”

This year marks the 80th anniversary of ENIAC, the first general-purpose digital computer. The computer was built during World War II to speed up ballistics calculations, but its contributions to computing extend well beyond military applications.

Two of ENIAC’s key architects—John W. Mauchly, its co-inventor, and Kathleen “Kay” McNulty, one of the six original programmers—married a few years after its completion and raised seven children together. Mauchly and McNulty’s grandchild Naomi Most delivered a talk as part of a celebration in honor of ENIAC’s anniversary on 15 February, which was held online and in-person at the American Helicopter Museum in West Chester, Pa. The following is adapted from that presentation.

RELATED: ENIAC, the First General-Purpose Digital Computer, Turns 80

There was a library at my grandparents’ farmhouse that felt like it went on forever. September light through the windows, beech leaves rustling outside on the stone porch, the sounds of cousins and aunts and uncles somewhere in the house. And in the corner of that library, an IBM personal computer.

When I spent summers there as a child, I didn’t yet know that the computer was closely tied to my family’s story.

My grandparents are known for their contributions to creating the Electronic Numerical Integrator and Computer, or ENIAC. But both were interested in more than just crunching numbers: My grandfather wanted to predict the weather. My grandmother wanted to be a good storyteller.

In Irish, the first language my grandmother Kathleen “Kay” McNulty ever spoke, a word existed to describe both of these impulses: ríomh.

I began to learn the Irish language myself five years ago, and I was struck by how certain words and phrases had multiple meanings. According to renowned Irish cultural historian Manchán Magan—from whom I took lessons—the word ríomh has at different times been used to mean to compute, but also to weave, to narrate, or to compose a poem. That one word that can tell the story of ENIAC, a machine with wires woven like thread that was built to compute, make predictions, and search for a signal in the noise.

John Mauchly’s Weather-Prediction Ambitions

Before working on ENIAC, John Mauchly spent years collecting rainfall data across the United States. His favorite pastime was meteorology, and he wanted to find patterns in storm systems to predict the weather.

The Army, however, funded ENIAC to make simpler predictions: calculating ballistic trajectory tables. Start there, co-inventors J. Presper Eckert and Mauchly realized, and perhaps the weather would soon be computable.

Co-inventors John Mauchly [left] and J. Presper Eckert look at a portion of ENIAC on 25 November 1966. Hulton Archive/Getty Images

Weather is a system unfolding through time, and a model of a storm is a story about how that system might unfold. There’s an old Irish saying related to this idea: Is maith an scéalaí an aimsir. Literally, “weather is a good storyteller.” But aimsir also means time. So the usual translation of this phrase into English becomes “time will tell.”

Mauchly wanted to ríomh an aimsire—to weave the weather into pattern, to compute the storm, to narrate the chaos. He realized that complex systems don’t reveal their full purpose at conception. They reveal it through aimsir—through weather, through time, through use.

ENIAC’s First Programmers Were Weavers

Kathleen “Kay” McNulty was born on 12 February 1921, in Creeslough, Ireland, on the night her father—an IRA training officer—was arrested and imprisoned in Derry Gaol.

Family oral history holds that her people were weavers. She spoke only Irish until her family reached Philadelphia when she was 4 years old, entering American school the following year knowing virtually no English. She graduated in 1942 from Chestnut Hill College with a mathematics degree, was recruited to compute artillery firing tables by hand for the U.S. Army, and was then selected—along with five other women—to program ENIAC.

They had no manual. They had only blueprints.

McNulty and her colleagues learned ENIAC and its quirks the way you learn a loom: by touch, by memory, by routing threads of electricity into patterns. They developed embodied knowledge the designers could only approximate. They could narrow a malfunction to a specific failed vacuum tube before any technician could locate it.

McNulty and Mauchly are also credited with conceiving the subroutine, the sequence of instructions that can be repeatedly recalled to perform a task, now essential in any programming. The subroutine was not in ENIAC’s blueprints, nor in the funding proposal. The concept emerged as highly determined people extended their imagination into the machine’s affordances.

The engineers designed the loom. Weavers discovered its true capabilities.

In 1950, four years after ENIAC was switched on, Mauchly’s dream was realized as it was used in the world’s first computer-assisted weather forecast. That was made possible after Klara von Neumann and Nick Metropolis reassembled and upgraded the ENIAC with a small amount of digital program memory. The programmers who transformed the math into operational code for the ENIAC were Norma Gilbarg, Ellen-Kristine Eliassen, and Margaret Smagorinsky. Their names are not as well-known as they should be.

Before programming ENIAC, Kay McNulty [left] was recruited by the U.S. Army to compute artillery firing tables. Here, she and two other women, Alyse Snyder [center] and Sis Stump, operate a mechanical analog computer designed to solve differential equations in the basement of the University of Pennsylvania’s Moore School of Electrical Engineering.University of Pennsylvania

Kay McNulty, Family Storyteller

Kay married John Mauchly in 1948, describing him as “the greatest delight of my life. He was so intelligent and had so many ideas.... He was not only lovable, he was loving.” She spent the rest of her life ensuring he, Eckert, and the ENIAC programmers would be recognized.

When she died in 2006, I came to her funeral in shock, not fully knowing what I’d lost. As she drifted away, it was said, she had been reciting her prayers in Irish. This understanding made it quickly over to Creeslough, in County Donegal, and awaited me when I visited to honor her memory with the dedication of a plaque right there in the center of town.

In her own memoir, she wrote: “If I am remembered at all, I would like to be remembered as my family storyteller.”

In Irish, the word for computer is ríomhaire. One who ríomhs. One who weaves, computes, and tells. My grandfather wanted to tell the story of the weather through computing. My grandmother wanted to be remembered as a storyteller. The language of her childhood already had a word that contained both of those ambitions.

Computers as Narrative Engines

When it was built, ENIAC looked like the back room of a textile production house. Panels. Switchboards. A room full of wires. Thread.

Thread does not tell you what it will become. We tend to think of computing as calculation—discrete and deterministic. But a model is a structured story about how something behaves.

Weather models, ballistic tables, economic forecasts, neural networks: These are all narrative engines, systems that take raw inputs and produce accounts of how the world might unfold. In complex systems, when parts are woven together through use, new structures arise that no one specified in advance.

Like ENIAC, the machines we are building now—the large models, the autonomous systems—are not merely calculators. They are looms.

Their most important properties will not be specified in advance. They will emerge through use, through the people who learn how to weave with them.

Through imagination.

Through aimsir.

A traditional data center protects the expensive hardware inside it with a “shell” constructed from steel and concrete. Constructing a data center’s shell is inexpensive compared to the cost of the hardware and infrastructure inside it, but it’s not trivial. It takes time for engineers to consider potential sites, apply for permits, and coordinate with construction contractors.

That’s a problem for those looking to quickly deploy AI hardware, which has led companies like Duos Edge AI and LG CNS to respond with a more modular approach. They use pre-fabricated, self-contained boxes that can be deployed in months instead of years. The boxes can operate alone or in tandem with others, providing the option to add more if required.

“I just came back from Nvidia’s GTC, and a lot of [companies] are sitting on their deployment because their data centers aren’t ready, or they can’t find the space,” said Doug Recker, CEO of Duos Edge AI. “We see the demand there, and we can deploy faster.”

GPUs shipped straight to you

Duos Edge AI’s modular compute pods are 55 feet long and 12.5 feet wide. Though they look similar to a shipping container, they’re actually a bit larger and designed primarily for transportation by truck. Each compute pod contains racks of GPUs much like those used in other data centers. Duos recently entered a deal with AI infrastructure company Hydra Host to deploy four pods with 576 GPUs per pod. That’s a total of 2,304 GPUs, with the option to later double the deployment to 4,608 GPUs.

Modular data centers aren’t new for Duos; the company previously deployed edge data centers for rural customers, such as the Amarillo, Texas school district. However, the pods for the Hydra Host deployment will be upgraded to handle more intense AI workloads. They’ll contain more racks, draw more power, and use liquid cooling to keep the GPUs running efficiently.

Across the Pacific, Korean technology giant LG is taking a similar approach. The company’s CNS subsidiary, which provides IT infrastructure and services, has announced the AI Modular Data Center which, like the Duos unit, contains racks of GPUs and supporting hardware in a pre-fabricated enclosure.

Also like Duos’ deployment, LG’s AI Modular Data Center contains 576 Nvidia GPUs with the option to scale up in the future. “We are currently developing an expanded version that can support more than 4,600 GPUs within a single unit, with a service launch planned within this year,” said Heon Hyeock Cho, vice president and head of the datacenter business unit at LG CNS. LG’s first Modular Data Center will roll out in the South Korean port city of Busan, where it could deploy up to 50 units.

LG and Duos are not alone. Hewlett Packard Enterprise, Vertiv, and Schneider Electric now have modular data centers available or in development. A report from market research firm Grand View Research estimates that the market for modular data centers could more than double by 2030.

On the grid, but under the radar

A modular data center site is quite different from traditional data center because there’s no need to construct a large steel-and-concrete shell. Instead, the site can be made ready by pouring a concrete pad. The pre-fabricated modules are delivered by truck, placed on the pad where desired, and then networked on-site.

Duos’ deployments, for instance, include power modules placed alongside the compute pods, and the pods are networked together with redundant fiber connections that allow the pods to operate in unison. Recker compared it to lining up school buses in a parking lot. “Everything is built off-site at a factory, and we can put it together like a jigsaw puzzle,” he said.

That simplicity is the point. Both Duos and LG CNS expect a modular data center can be deployed in about six months, compared to the roughly two or three years a conventional data center requires. Recker said that, for Duos, the turnaround is so quick that building the pre-fabricated unit isn’t always the constraint. While it’s possible to construct a pre-fabricated unit in 60 or 90 days, site preparation extends the timeline “because you can’t get the permits that fast.”

Modular data centers may also provide good value. Recker said a five-megawatt modular deployment can be built for about $25 million, and that Duos’ cost per megawatt is roughly half what larger facilities charge. For Duos, savings are possible in part because its modular data centers can target smaller deployments where the permitting is less complex. Smaller, modular deployments also meet less resistance from local governments, which are increasingly skeptical about data center construction.

While Duos targets smaller deployments, LG hopes to go big. Its planned Busan campus of 50 AI Modular Data Centers suggests an ambition to achieve deployments that rival the capacity of conventional facilities. A site with 50 units would bring the total number of GPUs to over 28,000. Here, the benefits of a modular approach could stem mostly from scalability, as a modular data center could start small and grow as required.

“By adopting a modular approach, the AI Modular Data Center can be incrementally expanded through the combination of dozens of AI Boxes,” Cho said. “It’s enabling the construction of even hyperscale-level AI data centers.”

Facial recognition technology (FRT) dates back 60 years. Just over a decade ago, deep-learning methods tipped the technology into more useful—and menacing—territory. Now, retailers, your neighbors, and law enforcement are all storing your face and building up a fragmentary photo album of your life.

Yet the story those photos can tell inevitably has errors. FRT makers, like those of any diagnostic technology, must balance two types of errors: false positives and false negatives. There are three possible outcomes.

In best-case scenarios—such as comparing someone’s passport photo to a photo taken by a border agent—false-negative rates are around two in 1,000 and false positives are less than one in 1 million.

In the rare event you’re one of those false negatives, a border agent might ask you to show your passport and take a second look at your face. But as people ask more of the technology, more ambitious applications could lead to more catastrophic errors. Let’s say that police are searching for a suspect, and they’re comparing an image taken with a security camera with a previous “mug shot” of the suspect.

Training-data composition, differences in how sensors detect faces, and intrinsic differences between groups, such as age, all affect an algorithm’s performance. The United Kingdom estimated that its FRT exposed some groups, such as women and darker-skinned people, to risks of misidentification as high as two orders of magnitude greater than it did to others.

Less clear photographs are harder for FRT to process.iStock

What happens with photos of people who aren’t cooperating, or vendors that train algorithms on biased datasets, or field agents who demand a swift match from a huge dataset? Here, things get murky.

Consider a busy trade fair using FRT to check attendees against a database, or gallery, of images of the 10,000 registrants, for example. Even at 99.9 percent accuracy you’ll get about a dozen false positives or negatives, which may be worth the trade-off to the fair organizers. But if police start using something like that across a city of 1 million people, the number of potential victims of mistaken identity rises, as do the stakes.

What if we ask FRT to tell us if the government has ever recorded and stored an image of a given person? That’s what U.S. Immigration and Customs Enforcement agents have done since June 2025, using the Mobile Fortify app. The agency conducted more than 100,000 FRT searches in the first six months. The size of the potential gallery is at least 1.2 billion images.

At that size, assuming even best-case images, the system is likely to return around 1 million false matches, but at a rate at least 10 times as high for darker-skinned people, depending on the subgroup.

Responsible use of this powerful technology would involve independent identity checks, multiple sources of data, and a clear understanding of the error thresholds, says computer scientist Erik Learned-Miller of the University of Massachusetts Amherst: “The care we take in deploying such systems should be proportional to the stakes.”

This sponsored article is brought to you by NYU Tandon School of Engineering.

Within a 6 mile radius of New York University’s (NYU) campus, there are more than 500 tech industry giants, banks, and hospitals. This isn’t just a fact about real estate, it’s the foundation for advancing quantum discovery and application.

While the world races to harness quantum technology, NYU is betting that the ultimate advantage lies not solely in a lab, but in the dense, demanding, and hyper-connected urban ecosystem that surrounds it. With the launch of its NYU Quantum Institute (NYUQI), NYU is positioning itself as the central node in this network; a “full stack” powerhouse built on the conviction that it has found the right place, and the right time, to turn quantum science into tangible reality.

Proximity advantage is essential because quantum science demands it. Globally, the quest for practical quantum solutions — whether for computing, sensing, or secure communications — has been stalled, in part, by fragmentation. Physicists and chemical engineers invent new materials, computer scientists develop new algorithms, and electrical engineers build new devices, but all three often work in isolated academic silos.

Gregory Gabadadze, NYU’s dean for science, NYU physicist and Quantum Institute Director Javad Shabani, and Juan de Pablo, Anne and Joel Ehrenkranz Executive Vice President for Global Science and Technology and executive dean of the Tandon School of Engineering.Veselin Cuparić/NYU

NYUQI’s premise is that breakthroughs happen “at the interfaces between different domains,” according to Juan de Pablo, Executive Vice President for Global Science and Technology at NYU and Executive Dean of the NYU Tandon School of Engineering. The Institute is built to actively force those necessary collisions — to integrate the physicists, engineers, materials scientists, computer scientists, biologists, and chemists vital to quantum research into one holistic operation. This institutional design ensures that the hardware built by one team can be immediately tested by software developed by another, accelerating progress in a way that isolated departments never could.

NYUQI’s premise is that breakthroughs happen at the interfaces between different domains. —Juan de Pablo, NYU Tandon School of Engineering

NYUQI’s integrated vision is backed by a massive physical commitment to the city. The NYUQI is not just a theoretical concept; its collaborators will be housed in a renovated, million-square-foot facility in the heart of Manhattan’s West Village, backed by a state-of-the-art Nanofabrication Cleanroom in Brooklyn serving as a high-tech foundry. This is where the theoretical meets physical devices, allowing the Institute to test and refine the process from materials science to deployment.

NYUQI will be housed in a renovated, million-square-foot facility in the heart of Manhattan’s West Village.Tracey Friedman/NYU

Leading this effort is NYUQI Director Javad Shabani, who, along with the other members, is turning the Institute into a hub for collaboration with private and public sector partners with quantum challenges that need solving. As de Pablo explains, “Anybody who wants to work on quantum with NYU, you come in through that door, and we’ll send you to the right place.” For New York’s vast ecosystem of tech giants and financial institutions, the NYUQI offers a resource they can’t build on their own: a cohesive team of experts in quantum phenomena, quantum information theory, communication, computing, materials, and optics, and a structured path to applying theoretical discoveries to advanced quantum technologies.

Solving the Challenge of Quantum Research

The NYUQI’s integrated structure is less about organizational management, and more about scientific requirement. The challenge of quantum is that the hardware, the software, and the programming are inherently interconnected — each must be designed to work with the other. To solve this, the Institute focuses on three applications of quantum science: Quantum Computing, Quantum Sensing, and Quantum Communications.

For Shabani, this means creating an integrated environment that bridges discovery with experimentation, starting with the physical components all the way to quantum algorithm centers. That will include a fabrication facility in the new building in Manhattan, as well as the NYU Nanofab in Brooklyn directed by Davood Shahjerdi. New York Senators Charles Schumer and Kirsten Gillibrand recently secured $1 million in congressionally-directed spending to bring Thermal Laser Epitaxy (TLE) technology — which allows for atomic-level purity, minimal defects, and streamlined application of a diverse range of quantum materials — to NYU, marking the first time the equipment will be used in the U.S.

NYU Nanofab manager Smiti Bhattacharya and Nanofab Director Davood Shahjerdi at the nanofab ribbon-cutting in 2023. The nanofab is the first academic cleanroom in Brooklyn, and serves as a prototyping facility for the NORDTECH Microelectronics Commons consortium.NYU WIRELESS

Tight control over fabrication, and can allow researchers to pivot quickly when a breakthrough in one area — say, finding a cheaper, more reliable material like silicon carbide — can be explored for use across all three applications, and offers unique access to academics and the private sector alike to sophisticated pieces of specialty equipment whose maintenance knowledge and costs make them all-but-impossible to maintain outside of the right staffing and environment.

The NYU Nanofab is Brooklyn’s first academic cleanroom, with a strategic focus on superconducting quantum technologies, advanced semiconductor electronics, and devices built from quantum heterostructures and other next-generation materials.NYU Nanofab

That speed and adaptability is the NYUQI’s competitive edge. It turns fragmented challenges into holistic solutions, positioning the Institute to solve real-world problems for its New York neighbors—from highly secure data transmission to next-generation drug discovery.

Testing Quantum Communication in NYC

The integrated approach also makes the NYUQI a testbed for the most critical near-term applications. Take Quantum Communications, which is essential for creating an “unhackable” quantum internet. In an industry first, NYU worked with the quantum start-up Qunnect to send quantum information through standard telecom fiber in New York City between Manhattan and Brooklyn through a 10-mile quantum networking link. Instead of simulating communication challenges in a lab, the NYUQI team is already leveraging NYU’s city-wide campus by utilizing existing infrastructure to test secure quantum transmission between Manhattan and Brooklyn.

The NYUQI team is already leveraging NYU’s city-wide campus by utilizing existing infrastructure to test secure quantum transmission between Manhattan and Brooklyn.

This isn’t just theory; it is building a functioning prototype in the most demanding, dense urban environment in the world. Real-time, real-world deployment is a critical component missing in other isolated institutions. When the NYUQI achieves results, the technology will be that much more readily available to the massive financial, tech, and communications organizations operating right outside their door.

NYUQI includes a state-of-the-art Nanofabrication Cleanroom in Brooklyn serving as a high-tech foundry.NYU Tandon

While the Institute has built the physical infrastructure and designed the necessary scientific architecture, its enduring contribution will be the specialized workforce it creates for the new quantum economy. This addresses the market’s greatest deficit: a lack of individuals trained not just in physics, but in the integrated, full-stack approach that quantum demands.

By creating a pipeline of 100 to 200 graduate and doctoral students who are encouraged to collaborate across Computing, Sensing, and Communications, the NYUQI is narrowing the skills gap. These will be future leaders who can speak the language of the physicist, the materials scientist, and the engineer simultaneously. This commitment to interdisciplinary talent is also fueled by the launch of the new Master of Science in Quantum Science & Technology program at NYU Tandon, positioning the university among a select group worldwide offering such a specialized degree.

Interdisciplinary education creates the shared language and understanding poised to make graduates coming from collaborations in the NYUQI extremely valuable in the current landscape. Quantum challenges are not just technical; they are managerial and philosophical as well. An engineer working with the NYUQI will understand the requirements of the nanofabrication cleanroom and the foundations of superconducting qubits for quantum computing, just as a physicist will understand the application needs of an industry partner like a large financial institution. In a field where the entire team must be able to communicate seamlessly, these are professionals truly equipped to rapidly translate discovery into deployable technology. Creating a talent pipeline at scale will provide a missing link that converts New York’s vast commercial energy into genuine quantum advantage.

NYUQI: Building Talent, Technology, and Structure

The vision for the NYUQI is an act of strategic geography that plays directly into the sheer volume of opportunity and demand right outside their new facility. By building the talent, the technology, and the structure necessary to capitalize on this dense environment, NYU is not just participating in the quantum race, it is actively steering it.

Attendees of NYU’s 2025 Quantum Summit.Tracey Friedman/NYU

The initial hypothesis for the NYUQI was simple: the ultimate advantage lies in pursuing the science in the right place at the right time. Now, the institute will ensure that the next wave of scientific discovery, capable of solving previously intractable problems in finance, medicine, and security, will be conceived, built, and tested in the heart of New York City.

A technical exploration of IEEE 802.11bn’s physical and MAC layer enhancements — including distributed resource units, enhanced long range, multi-AP coordination, and seamless roaming — that define Wi-Fi 8.

What Attendees will Learn

1. Why Wi-Fi 8 prioritizes reliability over raw throughput — Understand how IEEE 802.11bn shifts the design philosophy from peak data-rate gains to ultra-high reliability.
2. How new physical layer features overcome uplink power limitations — Learn how distributed resource units spread tones across wider distribution bandwidths to boost per-tone transmit power, and how enhanced long range protocol data units use power-boosted preamble fields and frequency-domain duplication to extend uplink coverage.
3. How advanced MAC coordination reduces interference and latency — Examine multi-access point coordination schemes — coordinated beamforming, spatial reuse, time division multiple access, and restricted target wake time — alongside non-primary channel access and priority enhanced distributed channel access.
4. What seamless roaming and power management mean for next-generation deployments — Discover how seamless mobility domains eliminate reassociation delays during access point transitions, and how dynamic power save and multi-link power management let devices trade capability for battery life without sacrificing connectivity.

Download this free whitepaper now!

Last week’s Nvidia GTC conference highlighted new chip architectures to power AI. But as the chips become faster and more powerful, the remainder of data center infrastructure is playing catch-up. The power-delivery community is responding: Announcements from Delta, Eaton, Schneider Electric, and Vertiv showcased new designs for the AI era. Complex and inefficient AC-to-DC power conversions are gradually being replaced by DC configurations, at least in hyperscale data centers.

“While AC distribution remains deeply entrenched, advances in power electronics and the rising demands of AI infrastructure are accelerating interest in DC architectures,” says Chris Thompson, vice president of advanced technology and global microgrids at Vertiv.

AC-to-DC Conversion Challenges

Today, nearly all data centers are designed around AC utility power. The electrical path includes multiple conversions before power reaches the compute load. Power typically enters the data center as medium-voltage AC (1 to 35 kilovolts), is stepped down to low-voltage AC (480 or 415 volts) using a transformer, converted to DC inside an uninterruptible power supply (UPS) for battery storage, converted back to AC, and converted again to low-voltage DC (typically 54 V DC) at the server, supplying the DC power computing chips actually require.

“The double conversion process ensures the output AC is clean, stable, and suitable for data center servers,” says Luiz Fernando Huet de Bacellar, vice president of engineering and technology at Eaton.

That setup worked well enough for the amounts of power required by traditional data centers. Traditional data center computational racks draw on the order of 10 kW each. For AI, that is starting to approach 1 megawatt. At that scale, the energy losses, current levels, and copper requirements of AC-to-DC conversions become increasingly difficult to justify. Every conversion incurs some power loss. On top of that, as the amount of power that needs to be delivered grows, the sheer size of the convertors, as well as the connector requirements of copper busbars, becomes untenable. According to an Nvidia blog, a 1-MW rack could require as much as 200 kilograms of copper busbar. For a 1-gigawatt data center, it could amount to 200,000 kg of copper.

Benefits of High-Voltage DC Power

By converting 13.8-kV AC grid power directly to 800 V DC at the data center perimeter, most intermediate conversion steps are eliminated. This reduces the number of fans and power-supply units, and leads to higher system reliability, lower heat dissipation, improved energy efficiency, and a smaller equipment footprint.

“Each power conversion between the electric grid or power source and the silicon chips inside the servers causes some energy loss,” says Bacellar.

Switching from 415-V AC to 800-V DC in electrical distribution enables 85 percent more power to be transmitted through the same conductor size. This happens because higher voltage reduces current demand, lowering resistive losses and making power transfer more efficient. Thinner conductors can handle the same load, reducing copper requirements by 45 percent, a 5 percent improvement in efficiency, and 30 percent lower total cost of ownership for gigawatt-scale facilities.

“In a high-voltage DC architecture, power from the grid is converted from medium-voltage AC to roughly 800-V DC and then distributed throughout the facility on a DC bus,” said Vertiv’s Thompson. “At the rack, compact DC-to-DC converters step that voltage down for GPUs and CPUs.”

A report from technology advisory group Omdia claims that higher voltage DC data centers have already appeared in China. In the Americas, the Mt. Diablo Initiative (a collaboration among Meta, Microsoft, and the Open Compute Project) is a 400-V DC rack power distribution experiment.

Innovations in DC Power Systems

A handful of vendors are trying to get ahead of the game. Vertiv’s 800-V DC ecosystem that integrates with Nvidia Vera Rubin Ultra Kyber platforms will be commercially available in the second half of 2026. Eaton, too, is well advanced in its 800-V DC systems innovation courtesy of a medium-voltage solid-state transformer (SST) that will sit at the heart of DC power distribution system. Meanwhile Delta, has released 800-V DC in-row 660-kW power racks with a total of 480 kW of embedded battery backup units. And, SolarEdge is hard at work on a 99%-efficient SST that will be paired with a native DC UPS and a DC power distribution layer.

But much of the industry is far behind. Patrick Hughes, senior vice president of strategy, technical, and industry affairs for the National Electrical Manufacturers Association, says most innovation is happening at the 400-V DC level, though some are preparing 800-V DC. He believes the industry needs a complete, coordinated ecosystem, including power electronics, protection, connectors, sensing, and service‑safe components that scale together rather than in isolation. That, in turn, requires retooling manufacturing capacity for DC‑specific equipment, expanding semiconductor and materials supply, and clear, long‑term demand commitments that justify major capital investment across the value chain.

“Many are taking a cautious approach, offering limited or adapted solutions while waiting for clearer standards, safety frameworks, and customer commitments,” said Hughes. “Building the supply chain will hinge on stabilizing standards and safety frameworks so suppliers can design, certify, manufacture, and install equipment with confidence.”

The undying thirst for smarter (historically, that means larger) AI models and greater adoption of the ones we already have has led to an explosion in data-center construction projects, unparalleled both in number and scale. Chief among them is Meta’s planned 5-gigawatt data center in Louisiana, called Hyperion, announced in June of 2025. Meta CEO Mark Zuckerberg said Hyperion will “cover a significant part of the footprint of Manhattan,” and the first phase—a 2-GW version—will be completed by 2030.

Though the project’s stated 5-GW scale is the largest among its peers, it’s just one of several dozen similar projects now underway. According to Michael Guckes, chief economist at construction-software company ConstructConnect, spending on data centers topped US $27 billion by July of 2025 and, once the full-year figures are tallied, will easily exceed $60 billion. Hyperion alone accounts for about a quarter of that.

For the engineers assigned to bring these projects to life, the mix of challenges involved represent a unique moment. The world’s largest tech companies are opening their wallets to pay for new innovations in compute, cooling, and network technology designed to operate at a scale that would’ve seemed absurd five years ago.

At the same time, the breakneck pace of building comes paired with serious problems. Modern data-center construction frequently requires an influx of temporary workers and sharply increases noise, traffic, pollution, and often local electricity prices. And the environmental toll remains a concern long after facilities are built due to the unprecedented 24/7 energy demands of AI data centers which, according to one recent study, could emit the equivalent of tens of millions of tonnes of CO₂ annually in the United States alone.

Regardless of these issues, large AI companies, and the engineers they hire, are going full steam ahead on giant data-center construction. So, what does it really take to build an unprecedentedly large data center?

AI Rewrites Building Design

The stereotypical data-center building rests on a reinforced concrete slab foundation. That’s paired with a steel skeleton and poured concrete wall panels. The finished building is called a “shell,” a term that implies the structure itself is a secondary concern. Meta has even used gigantic tents to throw up temporary data centers.

Still, the scale of the largest AI data centers brings unique challenges. “The biggest challenge is often what’s under the surface. Unstable, corrosive, or expansive soils can lead to delays and require serious intervention,” says Robert Haley, vice president at construction consulting firm Jacobs. Amanda Carter, a senior technical lead at Stantec, said a soil’s thermal conductivity is also important, as most electrical infrastructure is placed underground. “If the soil has high thermal resistivity, it’s going to be difficult to dissipate [heat].” Engineers may take hundreds or thousands of soil samples before construction can begin.

GPUs

Modern AI data centers often use rack-scale systems, such as the Nvidia GB200 NVL72, which occupy a single data-center rack. Each rack contains 72 GPUs, 36 CPUs, and up to 13.4 terabytes of GPU memory. The racks measure over 2.2 meters tall and weigh over one and a half tonnes, forcing AI data centers to use thicker concrete with more reinforcement to bear the load.

A single GB200 rack can use up to 120 kilowatts. If Hyperion meets its 5-gigawatt goals, the data-center campus could include over 41,000 rack-scale systems, for a total of more than 3 million GPUs. The final number of GPUs used by Hyperion is likely to be less than that, though only because future GPUs will be larger, more capable, and use more power.

Money

According to ConstructConnect, spending on data centers neared US $27 billion through July of 2025 and, according to the latest data, will tally close to $60 billion through the end of the year. Meta’s Hyperion project is a big slice of the pie, at $10 billion.

Data-center spending has become an important prop for the construction industry, which is seeing reduced demand in other areas, such as residential construction and public infrastructure. ConstructConnect’s third quarter 2025 financial report stated that the quarter’s decline “would have been far more severe without an $11 billion surge in data center starts.”

There’s apparently no shortage of eligible sites, however, as both the number of data centers under construction, and the money spent on them, has skyrocketed. The spending has allowed companies building data centers to throw out the rule book. Prior to the AI boom, most data centers relied on tried-and-true designs that prioritized inexpensive and efficient construction. Big tech’s willingness to spend has shifted the focus to speed and scale.

The loose purse strings open the door to larger and more robust prefabricated concrete wall and floor panels. Doug Bevier, director of development at Clark Pacific, says some concrete floor panels may now span up to 23 meters and need to handle floor loads up to 3,000 kilograms per square meter, which is more than twice the load international building codes normally define for manufacturing and industry. In some cases, the concrete panels must be custom-made for a project, an expensive step that the economics of pre-AI data centers rarely justified.

Simultaneously, the time scale for projects is also compressed: Jamie McGrath, senior vice president of data-center operations at Crusoe, says the company is delivering projects in “about 12 months,” compared to 30 to 36 months before. Not all projects are proceeding at that pace, but speed is universally a priority.

That makes it difficult to coordinate the labor and materials required. Meta’s Hyperion site, located in rural Richland Parish, Louisiana, is emblematic of this challenge. As reported by NOLA.com, at least 5,000 temporary workers have flocked to the area, which has only about 20,000 permanent residents. These workers earn above-average wages and bring a short-term boost for some local businesses, such as restaurants and convenience stores. However, they have also spurred complaints from residents about traffic and construction noise and pollution.

This friction with residents includes not only these obvious impacts, but also things you might not immediately suspect, such as light pollution caused by around-the-clock schedules. Also significant are changes to local water tables and runoff, which can reduce water quality for neighbors who rely on well water. These issues have motivated a few U.S. cities to enact data-center bans.

Data Centers Often Go BYOP (bring your own power)

Meta’s Richland Parish site also highlights a problem that’s priority No. 1 for both AI data centers and their critics: power.

Data centers have always drawn large amounts of power, which nudged data-center construction to cluster in hubs where local utilities were responsive to their demands. Virginia’s electric utility, Dominion Energy, met demand with agreements to build new infrastructure, often with a focus on renewable energy.

The power demands of the largest AI data centers, though, have caught even the most responsive utilities off guard. A report from the Lawrence Berkeley National Laboratory, in California, estimated the entire U.S. data-center industry consumed an average load of roughly 8 GW of power in 2014. Today, the largest AI data-center campuses are built to handle up to a gigawatt each, and Meta’s Hyperion is projected to require 5 GW.

“Data centers are exasperating issues for a lot of utilities,” says Abbe Ramanan, project director at the Clean Energy Group, a Vermont-based nonprofit.

Ramanan explains that utilities often use “peaker plants” to cope with extra demand. They’re usually older, less efficient fossil-fuel plants which, because of their high cost to operate and carbon output, were due for retirement. But Ramanan says increased electricity demand has kept them in service.

Meta secured power for Hyperion by negotiating with Entergy, Louisiana’s electric utility, for construction of three new gas-turbine power plants. Two will be located near the Richland Parish site, while a third will be located in southeast Louisiana.

Entergy frames the new plants as a win for the state. “A core pillar of Entergy and Meta’s agreement is that Meta pays for the full cost of the utility infrastructure,” says Daniel Kline, director of power-delivery planning and policy at Entergy. The utility expects that “customer bills will be lower than they otherwise would have been.” That would prove an exception, as a recent report from Bloomberg found electricity rates in regions with data centers are more likely to increase than in regions without.

CO2

Research published in Nature in 2025 projects that data-center emissions will range from 24 million to 44 million CO2-equivalent metric tonnes annually through 2030 in the United States alone. While some materials used in data centers, such as concrete, lead to significant emissions, the majority of these emissions will result from the high energy demands of AI servers.

Estimating the carbon emissions of Hyperion is difficult, as the project won’t be completed until 2030. Assuming that the three new natural gas plants that are planned for construction as part of the project produce emissions typical for their type, however, the plants could lead to full life-cycle emissions of between 4 million and 10 million metric tons of CO2 annually—roughly equivalent to the annual emissions of a country like Latvia.

Concrete

Data centers are typically built from concrete, with steel used as a skeleton to reinforce and shape the concrete shell. While the foundation is often poured concrete, the walls and floors are most often built from prefabricated concrete panels that can span up to 23 meters. Floors use a reinforced T-shape, similar to a steel girder, measuring up to 1.2 meters across at its thickest point. The largest data centers include hundreds of these concrete panels.

The America Cement Association projects that the current surge in building will require 1 million tonnes of cement over the next three years, though that’s still a tiny fraction of the overall cement industry, which weighed in at roughly 103 million tonnes in 2024.

The plants, which will generate a combined 2.26 GW, will use combined-cycle gas turbines that recapture waste heat from exhaust. This boosts thermal efficiency to 60 percent and beyond, meaning more fuel is converted to useful energy. Simple-cycle turbines, by contrast, vent the exhaust, which lowers efficiency to around 40 percent.

Even so, total life-cycle emissions for the Hyperion plants could range from 4 million to over 10 million tonnes of CO2 each year, depending on how frequently the plants are put in use and the final efficiency benchmarks once built. On the high end, that’s as much CO2 as produced by over 2 million passenger cars. Fortunately, not all of Meta’s data centers take the same approach to power. The company has announced a plan to power Prometheus, a large data-center project in Ohio scheduled to come online before the end of 2026, with nuclear energy.

But other big tech companies, spurred by the need to build data centers quickly, are taking a less efficient approach.

xAI’s Colossus 2, located in Memphis, is the most extreme example. The company trucked dozens of temporary gas-turbine generators to power the site located in a suburban neighborhood. OpenAI, meanwhile, has gas turbines capable of generating up to 300 megawatts at its new Stargate data center in Abilene, Texas, slated to open later in 2026. Both use simple-cycle turbines with a much lower efficiency rating than the combined-cycle plants Entergy will build to power Hyperion.

Demand for gas turbines is so intense, in fact, that wait times for new turbines are up to seven years. Some data centers are turning toward refurbished jet engines to obtain the turbines they need.

AI Racks Tip the Scales

The demand for new, reliable power is driven by the power-hungry GPUs inside modern AI data centers.

In January of 2025, Mark Zuckerberg announced in a post on Facebook that Meta planned to end 2025 with at least 1.3 million GPUs in service. OpenAI’s Stargate data center plans to use over 450,000 Nvidia GB200 GPUs, and xAI’s Colossus 2, an expansion of Colossus, is built to accommodate over 550,000 GPUs.

GPUs, which remain by far the most popular for AI workloads, are bundled into human-scale monoliths of steel and silicon which, much like the data centers built to house them, are rapidly growing in weight, complexity, and power consumption.

Memory

In addition to raw compute performance, Nvidia GB200 NVL72 racks also require huge amounts of memory. An Nvidia GB200 NVL72 rack may include up to 13.4 terabytes of high-bandwidth memory, which implies a data-center campus at Hyperion’s scale will require at least several dozen petabytes.

The immense demand has sent memory prices soaring: The price of DRAM, specifically DDR5, has increased 172 percent in 2025.

Power

Hyperion is expected to use 5 gigawatts of power across 11 buildings, which works out to just under 500 megawatts per building, assuming each will be similar to its siblings. That’s enough to power roughly 4.2 million U.S. homes.

Just one Hyperion data center built at the Richland Parish site will consume twice as much power as xAI’s Colossus which, at the time of its completion in the summer of 2024, was among the largest data centers yet built.

Nvidia’s GB200 NVL72—a rack-scale system—is currently a leading choice for AI data centers. A single GB200 rack contains 72 GPUs, 36 CPUs, and up to 17 terabytes of memory. It measures 2.2 meters tall, tips the scales at up to 1,553 kilograms, and consumes about 120 kilowatts—as much as around 100 U.S. homes. And this, according to Nvidia, is just the beginning. The company anticipates future racks could consume up to a megawatt each.

Viktor Petik, senior vice president of infrastructure solutions at Vertiv, says the rapid change in rack-scale AI systems has forced data centers to adapt. “AI racks consume far more power and weigh more than their predecessors,” says Petik. He adds that data centers must supply racks with multiple power feeds, without taking up extra space.

The new power demands from rack-scale systems have consequences that are reflected in the design of the data center—even its footprint.

In 2022 Meta broke ground on a new data center at a campus in Temple, Texas. According to SemiAnalysis, which studies AI data centers, construction began with the intent to build the data center in an H-shaped configuration common to other Meta data centers.

LAND

Meta CEO Mark Zuckerberg kicked off the buzz around Hyperion by saying it would cover a large chunk of Manhattan. Many took that to mean Hyperion would be a single building of that size, which isn’t correct. Hyperion will actually be a cluster of data centers—11 are currently planned—with over 370,000 square meters of floor space. That’s a lot smaller even than New York City’s Central Park, which covers 6 percent of Manhattan.

Meta has room to grow, however. The Richland Parish site spans 14.7 million square meters in total, which is about a quarter the area of Manhattan. And the 370,000 square meters of floor space Hyperion is expected to provide doesn’t include external infrastructure, such as the three new combined-cycle gas power plants Louisiana utility Entergy is building to power the project.

Construction was paused midway in December of 2022, however, as part of a company-wide review of its data-center infrastructure. Meta decided to knock down the structure it had built and start from scratch. The reasons for this decision were never made public, but analysts believe it was due to the old design’s inability to deliver sufficient electricity to new, power-hungry AI racks. Construction resumed in 2023.

Meta’s replacement ditches the H-shaped building for simple, long, rectangular structures, each flanked by rows of gas-turbine generators. While Meta’s plans are subject to change, Hyperion is currently expected to comprise 11 rectangular data centers, each packed with hundreds of thousands of GPUs, spread across the 13.6-square-kilometer Richland Parish campus.

Cooling, and Connecting, at Scale

Nvidia’s ultradense AI GPU racks are changing data centers not only with their weight, and power draw, but also with their intense cooling and bandwidth requirements.

Data centers traditionally use air cooling, but that approach has reached its limits. “Air as a cooling medium is inherently inferior,” says Poh Seng Lee, head of CoolestLAB, a cooling research group at the National University of Singapore.

Instead, going forward, GPUs will rely on liquid cooling. However, that adds a new layer of complexity. “It’s all the way to the facilities level,” says Lee. “You need pumps, which we call a coolant distribution unit. The CDU will be connected to racks using an elaborate piping network. And it needs to be designed for redundancy.” On the rack, pipes connect to cold plates mounted atop every GPU; outside the data-center shell, pipes route through evaporation cooling units. Lee says retrofitting an air-cooled data center is possible but expensive.

The networking used by AI data centers is also changing to cope with new requirements. Traditional data centers were positioned near network hubs for easy access to the global internet. AI data centers, though, are more concerned with networks of GPUs.

These connections must sustain high bandwidth with impeccable reliability. Mark Bieberich, a vice president at network infrastructure company Ciena, says its latest fiber-optic transceiver technology, WaveLogic 6, can provide up to 1.6 terabytes per second of bandwidth per wavelength. A single fiber can support 48 wavelengths in total, and Ciena’s largest customers have hundreds of fiber pairs, placing total bandwidth in the thousands of terabits per second.

This is a point where the scale of Meta’s Hyperion, and other large AI data centers, can be deceptive. It seems to imply the physical size of a single data center is what matters. But rather than being a single building, Hyperion is actually a set of buildings connected by high-speed fiber-optics.

“Interconnecting data centers is absolutely essential,” says Bieberich. “You could think about it as one logical AI training facility, but with geographically distributed facilities.” Nvidia has taken to calling this “scale across,” to contrast it with the idea that data centers must “scale up” to larger singular buildings.

The Big but Hazy Future

The full scale of the challenges that face Hyperion, and other future AI data centers of similar scale, remain hazy. Nvidia has yet to introduce the rack-scale AI GPU systems it will host. How much power will it demand? What type of cooling will it require? How much bandwidth must be provided? These can only be estimated.

In the absence of details, the gravity of AI data-center design is pulled toward one certainty: It must be big. New data-center designers are rewriting their rule book to handle power, cooling, and network infrastructure at a scale that would’ve seemed ridiculous five years ago.

This innovation is fueled by big tech’s fat wallet, which shelled out tens of billions of dollars in 2025 alone, leading to questions about whether the spending is sustainable. For the engineers in the trenches of data-center design, though, it’s viewed as an opportunity to make the impossible possible.

“I tell my engineers, this is peak. We’re being engineers. We’re being asked complicated questions,” says Stantec’s Carter. “We haven’t got to do that in a long time.”

This article appears in the April 2026 print issue.

This is a sponsored article brought to you by PNY Technologies.

Key Challenges for Data Science

• Data Preparation: Data preparation is a complex, time-consuming process that takes most of a data scientist’s time.
• Scaling: Volume of data is growing at a rapid pace. Data scientists may resort to downsampling datasets to make large datasets more manageable, leading to suboptimal results.
• Hardware: Demand for accelerated AI hardware for data centers and cloud service providers (CSPs) is exceeding supply. Current desktop computing resources may not be suitable for data science workflows.

Benefits of RTX PRO-Powered AI Workstations

NVIDIA RTX PRO 6000 Blackwell Workstation Edition delivers ultimate acceleration for data science and AI workflows. These powerful and robust workstations enable real-time rendering, rapid prototyping, and seamless collaboration. With support for up to four NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition GPUs, users can achieve data center-level performance right at their desk, making even the most demanding tasks manageable.

PNY is redefining professional computing with the ‪@NVIDIA‬ RTX PRO 6000 Blackwell Workstation Edition, the most powerful desktop GPU ever built. Engineered for unmatched compute power, massive memory capacity, and breakthrough performance, this cutting-edge solution delivers a quantum leap forward in workflow efficiency, enabling professionals to tackle the most demanding applications with ease.PNY

NVIDIA RTX PRO 6000 Blackwell Workstation Edition empowers data scientists to handle massive datasets, perform advanced visualizations, and support multi-user environments without compromise. It’s ideal for organizations scaling up their analytics or running complex models. NVIDIA RTX PRO 6000 Blackwell Workstation Edition is optimized for AI workflows, leveraging the NVIDIA AI software stack, including CUDA-X, and NVIDIA Enterprise software. These platforms enable zero-code-change acceleration for Python-based workflows and support over 100 AI-powered applications, streamlining everything from data preparation to model deployment.

Finally, NVIDIA RTX PRO 6000 Blackwell Workstation Edition offers significant advantages in security and cost control. By offloading compute from the data center and reducing reliance on cloud resources, organizations can lower expenses and keep sensitive data on-premises for enhanced protection.

Accelerate Every Step of Your Workflow

NVIDIA RTX PRO 6000 Blackwell Workstation Edition is designed to transform the entire data science pipeline, delivering end-to-end acceleration from data preparation to model deployment. With NVIDIA CUDA-X open-source data science cuDF library and other GPU-accelerated libraries, data scientists can process massive datasets at lightning speed, often achieving up to 50X faster performance compared to traditional CPU-based tools. This means tasks like cleaning data, managing missing values, and engineering features can be completed in seconds, not hours, allowing teams to focus on extracting insights and building better models.

NVIDIA RTX PRO 6000 Blackwell Workstation Edition is designed to transform the entire data science pipeline, delivering end-to-end acceleration from data preparation to model deployment

Exploratory data analysis is elevated with advanced analytics and interactive visualizations, powered by NVIDIA CUDA-X and PyData libraries. These tools enable users to create expansive, responsive visualizations that enhance understanding and support critical decision-making. When it comes to model training, GPU-accelerated XGBoost slashes training times from weeks to minutes, enabling rapid iteration and faster time-to-market AI solutions.

NVIDIA RTX PRO 6000 Blackwell Workstation Edition streamlines collaboration and scalability. With NVIDIA AI Workbench, teams can set up projects, develop, and collaborate seamlessly across desktops, cloud platforms, and data centers. The unified software stack ensures compatibility and robustness, while enterprise-grade hardware maximizes uptime and reliability for demanding workflows.

By integrating these advanced capabilities, NVIDIA RTX PRO 6000 Blackwell Workstation Edition empowers data scientists to overcome bottlenecks, boost productivity, and drive innovation, making them an essential foundation for modern, enterprise-ready AI development.

Performance Benchmarks

NVIDIA’s cuDF library offers zero-code change acceleration for pandas, delivering up to 50X performance gains. For example, a join operation that takes nearly 5 minutes on CPU completes in just 14 seconds on GPU. Advanced group by operations drop from almost 4 minutes to just 4 seconds.

Enterprise-Ready Solutions from PNY

Available from leading OEM manufacturers, NVIDIA RTX PRO 6000 Blackwell Workstation Edition Series GPUs are specifically engineered to meet the rigorous demands of enterprise environments. These systems incorporate NVIDIA Connect-X networking, now available at PNY and a comprehensive suite of deployment and support tools, ensuring seamless integration with existing IT infrastructure.

Designed for scalability, the latest generation of workstations can tackle complex AI development workflows at scale for training, development, or inferencing. Enterprise-grade hardware maximizes uptime and reliability.

To learn more about NVIDIA RTX PRO™ Blackwell solutions, visit: NVIDIA RTX PRO Blackwell | PNY Pro | pny.com or email GOPNY@PNY.COM

Light-warping physics made “invisibility cloaks” a possibility. Now two startups hope to harness the science underlying this advance to boost the bandwidth of data centers and speed artificial intelligence.

Roughly 20 years ago, scientists developed the first structures capable of curving light around objects to conceal them. These are composed of optical metamaterials—materials with structures smaller than the wavelengths they are designed to manipulate, letting them bend light in unexpected ways.

The problem with optical cloaks? “There’s no market for them,” says Patrick Bowen, cofounder and CEO of photonic computing startup Neurophos in Austin, Texas. For instance, each optical cloak typically works only on a single color of light instead of on all visible colors as you might want for stealth applications.

Now companies are devising more practical uses for the science behind cloaks, such as improving the switches that connect computers in data centers for AI and other cloud services. Increasingly, data centers are looking to use optical circuit switches to overcome the bandwidth limits and power consumption of conventional electronic switches and networks that require converting data between light to electrons multiple times.

RELATED: Semiconductor Industry Closes in on 400 Gb/s Photonics Milestone

However, today’s optical switching technologies have drawbacks of their own. For instance, ones that depend on silicon photonics face problems with energy efficiency, while those that rely on microelectromechanical systems (MEMS) can prove unreliable, says Sam Heidari, CEO of optical metasurface startup Lumotive in Redmond, Wash.

Instead, Lumotive has developed metamaterials with adjustable properties. Its new microchip, which debuted 19 March, is covered with copper structures built using standard chipmaking techniques. Between these copper features are liquid crystal elements. The structure of these elements are electronically programmable, just like in liquid crystal displays (LCDs), to alter the optical properties of the metamaterial chip.

The microchip can precisely steer, lens, shape, and split beams of light reflected off its surface. It can perform all the same functions as multiple optical components with no moving parts in a programmable way in real time, according to Lumotive. “Having no moving parts significantly improves reliability,” Heidari says.

“We had to go through a lot of R&D at the foundries to not only make our devices functional, but also commercially viable in terms of the right cost and right reliability,” Heidari says.

The company says its new chips are capable of manipulating not only the industry’s standard of 256 by 256 ports, but could scale up to 10,000 by 10,000. “We think this is game-changing for data centers,” Heidari says. Lumotive plans to launch its first optical switches at the end of 2026.

Optical Computing With Metamaterials

Similarly, Neurophos hopes its technology may be transformative for artificial intelligence. Since AI is proving energy hungry when run on conventional electronics, scientists are exploring optical computing as a low-power alternative by processing data with light instead of electrons.

However, optical processors in the works today are typically far too bulky to achieve a compute density competitive with the best modern electronic processors, Bowen says. Neurophos says it can use metamaterials to build optical modulators—the optical equivalent of a transistor—that are 1/10,000th the size of today’s designs using standard chipmaking processes. “It’s entirely CMOS,” Bowen says. “There are no exotic materials in it.”

When a laser beam encoding data shines on a Neurophos chip, the way in which each metamaterial element is configured alters the reflected beam to encode results from complex AI tasks. “We basically fit a 1,000- by-1,000 array of optical modulators on a tiny 5-by-5-millimeter area on a chip,” Bowen says. “If you wanted to do that with off-the-shelf silicon photonics, your chip would be a square meter in size.”

All in all, Bowen claims the Neurophos microchip will offer 50 times greater compute density and 50 times greater energy efficiency than Nvidia’s Blackwell-generation GPU. The company says that hyperscalers—the world’s biggest cloud service providers—will evaluate two upcoming proof-of-concept chips this year. Neurophos is targeting its first systems for early 2028, with production ramping mid-2028.

Happy 80th anniversary, ENIAC! The Electronic Numerical Integrator and Computer, the first large-scale, general-purpose, programmable electronic digital computer, helped shape our world.

On 15 February 1946, ENIAC—developed in the Moore School of Electrical Engineering at the University of Pennsylvania, in Philadelphia—was publicly demonstrated for the first time. Although primitive by today’s standards, ENIAC’s purely electronic design and programmability were breakthroughs in computing at the time. ENIAC made high-speed, general-purpose computing practicable and laid the foundation for today’s machines.

On the eve of its unveiling, the U.S. Department of War issued a news release hailing it as a new machine “expected to revolutionize the mathematics of engineering and change many of our industrial design methods.” Without a doubt, electronic computers have transformed engineering and mathematics, as well as practically every other domain, including politics and spirituality.

ENIAC’s success ushered the modern computing industry and laid the foundation for today’s digital economy. During the past eight decades, computing has grown from a niche scientific endeavor into an engine of economic growth, the backbone of billion-dollar enterprises, and a catalyst for global innovation. Computing has led to a chain of innovations and developments such as stored programs, semiconductor electronics, integrated circuits, networking, software, the Internet, and distributed large-scale systems.

Inside the ENIAC

The motivation for developing ENIAC was the need for faster computation during World War II. The U.S. military wanted to produce extensive artillery firing tables for field gunners to quickly determine settings for a specific weapon, a target, and conditions. Calculating the tables by hand took “human computers” several days, and the available mechanical machines were far too slow to meet the demand.

80 Years of Electronic Computer Milestones

1946

ENIAC operational

Birth of electronic computing

1951

UNIVAC I

Start of commercial computing

1958

Integrated circuit

Foundation for modern computer hardware

1964

IBM System/360

Popular mainframe computer

1970

Programmed Data Processor (PDP-11)

Popular 16-bit minicomputer

1971

Intel 4004

Beginning of the microprocessor and microcomputer era

1975

Cray-1

First supercomputer

1977

VAX

Popular 32-bit minicomputer

1981

IBM PC

Personal and small-business computing

1989

World Wide Web

Digital communication, interaction, and transaction (e-commerce)

2002

Amazon Web Services

Beginning of the cloud computing revolution

2010

Apple iPad

Handheld computer/tablet

2010

Industry 4.0

Delivered real-time decision-making, smart manufacturing, and logistics

2016

First reprogrammable quantum computer demonstrated

Ignited interest in quantum computing

2023

Generative AI boom

Widespread use of GenAI by individuals, businesses, and academia

2026

ENIAC’s 80th anniversary

80 years of computing evolution

In 1942 John Mauchly, an associate professor of electrical engineering at Penn’s Moore School, suggested using vacuum tubes to speed up computer calculations. Following up on his theory, the U.S. Army Ballistic Research Laboratory, which was responsible for providing artillery settings to soldiers in the field, commissioned Mauchly and his colleagues J. Presper Eckert and Adele Katz Goldstine, to work on a new high-speed computer. Eckert was a lab instructor at Moore, and Goldstine became one of ENIAC’s programmers. It took them a year to design ENIAC and 18 months to build it.

The computer contained about 18,000 vacuum tubes, which were cooled by 80 air blowers. More than 30 meters long, it filled a 9 m by 15 m room and weighed about 30 kilograms. It consumed as much electricity as a small town.

Programming the machine was difficult. ENIAC did not have stored programs, so to reprogram the machine, operators manually reconfigured cables with switches and plugboards, a process that took several days.

By the 1950s, large universities either had acquired or built their own machines to rival ENIAC. The schools included Cambridge (EDSAC), MIT (Whirlwind), and Princeton (IAS). Researchers used the computers to model physical phenomena, solve mathematical problems, and perform simulations.

After almost nine years of operation, ENIAC officially was decommissioned on 2 October 1955.

ENIAC in Action: Making and Remaking the Modern Computer, a book by Thomas Haigh, Mark Priestley, and Crispin Rope, describes the design, construction, and testing processes and dives into its afterlife use. The book also outlines the complex relationship between ENIAC and its designers, as well as the revolutionary approaches to computer architecture.

In the early 1970s, there was a controversy over who invented the electronic computer and who would be assigned the patent. In 1973 Judge Earl Richard Larson of U.S. District Court in Minnesota ruled in the Honeywell v. Sperry Rand case that Eckert and Mauchly did not invent the automatic electronic digital computer but instead had derived their subject matter from a computer prototyped in 1939 by John Vincent Atanasoff and Clifford Berry at Iowa State College (now Iowa State University). The ruling granted Atanasoff legal recognition as the inventor of the first electronic digital computer.

IEEE’s ENIAC Milestone

In 1987 IEEE designated ENIAC as an IEEE Milestone, citing it as “a major advance in the history of computing” and saying the machine “established the practicality of large-scale electronic digital computers and strongly influenced the development of the modern, stored-program, general-purpose computer.”

The commemorative Milestone plaque is displayed at the Moore School, by the entrance to the classroom where ENIAC was built.

“The ENIAC legacy heralded the computer age, transforming not only science and industry but also education, research, and human communication and interaction.”

A paper on the machine, published in 1996 in IEEE Annals of the History of Computing and available in the IEEE Xplore Digital Library, is a valuable source of technical information.

“The Second Life of ENIAC,” an article published in the annals in 2006, covers a lesser-known chapter in the machine’s history, about how it evolved from a static system—configured and reconfigured through laborious cable plugging—into a precursor of today’s stored-program computers.

A classic history paper on ENIAC was published in the December 1995 IEEE Technology and Society Magazine.

The IEEE Inspiring Technology: 34 Breakthroughs book, published in 2023, features an ENIAC chapter.

The women behind ENIAC

One of the most remarkable aspects of the ENIAC story is the pivotal role women played, according to the book Proving Ground: The Untold Story of the Six Women Who Programmed the World’s First Modern Computer, highlighted in an article in The Institute. There were no “programmers” at that time; only schematics existed for the computer. Six women, known as the ENIAC 6, became the machine’s first programmers.

The ENIAC 6 were Kathleen Antonelli, Jean Bartik, Betty Holberton, Marlyn Meltzer, Frances Spence, and Ruth Teitelbaum.

“These six women found out what it took to run this computer, and they really did incredible things,” a Penn professor, Mitch Marcus, said in a 2006 PhillyVoice article. Marcus teaches in Penn’s computer and information science department.

In 1997 all six female programmers were inducted into the Women in Technology International Hall of Fame, in Los Angeles.

Two other women contributed to the programming. Goldstine wrote ENIAC’s five-volume manual, and Klára Dán von Neumann, wife of John von Neumann, helped train the programmers and debug and verify their code.

To honor the women of ENIAC, the IEEE Computer Society established the annual Computer Pioneer Award in 1981. Eckert and Mauchly were among the award’s first recipients. In 2008 Bartik was honored with the award. Nominations are open to all professionals, regardless of gender.

An ENIAC replica

Last year a group of 80 autistic students, ages 12 to 16, from PS Academy Arizona, in Gilbert, recreated the ENIAC using 22,000 custom parts. It took the students almost six months to assemble.

A ceremony was held in January to display their creation. The full-scale replica features actual-size panels made from layered cardboard and wood. Although all electronic components are simulated, they are not electrically active. The machine, illuminated by hundreds of LEDs, is accompanied by a soundtrack that simulates the deep hum of ENIAC’s transformers and the rhythmic clicking of relays.

“Every major unit, accumulators, function tables, initiator, and master programmer is present and placed exactly where it was on the original machine,” Tom Burick, the teacher who mentored the project, said at the ceremony.

The replica, still on display at the school, is expected to be moved to a more permanent spot in the near future.

ENIAC’s legacy

ENIAC’s significance is both technical and symbolic. Technically, it marks the beginning of the chain of innovations that created today’s computational infrastructure. Symbolically, it made governments, militaries, universities, and industry view computation as a tool for improvement and for innovative applications that had previously been impossible. It marked a tectonic shift in the way humans approach problem-solving, modeling, and scientific reasoning.

The ENIAC legacy heralded the computer age, transforming not only science and industry but also education, research, and human communication and interaction.

As Eckert is reported to have said, “There are two epochs in computer history: Before ENIAC and After ENIAC.”

Coevolution of programming languages

The remarkable evolution of computer hardware during the past 80 years has been sparked by advances in programming languages—the essential drivers of computing.

From the manual rewiring of ENIAC to the orchestration of intelligent, distributed systems, programming languages have steadily evolved to make computers more powerful, expressive, and accessible.

Lessons From Computing’s Remarkable Journey

Computing history teaches us that flexibility, accessibility, collaboration, sound governance, and forward thinking are essential for sustained technological progress. In a recent Communications of the ACM article, Richa Gupta identified four historic shifts that led to computing’s rapid, transformative progress:

1. Programmable machines taught us that flexibility is key; technologies that adapt and are repurposed scale better.
2. The Internet showed that connection and standard protocols drive explosive growth but also bring new risks such as data security issues, invasion of privacy, and misuse.
3. Personal computers illustrated that accessibility and usability matter more than raw power. When nonexperts can use a tool easily, adoption rises.
4. The open-source movement revealed that collaborative innovation accelerates growth and helps spot problems early.

Predictions for computing in the decades ahead

The evolution of computing will continue along multiple trajectories, with the emphasis moving from generalization to specialization (for AI, graphics, security, and networking), from monolithic system design to modular integration, and from performance-centric metrics alone to energy efficiency and sustainability as primary objectives.

Increasingly, security will be built into hardware by design. Computing paradigms will expand beyond traditional deterministic models to embrace probabilistic, approximate, and hybrid approaches for certain tasks.

Those developments will usher in a new era of computing and a new class of applications.

Looming over the internet lasers and firestarting phones companies were touting at Mobile World Congress in Barcelona this month, was a more nebulous but much larger announcement: a pan-European cloud called EURO-3C.

EURO-3C’s backers—Spanish telecoms giant Telefónica, dozens of other European companies, and the European Commission (EC)—aim to fill a gap. U.S.-based cloud giants dominate in the EU, and European policymakers want their growing portfolio of digital government services on a “sovereign cloud” under full EU control.

But the EU lacks a real equivalent to the likes of AWS or Microsoft Azure. Indeed, any effort to build one will inevitably run up against the same U.S. cloud giants.

Just four U.S.-based hyperscalers—AWS, Microsoft Azure, Google Cloud, and IBM Cloud—together account for some 70 percent of EU cloud services. This is despite the fact that the 2018 U.S. CLOUD Act allows U.S. federal law enforcement—at least in theory—to compel U.S.-based firms to hand over data that’s stored abroad.

Who Do You Trust?

But those hypothetical risks to digital services have become more real as transatlantic relations have soured under the second Trump administration. The U.S. has openly threatened to invade an EU member state and sanctioned a European Commissioner for passing legislation the White House dislikes.

After the White House sanctioned the Netherlands-based International Criminal Court in February 2025, Court staffers claimed Microsoft locked the Court’s chief prosecutor out of his email (Microsoft has denied this). Around the same time, the U.S. reportedly threatened to sever EU ally Ukraine’s access to crucial Starlink satellite internet as leverage during trade negotiations.

“The geopolitical risk isn’t just the most extreme form of a doomsday ‘kill switch’ where Washington turns off Europe’s internet,” says Stéfane Fermigier of EuroStack, an industry group that supports European digital independence. “It is the selective degradation of services and a total lack of retaliatory leverage.”

What, then, is the EU to do? France offers an example. Even before 2025, France implemented harsh restrictions on non-EU cloud providers in public services—providers must locate data in the EU, rely on EU-based staff, and may not have majority non-EU shareholders. Now, EU policymakers are following France’s lead.

In October 2025, the EC issued a two-part framework for judging cloud providers bidding for public-sector contracts. In the first part, the framework lays out a sort of sovereignty ladder. The more that a provider is subject to EU law, the higher its sovereignty level on this ladder. Any prospective bidder must first meet a certain level, depending on the tender.

Qualifying bidders then move to the second part, where their “sovereignty” is scored in more detail. Using too much proprietary software; over-relying on supply chains from outside the EU; having non-EU support staff; liability to non-EU laws like the CLOUD Act: All hurt a bidder’s score.

The framework was created for one tender, but observers say it sets a major precedent. Cloud providers bidding for state contracts across Europe may need to follow it, and it may influence legislation on both national and EU-wide levels.

A Question of Scale

Who, then, will receive high marks? At the moment, the answer is not simple. The EU cloud scene is quite fragmented. Numerous modest EU providers offer “sovereign cloud” services—such as Deutsche Telekom’s T-Systems, OVHcloud, and Scaleway—but none are on the scale of AWS or Google Cloud.

Inertia is on the side of the U.S. cloud giants, which can invest in their infrastructure and services on a far grander scale than their European counterparts. Some U.S. providers now offer cloud services they say comply with the Commission’s “cloud sovereignty” demands.

Some European observers, like EuroStack, say such promises are hollow so long as a provider’s parent company is subject to the likes of the CLOUD Act and loopholes in the Commission’s process remain open. An AWS spokesperson told IEEE Spectrum it had not disclosed any non-US enterprise or government data to the U.S. government under the CLOUD Act; a Google spokesperson said that its most sensitive EU offerings “are subject to local laws, not U.S. law”.

Even if a project like EURO-3C can offer a large-scale alternative, the U.S. cloud giants have another sort of inertia. Many developers—and many public purchasers of their services—will need convincing to leave behind a familiar environment.

“If you look at AWS, you look at Google, they’ve created some super technology. It’s very convenient, it’s easy to use,” says Arnold Juffer, CEO of the Netherlands-based cloud provider Nebul. “Once you’re in that platform, in that ecosystem, it’s very hard to get out.”

Martyna Chmura, an analyst at the Bloomsbury Intelligence and Security Institute, a London-based think tank, sees some EU developers taking a mixed approach. “Many organizations are already moving toward multicloud setups, using European or sovereign providers for sensitive workloads while still relying on hyperscalers for certain services,” she says.

In that case, the EU’s top-down demands may encourage developers to use EU providers for sensitive applications—like government services, transport, autonomous vehicles, and some industrial automation—even if it’s inconvenient in the short term, or if it causes even more fragmentation of the EU cloud scene. “Running systems across different platforms can increase integration costs and make security and data governance more complicated. In some cases, organisations could lose some of the efficiency and cost advantages that come from using large hyperscale platforms,” Chmura says.

“Overall, the EU appears willing to accept some of these trade-offs,” Chmura says.

This week, over 30,000 people are descending upon San Jose, Calif., to attendNvidia GTC, the so-called Superbowl of AI—a nickname that may or may not have been coined by Nvidia. At the main event Jensen Huang, Nvidia CEO, took the stage to announce (among other things) a new line ofnext-generation Vera Rubin chips that represent a first for the GPU giant: a chip designed specifically to handle AI inference. The Nvidia Groq 3 language processing unit (LPU) incorporates intellectual property Nvidialicensed from the startup Groq last Christmas Eve for US $20 billion.

“Finally, AI is able to do productive work, and therefore the inflection point of inference has arrived,” Huang told the crowd. “AI now has to think. In order to think, it has to inference. AI now has to do; in order to do, it has to inference.”

Training and inference tasks have distinct computational requirements. While training can be done on huge amounts of data at the same time and can take weeks, inference must be run on a user’s query when it comes in. Unlike training, inference doesn’t require running costlybackpropagation. With inference, the most important thing is low latency—users expect the chatbot to answer quickly, and for thinking or reasoning models, inference runs many times before the user even sees an output.

Over the past few years, inference-specific chip startups were experiencing a sort of Cambrian explosion, with different companies exploring distinct approaches to speed up the task. The startups includeD-matrix, with digital in-memory compute;Etched, with an ASIC for transformer inference;RainAI, with neuromorphic chips;EnCharge, with analog in-memory compute;Tensordyne, with logarithmic math to make AI computations more efficient;FuriosaAI, with hardware optimized for tensor operation rather than vector-matrix multiplication, and others.

Late last year, it looked like Nvidia had picked one of the winners among the crop of inference chips when it announced its deal with Groq. The Nvidia Groq 3 LPU reveal came a mere two and a half months after, highlighting the urgency of the growing inference market.

Memory bandwidth and data flow

Groq’s approach to accelerating inference relies on interleaving processing units with memory units on the chip. Instead of relying on high-bandwidth memory (HBM) situated next to GPUs, it leans on SRAM memory integrated within the processor itself. This design greatly simplifies the flow of data through the chip, allowing it to proceed in a streamlined, linear fashion.

“The data actually flows directly through the SRAM,”Mark Heaps said at the Supercomputing conference in 2024. Heaps was a chief technology evangelist at Groq at the time and is now director of developer marketing at Nvidia. “When you look at a multicore GPU, a lot of the instruction commands need to be sent off the chip, to get into memory and then come back in. We don’t have that. It all passes through in a linear order.”

Using SRAM allows that linear data flow to happen exceptionally fast, leading to the low latency required for inference applications. “The LPU is optimized strictly for that extreme low latency token generation,” saysIan Buck, VP and general manager of hyperscale and high-performance computing at Nvidia.

Comparing the Rubin GPU and Groq 3 LPU side by side highlights the difference. The Rubin GPU has access to a whopping 288 gigabytes of HBM and is capable of 50 quadrillion floating-point operations per second (petaFLOPS) of 4-bit computation. The Groq 3 LPU contains a mere 500 megabytes of SRAM memory and is capable of 1.2 petaFLOPS of 8-bit computation. On the other hand, while the Rubin GPU has a memory bandwidth of 22 terabytes per second, at 150 TB/s the Groq 3 LPU is seven times as fast. The lean, speed-focused design is what allows the LPU to excel at inference.

The new inference chip underscores the ongoing trend of AI adoption, which shifts the computational load from just building ever bigger models to actually using those models at scale. “Nvidia’s announcement validates the importance of SRAM-based architectures for large-scale inference, and no one has pushed SRAM density further than d-Matrix,” says d-Matrix CEO Sid Sheth. He’s betting that data center customers will want a variety of processors for inference. “The winning systems will combine different types of silicon and fit easily into existing data centers alongside GPUs.”

Inference-only chips may not be the only solution. Late last week, Amazon Web Services said that it will deploy a new kind of inferencing system in its data centers. The system is a combination of AWS’s Tranium AI accelerator and Cerebras Systems’ third generation computer CS-3, which is built around the largest single chip ever made. The two-part system is meant to take advantage of a technique called inference disaggregation. It separates inference into two parts—processing the prompt, called prefill, and generating the output, called decode. Prefill is inherently parallel, computationally intensive, and doesn’t need much memory bandwidth, while decode is a more serial process that needs a lot of memory bandwidth. Cerebras has maximized the memory bandwidth issue by building 44 GB of SRAM on its chip connected by a 21 PB/s network.

Nvidia, too, intends to take advantage of inference disaggregation in its new compute rack, called the Nvidia Groq 3 LPX. Each tray within the rack will house 8 Groq 3 LPUs. The LPX will split the inference task with a Vera Rubin NVL72, Nvidia’s existing GPU and CPU rack. The prefill and the more computationally intensive parts of the decode are done on Vera Rubin, while the final part is done on the Groq 3 LPU, leveraging the strengths of each chip. “We’re in volume production now,” Huang said.

Correction on 4/8/26: a previous version of this article incorrectly stated that the Nvidia Groq 3 LPX contains a Vera Rubin chip in each tray. In fact, each tray contains 8 Groq 3 LPUs and no Vera Rubins, but the whole rack is designed to work in concert with an NVL72 rack, which houses Vera Rubin chips.

Worried that your latest ask to a cloud-based AI reveals a bit too much about you? Want to know your genetic risk of disease without revealing it to the services that compute the answer?

There is a way to do computing on encrypted data without ever having it decrypted. It’s called fully homomorphic encryption, or FHE. But there’s a rather large catch. It can take thousands—even tens of thousands—of times as long to compute on today’s CPUs and GPUs than simply working with the decrypted data.

So universities, startups, and at least one processor giant have been working on specialized chips that could close that gap. Last month at the IEEE International Solid-State Circuits Conference (ISSCC) in San Francisco, Intel demonstrated its answer, Heracles, which sped up FHE computing tasks as much as 5,000-fold compared to a top-of the-line Intel server CPU.

Startups are racing to beat Intel and each other to commercialization. But Sanu Mathew, who leads security circuits research at Intel, believes the CPU giant has a big lead, because its chip can do more computing than any other FHE accelerator yet built. “Heracles is the first hardware that works at scale,” he says.

The scale is measurable both physically and in compute performance. While other FHE research chips have been in the range of 10 square millimeters or less, Heracles is about 20 times that size and is built using Intel’s most advanced, 3-nanometer FinFET technology. And it’s flanked inside a liquid-cooled package by two 24-gigabyte high-bandwidth memory chips—a configuration usually seen only in GPUs for training AI.

RELATED: How to Compute with Data You Can’t See

In terms of scaling compute performance, Heracles showed muscle in live demonstrations at ISSCC. At its heart the demo was a simple private query to a secure server. It simulated a request by a voter to make sure that her ballot had been registered correctly. The state, in this case, has an encrypted database of voters and their votes. To maintain her privacy, the voter would not want to have her ballot information decrypted at any point; so using FHE, she encrypts her ID and vote and sends it to the government database. There, without decrypting it, the system determines if it is a match and returns an encrypted answer, which she then decrypts on her side.

On an Intel Xeon server CPU, the process took 15 milliseconds. Heracles did it in 14 microseconds. While that difference isn’t something a single human would notice, verifying 100 million voter ballots adds up to more than 17 days of CPU work versus a mere 23 minutes on Heracles.

Looking back on the five-year journey to bring the Heracles chip to life, Ro Cammarota, who led the project at Intel until last December and is now at University of California, Irvine, says “We have proven and delivered everything that we promised.”

FHE Data Expansion

FHE is fundamentally a mathematical transformation, sort of like the Fourier transform. It encrypts data using a quantum-computer-proof algorithm, but, crucially, uses corollaries to the mathematical operations usually used on unencrypted data. These corollaries achieve the same ends on the encrypted data.

One of the main things holding such secure computing back is the explosion in the size of the data once it’s encrypted for FHE, Anupam Golder, a research scientist at Intel’s circuits research lab, told engineers at ISSCC. “Usually, the size of cipher text is the same as the size of plain text, but for FHE it’s orders of magnitude larger,” he said.

While the sheer volume is a big problem, the kinds of computing you need to do with that data is also an issue. FHE is all about very large numbers that must be computed with precision. While a CPU can do that, it’s very slow going—integer addition and multiplication take about 10,000 more clock cycles in FHE. Worse still, CPUs aren’t built to do such computing in parallel. Although GPUs excel at parallel operations, precision is not their strong suit. (In fact, from generation to generation, GPU designers have devoted more and more of the chip’s resources to computing less-and-less-precise numbers.)

FHE also requires some oddball operations with names like “twiddling” and “automorphism,” and it relies on a compute-intensive noise-cancelling process called bootstrapping. None of these things are efficient on a general-purpose processor. So, while clever algorithms and libraries of software cheats have been developed over the years, the need for a hardware accelerator remains if FHE is going to tackle large-scale problems, says Cammarota.

The Labors of Heracles

Heracles was initiated under a Defense Advanced Research Projects Agency (DARPA) program five years ago to accelerate FHE using purpose-built hardware. It was developed as “a whole system-level effort that went all the way from theory and algorithms down to the circuit design,” says Cammarota.

Among the first problems was how to compute with numbers that were larger than even the 64-bit words that are today a CPU’s most precise. There are ways to break up these gigantic numbers into chunks of bits that can be calculated independently of each other, providing a degree of parallelism. Early on, the Intel team made a big bet that they would be able to make this work in smaller, 32-bit chunks, yet still maintain the needed precision. This decision gave the Heracles architecture some speed and parallelism, because the 32-bit arithmetic circuits are considerably smaller than 64-bit ones, explains Cammarota.

At Heracles’s heart are 64 compute cores—called tile-pairs—arranged in an eight-by-eight grid. These are what are called single instruction multiple data (SIMD) compute engines designed to do the polynomial math, twiddling, and other things that make up computing in FHE and to do them in parallel. An on-chip 2D mesh network connects the tiles to each other with wide, 512-byte, buses.

RELATED: Tech Keeps Chatbots From Leaking Your Data

Important to making encrypted computing efficient is feeding those huge numbers to the compute cores quickly. The sheer amount of data involved meant linking 48-GB-worth of expensive high-bandwidth memory to the processor with 819-GB-per-second connections. Once on the chip, data musters in 64 megabytes of cache memory—somewhat more than an Nvidia Hopper-generation GPU. From there it can flow through the array at 9.6 terabytes per second by hopping from tile-pair to tile-pair.

To ensure that computing and moving data don’t get in each other’s way, Heracles runs three synchronized streams of instructions simultaneously, one for moving data onto and off of the processor, one for moving data within it, and a third for doing the math, Golder explained.

It all adds up to some massive speedups, according to Intel. Heracles—operating at 1.2 gigahertz—takes just 39 microseconds to do FHE’s critical math transformation, a 2,355-fold improvement over an Intel Xeon CPU running at 3.5 GHz. Across seven key operations, Heracles was 1,074 to 5,547 times as fast.

The differing ranges have to do with how much data movement is involved in the operations, explains Mathew. “It’s all about balancing the movement of data with the crunching of numbers,” he says.

FHE Competition

“It’s very good work,” Kurt Rohloff, chief technology officer at FHE software firm Duality Technology, says of the Heracles results. Duality was part of a team that developed a competing accelerator design under the same DARPA program that brought forth Intel’s Heracles. “When Intel starts talking about scale, that usually carries quite a bit of weight.”

Duality’s focus is less on new hardware than on software products that do the kind of encrypted queries that Intel demonstrated at ISSCC. At the scale in use today “there’s less of a need for [specialized] hardware,” says Rohloff. “Where you start to need hardware is emerging applications around deeper machine-learning-oriented operations like neural net, LLMs, or semantic search.”

Last year, Duality demonstrated an FHE-encrypted language model called BERT. Like more famous LLMs such as ChatGPT, BERT is a transformer model. However it’s only one-tenth the size of even the most compact LLMs.

John Barrus, vice president of product at Dayton, Ohio–based Niobium Microsystems, an FHE chip startup spun out of another DARPA competitor, agrees that encrypted AI is a key target of FHE chips. “There are a lot of smaller models that, even with FHE’s data expansion, will run just fine on accelerated hardware,” he says.

With no stated commercial plans from Intel, Niobium expects its chip to be “the world’s first commercially viable FHE accelerator, designed to enable encrypted computations at speeds practical for real-world cloud and AI infrastructure.” Although it hasn’t announced when a commercial chip will be available, last month the startup revealed that it had inked a deal worth 10 billion South Korean won (US $6.9 million) with Seoul-based chip design firm Semifive to develop the FHE accelerator for fabrication using Samsung Foundry’s 8-nanometer-process technology.

Other startups including Cornami, Fabric Cryptography, and Optalysys have been working on chips to accelerate FHE. Optalysys CEO Nick New says Heracles hits about the level of speedup you could hope for using an all-digital system. “We’re looking at pushing way past that digital limit,” he says. His company’s approach is to use the physics of a photonic chip to do FHE’s compute-intensive transform steps. That photonics chip is on its seventh generation, he says, and among the next steps is to 3D integrate it with custom silicon to do the nontransform steps and coordinate the whole process. A full 3D-stacked commercial chip could be ready in two or three years, says New.

While competitors develop their chips, so will Intel, says Mathew. It will be improving on how much the chip can accelerate computations by fine-tuning the software. It will also be trying out more massive FHE problems, and exploring hardware improvements for a potential next generation. “This is like the first microprocessor…the start of a whole journey,” says Mathew.

Explore structured finite-element methodologies for analyzing transformer behavior under harmonic and transient conditions — covering modelling, solver configuration, and result validation techniques.

What Attendees will Learn

1. How FEM enables pre-fabrication performance evaluation — Assess magnetic field distribution, current behavior, and turns-ratio accuracy through simulation rather than physical testing.
2. How harmonic analysis uncovers saturation and imbalance — Identify high-flux regions and current asymmetries that analytical methods may not capture.
3. How transient simulations characterize dynamic response — Examine time-domain current waveforms, inrush behavior, and multi-cycle stabilization.
4. How modelling choices affect simulation fidelity — Understand the impact of coil definitions, winding configurations, solver type, and material models on accuracy.

Download this free whitepaper now!

Move over, Pixar. The ants that animators once morphed into googly-eyed caricatures in films such as A Bug’s Life and Antz just received a meticulously precise anatomical reboot.

Writing today in Nature Methods, an international team of entomologists, accelerator physicists, computer scientists, and biological-imaging specialists describe a new 3D atlas of ant morphology.

Dubbed Antscan, the platform features micrometer-resolution reconstructions that lay bare not only the insects’ armored exoskeletons but also their muscles, nerves, digestive tracts, and needlelike stingers poised at the ready.

Those high-resolution images—spanning 792 species across 212 genera and covering the bulk of described ant diversity—are now available free of charge through an interactive online portal, where anyone can rotate, zoom, and virtually “dissect” the insects from a laptop.

“Antscan is exciting!” says Cameron Currie, an evolutionary biologist at McMaster University in Hamilton, Ont., Canada, who was not involved in the research. “It provides an outstanding resource for comparative work across ants.”

Digital Access to Natural History Collections

It also provides broader access to natural history collections.

No longer must these vast archives of preserved life be confined to drawers and jars in museums scattered around the world, available only to specialists able to visit in person. All these specimens can now be explored digitally by anyone with an internet connection, adding fresh scientific value to museum holdings.

“The more people that access and work with the stuff in our museums, whether it’s physically or digitally, the greater value they add,” says David Blackburn, the curator of herpetology at the Florida Museum of Natural History who, like Currie, was not involved in the research.

Some of those people may be professional myrmecologists (scientists who specialize in the study of ants) and fourmiculture (ant-farming) enthusiasts. But others may be schoolteachers, video-game designers, tattoo artists, or curious members of the public.

“It is an extremely rich dataset that can be used for a number of different applications in science, but also for the arts and outreach and education.” says Julian Katzke, an entomologist at the National Museum of Natural History in Washington, D.C.

Card-carrying members of IEEE should find plenty to explore in Antscan as well, says Evan Economo, a biodiversity scientist at the University of Maryland in College Park who, along with Katzke, co-led the project. With the dataset now publicly available and standardized at scale, “I would really like to see these big libraries of organismal form one day be useful for people in robotics and engineering, so they can mine these data for new kinds of biomechanical designs,” he says.

These renderings reveal different structures within the body of an army ant (Eciton hamatum) subsoldier, based on Antscan data.Katzke et al.

Advancements in Ant Imaging Technology

Researchers have been digitizing natural history collections for years: photographing drawers of pinned specimens, building surface-level models from overlapping image stacks, and using computed tomography (CT) to scan select species one at a time. But those efforts are typically slow, piecemeal, and often limited to external features.

To capture entire organisms, inside and out, Economo and his team—then based at the Okinawa Institute of Science and Technology in Japan and including former lab members Katzke and Francisco Hita Garcia (now at the Museum für Naturkunde in Berlin)—built an automated imaging pipeline that effectively turned a particle accelerator into an anatomical assembly line.

They scoured museum collections around the world for ant specimens—workers, queens, and males alike—and sent some 2,200 preserved samples through a pair of micro-CT beamlines at the Karlsruhe Institute of Technology’s synchrotron light source facility in Germany.

There, biological imaging specialist Thomas van de Kamp oversaw the operation, as intense X-ray beams swept through each specimen and high-speed detectors recorded thousands of projection images from multiple angles. Robotic handlers moved vials of alcohol-preserved ants into position, one after another, all in a matter of days.

Software then reconstructed 200-plus terabytes of data generated into 3D volumes, with neural networks helping to automate the identification and analysis of anatomical structures.

Similar large-scale digitization efforts—such as the openVertebrate Project, led by the Florida Museum of Natural History’s Blackburn, which involved scanning thousands of birds, fish, mammals, reptiles, and amphibians—have begun transforming how biologists study anatomy. But applying conventional micro-CT at comparable scale to insects, which are smaller and harder to scan at useful resolutions, required a leap in speed and throughput.

That’s where the synchrotron came in. By harnessing a particle accelerator to generate extraordinarily bright, coherent X-rays, the team was able to capture high-resolution internal anatomy in seconds, without the lengthy staining or other preprocessing steps often required for soft-tissue contrast in standard lab scanners.

“It is an impressive piece of work,” says Vladimir Blagoderov, principal curator of invertebrates at the National Museums Scotland in Edinburgh, who was not involved in the research. “This project adds an industrial dimension to CT scanning by combining robotics, standardized sampling, automated image-processing pipelines, and machine learning.”

The sheer taxonomic breadth of the Antscan dataset now makes it possible to spot patterns across the entire ant family tree, as Economo and his colleagues have already demonstrated.

In a separate paper published last December, for example, the researchers drew on the newly generated scans to measure how much ants invest in their outer protective casing. Reporting in Science Advances, they showed that species with lighter, less costly cuticles tend to form larger colonies and diversify more rapidly over evolutionary time.

In their latest study, the Antscan team turned to a different evolutionary question: The distribution of a biomineral “armor” layer first described by Currie and his colleagues in 2020 in a Central American leaf-cutter ant. A quick sweep through the Antscan database revealed that this coating—which absorbs X-rays and is visible as a bright sheath over the cuticle—is not an oddity confined to one species.

Instead, it is common among fungus-farming ants, the evolutionary lineage from which leaf-cutting ants arose roughly 20 million years ago, but largely absent in most other branches of the ant tree. (Currie’s team independently confirmed the pattern using X-ray diffraction, a technique that can precisely reveal a material’s mineral composition, as the group reported last month in a preprint posted to bioRxiv.)

Those are only early demonstrations of what the database can do, though. And with AI tools increasingly capable of parsing enormous, information-rich data troves, the real analytical power of Antscan may still lie ahead, says Marek Borowiec, director of the C.P. Gillette Museum of Arthropod Diversity at Colorado State University, who has chronicled the rise of deep learning tools in ecology and evolution.

“The full advantage of this dataset will be realized when these methods are deployed,” he says.

Transforming Morphology with Antscan

The ambitions behind Antscan extend well beyond ant biology. Economo and his colleagues see it as a blueprint for digitizing, standardizing, and scaling anatomy itself.

Just as large-scale sequencing projects and genomic databases transformed the study of DNA over the past two decades, they hope Antscan will catalyze a comparable shift for morphology. “This is kind of like having a genome for shape,” Economo says.

Museum collections house millions of alcohol-preserved insects and other small invertebrates—beetles, flies, wasps, spiders, crustaceans—many of them representing rare or extinct populations. Following the Antscan playbook, each could be converted into a high-resolution library of “digital twins.“

In each case, synchrotron micro-CT would offer a rapid way to peer inside fragile specimens without cutting them open, capturing both hard exoskeleton and soft tissue in exquisite detail across vast swaths of biological diversity.

Antscan/YouTube

Beam time at major synchrotron facilities is scarce and fiercely competitive, a practical bottleneck for any effort to digitize biodiversity at scale, notes National Museums Scotland’s Blagoderov. What’s more, “even once the scans exist, the downstream burden is nontrivial: Moving, storing, and processing hundreds of terabytes of data can become a bottleneck in its own right,” he says.

But if access can be secured and the computational infrastructure scaled to match, such efforts could transform natural history museums from static repositories into dynamic digital biomes.

That transformation may prove especially important at a time of accelerating species loss on Earth. By capturing organisms in extraordinary detail, resources like Antscan create a permanent, high-resolution record of life’s architecture—an anatomical time capsule that can be queried and revisited long after fragile specimens degrade or wild populations vanish.

And should Pixar ever greenlight A Bug’s Life 2 (suggested title: Even Buggier), the studio’s character designers may not need to take much artistic license at all. Thanks to a particle accelerator and a small cadre of dedicated scientists, the reference models are already in hand—rendered not in animation software but in micrometer-perfect anatomical form.

When Ukrainian mathematician Maryna Viazovska received a Fields Medal—widely regarded as the Nobel Prize for mathematics—in July 2022, it was big news. Not only was she the second woman to accept the honor in the award’s 86-year history, but she collected the medal just months after her country had been invaded by Russia. Nearly four years later, Viazovska is making waves again. Today, in a collaboration between humans and AI, Viazovska’s proofs have been formally verified, signaling rapid progress in AI’s abilities to assist with mathematical research.

“These new results seem very, very impressive, and definitely signal some rapid progress in this direction,” says AI-reasoning expert and Princeton University postdoc Liam Fowl, who was not involved in the work.

In her Fields Medal–winning research, Viazovska had tackled two versions of the sphere-packing problem, which asks: How densely can identical circles, spheres, et cetera, be packed in n-dimensional space? In two dimensions, the honeycomb is the best solution. In three dimensions, spheres stacked in a pyramid are optimal. But after that, it becomes exceedingly difficult to find the best solution, and to prove that it is in fact the best.

In 2016, Viazovska solved the problem in two cases. By using powerful mathematical functions known as (quasi-)modular forms, she proved that a symmetric arrangement known as E₈ is the best 8-dimensional packing, and soon after proved with collaborators that another sphere packing called the Leech lattice is best in 24 dimensions. Though seemingly abstract, this result has potential to help solve everyday problems related to dense sphere packing, including error-correcting codes used by smartphones and space probes.

The proofs were verified by the mathematical community and deemed correct, leading to the Fields Medal recognition. But formal verification—the ability of a proof to be verified by a computer—is another beast altogether. Since 2022, much progress has been made in AI-assisted formal proof verification.

Serendipity leads to formalization project

A few years later, a chance meeting in Lausanne, Switzerland, between third-year undergraduate Sidharth Hariharan and Viazovska would reignite her interest in sphere-packing proofs. Though still very early in his career, Hariharan was already becoming adept at formalizing proofs.

“Formal verification of a proof is like a rubber stamp,” Fowl says. “It’s a kind of bona fide certification that you know your statements of reasoning are correct.”

Hariharan told Viazovska how he had been using the process of formalizing proofs to learn and really understand mathematical concepts. In response, Viazovska expressed an interest in formalizing her proofs, largely out of curiosity. From this, in March 2024 the Formalising Sphere Packing in Lean project was born. Lean is a popular programming language and “proof assistant” that allows mathematicians to write proofs that are then verified for absolute correctness by a computer.

A collaboration formed to write a human-readable “blueprint” that could be used to map the 8-dimensional proof’s various constituents and figure out which of them had and had not been formalized and/or proven, and then prove and formalize those missing elements in Lean.

“We had been building the project’s repository for about 15 months when we enabled public access in June 2025,” recalls Hariharan, now a first-year Ph.D. student at Carnegie Mellon University. “Then, in late October we heard from Math, Inc. for the first time.”

The AI speedup

Math, Inc. is a startup developing Gauss, an AI specifically designed to automatically formalize proofs. “It’s a particular kind of language model called a reasoning agent that’s meant to interleave both traditional natural-language reasoning and fully formalized reasoning,” explains Jesse Han, Math, Inc. CEO and cofounder. “So it’s able to conduct literature searches, call up tools, and use a computer to write down Lean code, take notes, spin up verification tooling, run the Lean compiler, et cetera.”

Math, Inc. first hit the headlines when it announced that Gauss had completed a Lean formalization of the strong prime number theorem (PNT) in three weeks last summer, a task that Fields Medalist Terence Tao and Alex Kontorovich had been working on. Similarly, Math, Inc. contacted Hariharan and colleagues to say that Gauss had proven several facts related to their sphere-packing project.

“They told us that they had finished 30 “sorrys,” which meant that they proved 30 intermediate facts that we wanted proved,” explains Hariharan. A proportion of these sorrys were shared with the project team and merged with their own work. “One of them helped us identify a typo in our project, which we then fixed,” adds Hariharan. “So it was a pretty fruitful collaboration.”

From 8 to 24 dimensions

But then, radio silence followed. Math, Inc. appeared to lose interest. However, while Hariharan and colleagues continued their labor of love, Math, Inc. was building a new and improved version of Gauss. “We made a research breakthrough sometime mid-January that produced a much stronger version of Gauss,” says Han. “This new version reproduced our three-week PNT result in two to three days.”

Days later, the new Gauss was steered back to the sphere-packing formalization. Working from the invaluable preexisting blueprint and work that Hariharan and collaborators had shared, Gauss not only autoformalized the 8-dimensional case, but also found and fixed a typo in the published paper, all in the space of five days.

“When they reached out to us in late January saying that they finished it, to put it very mildly, we were very surprised,” says Hariharan. “But at the end of the day, this is technology that we’re very excited about, because it has the capability to do great things and to assist mathematicians in remarkable ways.”

Hariharan was working on sphere-packing proof verification as the sun was setting behind Carnegie Mellon’s Hamerschlag Hall.Sidharth Hariharan

The 8-dimensional sphere-packing proof formalization alone, announced on February 23, represents a watershed moment for autoformalization and AI–human collaboration. But today, Math, Inc. revealed an even more impressive accomplishment: Gauss has autoformalized Viazovska’s 24-dimensional sphere-packing proof—all 200,000+ lines of code of it—in just two weeks.

There are commonalities between the 8- and 24-dimensional cases in terms of the foundational theory and overall architecture of the proof, meaning some of the code from the 8-dimensional case could be refactored and reused. However, Gauss had no preexisting blueprint to work from this time. “And it was actually significantly more involved than the 8-dimensional case, because there was a lot of missing background material that had to be brought on line surrounding many of the properties of the Leech lattice, in particular its uniqueness,” explains Han.

Though the 24-dimensional case was an automated effort, both Han and Hariharan acknowledge the many contributions from humans that laid the foundations for this achievement, regarding it as a collaborative endeavor overall between humans and AI.

But for Han, it represents even more: the beginning of a revolutionary transformation in mathematics, where extremely large-scale formalizations are commonplace. “A programmer used to be someone who punched holes into cards, but then the act of programming became separated from whatever material substrate was used for recording programs,” he concludes. “I think the end result of technology like this will be to free mathematicians to do what they do best, which is to dream of new mathematical worlds.”

Sometimes a visually compelling metaphor is all you need to get an otherwise complicated idea across. In the summer of 2001, a Tulane physics professor named John P. Perdew came up with a banger. He wanted to convey the hierarchy of computational complexity inherent in the behavior of electrons in materials. He called it “Jacob’s Ladder.” He was appropriating an idea from the Book of Genesis, in which Jacob dreamed of a ladder “set up on the earth, and the top of it reached to heaven. And behold the angels of God ascending and descending on it.”

Jacob’s Ladder represented a gradient and so too did Perdew’s ladder, not of spirit but of computation. At the lowest rung, the math was the simplest and least computationally draining, with materials represented as a smoothed-over, cartoon version of the atomic realm. As you climbed the ladder, using increasingly more intensive mathematics and compute power, descriptions of atomic reality became more precise. And at the very top, nature was perfectly described via impossibly intensive computation—something like what God might see.

With this metaphor in mind, we propose to extend Jacob’s Ladder beyond Perdew’s version, to encompass all computational approaches to simulating the behavior of electrons. And instead of climbing rung by rung toward an unreachable summit, we have an idea to bend the ladder so that even the very top lies within our grasp. Specifically, we at Microsoft envision a hybrid approach. It starts with using quantum computers to generate exquisitely accurate data about the behavior of electrons—data that would be prohibitively expensive to compute classically. This quantum-generated data will then train AI models running on classical machines, which can predict the properties of materials with remarkable speed. By combining quantum accuracy with AI-driven speed, we can ascend Jacob’s Ladder faster, designing new materials with novel properties and at a fraction of the cost.

At the base of Jacob’s Ladder are classical models that treat atoms as simple balls connected by springs—fast enough to handle millions of atoms over long times but with the lowest precision. Moving up along the black line, semiempirical methods add some quantum mechanical calculations. Next are approximations based on Hartree-Fock (HF) and density functional theory (DFT), which include full quantum behavior of individual electrons but model their interactions in an averaged way. The greater accuracy requires significant computing power, which limits them to simulating molecules with no more than a few hundred atoms. At the top are coupled-cluster and full configuration interaction (FCI) methods—exquisitely accurate but, at the moment, restricted to tiny molecules or subsets of electrons due to the large computational costs involved. Quantum computing can bend the accuracy-versus-cost curve at the top [orange line], making highly accurate calculations feasible for large systems. AI, trained on this quantum-accurate data, can flatten this curve [purple line], enabling rapid predictions for similar systems at a fraction of the cost of classical computing.

In our approach, the base of Jacob’s Ladder still starts with classical models that treat atoms as simple balls connected by springs—models that are fast enough to handle millions of atoms over long times, but with the lowest precision. As we ascend the ladder, some quantum mechanical calculations are added to semiempirical methods. Eventually, we’ll get to the full quantum behavior of individual electrons but with their interactions modeled in an averaged way; this greater accuracy requires significant compute power, which means you can only simulate molecules of no more than a few hundred atoms. At the top will be the most computationally intensive methods—prohibitively expensive on classical computers but tractable on quantum computers.

In the coming years, quantum computing and AI will become critical tools in the pursuit of new materials science and chemistry. When combined, their forces will multiply. We believe that by using quantum computers to train AI on quantum data, the result will be hyperaccurate AI models that can reach ever higher rungs of computational complexity without the prohibitive computational costs.

This powerful combination of quantum computing and AI could unlock unprecedented advances in chemical discovery, materials design, and our understanding of complex reaction mechanisms. Chemical and materials innovations already play a vital—if often invisible—role in our daily lives. These discoveries shape the modern world: new drugs to help treat disease more effectively, improving health and extending life expectancy; everyday products like toothpaste, sunscreen, and cleaning supplies that are safe and effective; cleaner fuels and longer-lasting batteries; improved fertilizers and pesticides to boost global food production; and biodegradable plastics and recyclable materials to shrink our environmental footprint. In short, chemical discovery is a behind-the-scenes force that greatly enhances our everyday lives.

The potential is vast. Anywhere AI is already in use, this new quantum-enhanced AI could drastically improve results. These models could, for instance, scan for previously unknown catalysts that could fix atmospheric carbon and so mitigate climate change. They could discover novel chemical reactions to turn waste plastics into useful raw materials and remove toxic “forever chemicals” from the environment. They could uncover new battery chemistries for safer, more compact energy storage. They could supercharge drug discovery for personalized medicine.

And that would just be the beginning. We believe quantum-enhanced AI will open up new frontiers in materials science and reshape our ability to understand and manipulate matter at its most fundamental level. Here’s how.

How Quantum Computing Will Revolutionize Chemistry

To understand how quantum computing and AI could help bend Jacob’s Ladder, it’s useful to look at the classical approximation techniques that are currently used in chemistry. In atoms and molecules, electrons interact with one another in complex ways called electron correlations. These correlations are crucial for accurately describing chemical systems. Many computational methods, such as density functional theory (DFT) or the Hartree-Fock method, simplify these interactions by replacing the intricate correlations with averaged ones, assuming that each electron moves within an average field created by all other electrons. Such approximations work in many cases, but they can’t provide a full description of the system.

A joint project between Microsoft and Pacific Northwest National Laboratory used AI and high-performance computing to identify potential materials for battery electrolytes. The most promising were synthesized [top and middle] and tested [bottom] at PNNL. Dan DeLong/Microsoft

Electron correlation is particularly important in systems where the electrons are strongly interacting—as in materials with unusual electronic properties, like high-temperature superconductors—or when there are many possible arrangements of electrons with similar energies—such as compounds containing certain metal atoms that are crucial for catalytic processes.

In these cases, the simplified approach of DFT or Hartree-Fock breaks down, and more sophisticated methods are needed. As the number of possible electron configurations increases, we quickly reach an “exponential wall” in computational complexity, beyond which classical methods become infeasible.

Enter the quantum computer. Unlike classical bits, which are either on or off, qubits can exist in superpositions—effectively coexisting in multiple states simultaneously. This should allow them to represent many electron configurations at once, mirroring the complex quantum behavior of correlated electrons. Because quantum computers operate on the same principles as the electron systems they will simulate, they will be able to accurately simulate even strongly correlated systems—where electrons are so interdependent that their behavior must be calculated collectively.

AI’s Role in Advancing Computational Chemistry

At present, even the computationally cheap methods at the bottom of Jacob’s Ladder are slow, and the ones higher up the ladder are slower still. AI models have emerged as powerful accelerators to such calculations because they can serve as emulators that predict simulation outcomes without running the full calculations. The models can speed up the time it takes to solve problems up and down the ladder by orders of magnitude.

This acceleration opens up entirely new scales of scientific exploration. In 2023 and 2024, we collaborated with researchers at Pacific Northwest National Laboratory (PNNL) on using advanced AI models to evaluate over 32 million potential battery materials, looking for safer, cheaper, and more environmentally friendly options. This enormous pool of candidates would have taken about 20 years to explore using traditional methods. And yet, within less than a week, that list was narrowed to 500,000 stable materials and then to 800 highly promising candidates. Throughout the evaluation, the AI models replaced expensive and time-consuming quantum chemistry calculations, in some cases delivering insights half a million times as fast as would otherwise have been the case.

We then used high-performance computing (HPC) to validate the most promising materials with DFT and AI-accelerated molecular dynamics simulations. The PNNL team then spent about nine months synthesizing and testing one of the candidates—a solid-state electrolyte that uses sodium, which is cheap and abundant, and some other materials, with 70 percent less lithium than conventional lithium-ion designs. The team then built a prototype solid-state battery that they tested over a range of temperatures.

This potential battery breakthrough isn’t unique. AI models have also dramatically accelerated research in climate science, fluid dynamics, astrophysics, protein design, and chemical and biological discovery. By replacing traditional simulations that can take days or weeks to run, AI is reshaping the pace and scope of scientific research across disciplines.

However, these AI models are only as good as the quality and diversity of their training data. Whether sourced from high-fidelity simulations or carefully curated experimental results, these data must accurately represent the underlying physical phenomena to ensure reliable predictions. Poor or biased data can lead to misleading outcomes. By contrast, high-quality, diverse datasets—such as those full-accuracy quantum simulations—enable models to generalize across systems and uncover new scientific insights. This is the promise of using quantum computing for training AI models.

How to Accelerate Chemical Discovery

The real breakthrough will come from strategically combining quantum computing’s and AI’s unique strengths. AI already excels at learning patterns and making rapid predictions. Quantum computers, which are still being scaled up to be practically useful, will excel at capturing electron correlations that classical computers can only approximate. So if you train classical models on quantum-generated data, you’ll get the best of both worlds: the accuracy of quantum delivered at the speed of AI.

As we learned from the Microsoft-PNNL collaboration on electrolytes, AI models alone can greatly speed up chemical discovery. In the future, quantum-accurate AI models will tackle even bigger challenges. Consider the basic discovery process, which we can think of as a funnel. Scientists begin with a vast pool of candidate molecules or materials at the wide-mouthed top, narrowing them down using filters based on desired properties—such as boiling point, conductivity, viscosity, or reactivity. Crucially, the effectiveness of this screening process depends heavily on the accuracy of the models used to predict these properties. Inaccurate predictions can create a “leaky” funnel, where promising candidates are mistakenly discarded or poor ones are mistakenly advanced.

Quantum-accurate AI models will dramatically improve the precision of chemical-property predictions. They’ll be able to help identify “first-time right” candidates, sending only the most promising molecules to the lab for synthesis and testing—which will save both time and cost.

Another key aspect of the discovery process is understanding the chemical reactions that govern how new substances are formed and behave. Think of these reactions as a network of roads winding through a mountainous landscape, where each road represents a possible reaction step, from starting materials to final products. The outcome of a reaction depends on how quickly it travels down each path, which in turn is determined by the energy barriers along the way—like mountain passes that must be crossed. To find the most efficient route, we need accurate calculations of these barrier heights, so that we can identify the lowest passes and chart the fastest path through the reaction landscape.

Even small errors in estimating these barriers can lead to incorrect predictions about which products will form. Case in point: A slight miscalculation in the energy barrier of an environmental reaction could mean the difference between labeling a compound a “forever chemical” or one that safely degrades over time.

Accurate modeling of reaction rates is also essential for designing catalysts—substances that speed up and steer reactions in desired directions. Catalysts are crucial in industrial chemical production, carbon capture, and biological processes, among many other things. Here, too, quantum-accurate AI models can play a transformative role by providing the high-fidelity data needed to predict reaction outcomes and design better catalysts.

Once trained, these AI models, powered by quantum-accurate data, will revolutionize computational chemistry by delivering quantum-level precision. And once the AI models, which run on classical computers, are trained with quantum computing data, researchers will be able to run high-accuracy simulations on laptops or desktop computers, rather than relying on massive supercomputers or future quantum hardware. By making advanced chemical modeling more accessible, these tools will democratize discovery and empower a broader community of scientists to tackle some of the most pressing challenges in health, energy, and sustainability.

Remaining Challenges for AI and Quantum Computing

By now, you’re probably wondering: When will this transformative future arrive? It’s true that quantum computers still struggle with error rates and limited lifetimes of usable qubits. And they still need to scale to the size required for meaningful chemistry simulations. Meaningful chemistry simulations beyond the reach of classical computation will require hundreds to thousands of high-quality qubits with error rates of around 10^-15, or one error in a quadrillion operations. Achieving this level of reliability will require fault tolerance through redundant encoding of quantum information in logical qubits, each consisting of hundreds of physical qubits, thus requiring a total of about a million physical qubits. Current AI models for chemical-property predictions may not have to be fully redesigned. We expect that it will be sufficient to start with models pretrained on classical data and then fine-tune them with a few results from quantum computers.

Despite some open questions, the potential rewards in terms of scientific understanding and technological breakthroughs make our proposal a compelling direction for the field. The quantum computing industry has begun to move beyond the early noisy prototypes, and high-fidelity quantum computers with low error rates could be possible within a decade.

Realizing the full potential of quantum-enhanced AI for chemical discovery will require focused collaboration between chemists and materials scientists who understand the target problems, experts in quantum computing who are building the hardware, and AI researchers who are developing the algorithms. Done right, quantum-enhanced AI could start to tackle the world’s toughest challenges—from climate change to disease—years ahead of anyone’s expectations.

In the time it takes you to read this sentence, the Large Hadron Collider (LHC) will have smashed billions of particles together. In all likelihood, it will have found exactly what it found yesterday: more evidence to support the Standard Model of particle physics.

For the engineers who built this 27-kilometer-long ring, this consistency is a triumph. But for theoretical physicists, it has been rather frustrating. As Matthew Hutson reports in “AI Hunts for the Next Big Thing in Physics,” the field is currently gripped by a quiet crisis. In an email discussing his reporting, Hutson explains that the Standard Model, which describes the known elementary particles and forces, is not a complete picture. “So theorists have proposed new ideas, and experimentalists have built giant facilities to test them, but despite the gobs of data, there have been no big breakthroughs,” Hutson says. “There are key components of reality we’re completely missing.”

That’s why researchers are turning artificial intelligence loose on particle physics. They aren’t simply asking AI to comb through accelerator data to confirm existing theories, Hutson explains. They’re asking AI to point the way toward theories that they’ve never imagined. “Instead of looking to support theories that humans have generated,” he says, “unsupervised AI can highlight anything out of the ordinary, expanding our reach into unknown unknowns.” By asking AI to flag anomalies in the data, researchers hope to find their way to “new physics” that extends the Standard Model.

On the surface, this article might sound like another “AI for X” story. As IEEE Spectrum’s AI editor, I get a steady stream of pitches for such stories: AI for drug discovery, AI for farming, AI for wildlife tracking. Often what that really means is faster data processing or automation around the edges. Useful, sure, but incremental.

What struck me in Hutson’s reporting is that this effort feels different. Instead of analyzing experimental data after the fact, the AI essentially becomes part of the instrument, scanning for subtle patterns and deciding in real time what’s interesting. At the LHC, detectors record 40 million collisions per second. There’s simply no way to preserve all that data, so engineers have always had to build filters to decide which events get saved for analysis and which are discarded; nearly everything is thrown away.

Now those split-second decisions are increasingly handed to machine learning systems running on field-programmable gate arrays (FPGAs) connected to the detectors. The code must run on the chip’s limited logic and memory, and compressing a neural network into that hardware isn’t easy. Hutson describes one theorist pleading with an engineer, “Which of my algorithms fits on your bloody FPGA?”

This moment is part of a much older pattern. As Hutson writes in the article, new instruments have opened doors to the unexpected throughout the history of science. Galileo’s telescope revealed moons circling Jupiter. Early microscopes exposed entire worlds of “animalcules” swimming around. These better tools didn’t just answer existing questions; they made it possible to ask new ones.

If there’s a crisis in particle physics, in other words, it may not just be about missing particles. It’s about how to look beyond the limits of the human imagination. Hutson’s story suggests that AI might not solve the mysteries of the universe outright, but it could change how we search for answers.

Late last year, Sandia National Laboratories started testing an unusual type of supercomputer. Unlike conventional supercomputers, which consist of large interconnected clusters of CPUs and GPUs, this machine incorporates reconfigurable accelerators that optimize their operations for the particular computation that’s being run. This new architecture, which is similar to field-programmable gate arrays (FPGAs), is built by startup NextSilicon. A key benefit of the approach is that it doesn’t require a software rewrite: the hardware optimizes itself for the software, not vice versa.

Spectra, which incorporates 128 NextSilicon Maverick-2 accelerators, is still in the investigative phase, says program leader and Sandia senior scientist James Laros. NextSilicon, which has headquarters in Tel Aviv and Minneapolis, claims its accelerators generally use half as much power as Nvidia’s Blackwell while offering a quadruple speed advantage. The power and speed vary depending on the particular application.

The Power of Flexibility

NextSilicon CEO Elad Raz says typical architectures predict the next instruction then fetch and cache data. “What if you can remove all that overhead?” he wondered. “A lot of people are trying to build a new CPU or a better GPU. Other companies have a software solution,” says Raz. “I wanted to build something with software and hardware collaborating together.”

The company’s Maverick-2 first runs the application on a CPU and identifies which operations run most frequently. Then, it reconfigures the chip to schedule its work in a way that optimizes data flow. Instead of back-and-forth fetching of data, he says, “you can generate a pipeline.”

And a key advantage of the company’s design is that users do not have to rewrite their software in order to run it more efficiently on the system. The hardware adapts to the software, not vice versa.

Most of the applications Sandia runs are constrained by memory bandwidth, says Laros. “What if we can go faster because we don’t have to go back to the main memory?” That’s the potential of the Spectra architecture.

Raz says Maverick uses half as much power as Nvidia’s Blackwell and can perform HPCG, a supercomputing benchmark, twice as fast; it performs PageRank, another benchmark, 10 times as fast. Sandia scientists are currently assessing Spectra’s performance when running molecular dynamics simulations, which predict the movements of atoms and are widely used in physics and materials science, and other core codes used by the U.S. Department of Energy. “Where it will provide a benefit is if we can get better performance for types of apps that don’t run well on GPUs, or if we can get the same performance with better energy efficiency,” Laros says.

Supporting the Mission Through Experimentation

Sandia performs computer simulations to maintain the United States’ nuclear arsenal. “We’ve replaced testing with simulation and computing,” Laros says. Because of the high stakes of this mission, the lab has to “make sure we’re not putting all our eggs in one basket,” he says. If a company whose technology the U.S. government relies on for nuclear stockpile stewardship should go out of business, the government needs to have alternatives. “We maintain a pipeline of overlapping technologies,” Laros says.

Spectra is part of Sandia’s Vanguard program, which allows the government to partner with startups to test out and help develop early-stage high-performance computing technologies. “The goal is to test them for our advanced simulation and computing mission codes,” Laros says.

Penguin Solutions integrated the thermal-management and power-distribution systems for Spectra, and led the installation at Sandia National Laboratories. Craig Fritz/Sandia National Laboratories

Sandia runs a large portion of its code on CPUs, says Laros. They’ve adopted GPU based systems built by Nvidia as well. These systems offer speed advantages, but they require lab staff to port their code. “It took us hundreds of hours,” says Laros. And there are important scientific simulations that don’t run well on GPUs, including Monte Carlo methods, which can be used to assess complex risks.

Laros says it’s unusual right now to find a computing startup focusing on high-performance scientific computing—“It’s all about AI” these days, he says. Next Silicon is developing hardware that the company hopes will have advantages for both, thanks to its promised power savings. Power availability is a major constraint on large-scale AI data centers today. Raz hopes NextSilicon’s accelerators will offer an advantage by enabling more efficient performance for a given amount of electricity consumption.

Mathematics is often regarded as the ideal domain for measuring AI progress effectively. Math’s step-by-step logic is easy to track, and its definitive, automatically verifiable answers remove any human or subjective factors. But AI systems are improving at such a pace that math benchmarks are struggling to keep up.

Way back in November 2024, nonprofit research organization Epoch AI quietly released FrontierMath. A standardized, rigorous benchmark, FrontierMath was designed to measure the mathematical reasoning capabilities of the latest AI tools.

“It’s a bunch of really hard math problems,” explains Greg Burnham, Epoch AI senior researcher. “Originally, it was 300 problems that we now call tiers 1–3, but having seen AI capabilities really speed up, there was a feeling that we had to run to stay ahead, so now there’s a special challenge set of extra carefully constructed problems that we call tier 4.”

To a rough approximation, tiers 1–4 go from advanced undergraduate through to early postdoc-level mathematics. When introduced, state-of-the-art AI models were unable to solve more than 2 percent of the problems FrontierMath contained. Fast forward to today: The best publicly available AI models, such as GPT-5.2 and Claude Opus 4.6, are solving over 40 percent of FrontierMath’s 300 tier 1–3 problems, and over 30 percent of the 50 tier 4 problems.

AI takes on Ph.D.-level mathematics

And this dizzying pace of advancement is showing no signs of abating. For example, just recently Google DeepMind announced that Aletheia, an experimental AI system derived from Gemini Deep Think, achieved publishable Ph.D.-level research results. Though obscure mathematically—it was calculated with certain structure constants in arithmetic geometry called eigenweights—the result is significant in terms of AI development.

“They’re claiming it was essentially autonomous, meaning a human wasn’t guiding the work, and it’s publishable,” Burnham says. “It’s definitely at the lower end of the spectrum of work that would get a mathematician excited, but it’s new—it’s something we truly haven’t really seen before.”

To place this achievement in context, every FrontierMath problem has a known answer that a human has derived. Though a human could probably have achieved Aletheia’s result “if they sat down and steeled themselves for a week,” says Burnham, no human had ever done so.

Aletheia’s results and other recent achievements by AI mathematicians point to new, tougher benchmarks being needed to understand AI capabilities—and fast, because existing ones will soon become irrelevant. “There are easier math benchmarks that are already obsolete, several generations of them,” says Burnham. “FrontierMath will probably saturate [Ed. note: This means that state-of-the-art AI models score 100 percent] within the next two years—could be faster.”

The First Proof challenge

To begin to address this problem, on 6 February, a group of 11 highly distinguished mathematicians proposed the First Proof challenge, a set of 10 extremely difficult math questions that arose naturally in the authors’ research processes, and whose proofs are roughly five pages or less and had not been shared with anyone. The First Proof challenge was a preliminary effort to assess the capabilities of AI systems in solving research-level math questions on their own.

Generating serious buzz in the math community, professional and amateur mathematicians, and teams including OpenAI, all stepped up to the challenge. But by the time the authors posted the proofs on 14 February, no one had submitted correct solutions to all 10 problems.

In fact, far from it. The authors themselves only solved two of the 10 problems using Gemini 3.0 Deep Think and ChatGPT 5.2 Pro. And most outside submissions fared little better, apart from OpenAI and a small Aletheia team at Google DeepMind. With “limited human supervision,” OpenAI’s most advanced internal AI system solved five of the 10 problems, with Aletheia achieving similar outcomes—results met with a spectrum of emotions by different members of the mathematics community, from awe to disappointment. The team behind First Proof plans an even tougher second round on 14 March.

A new frontier for AI

“I think First Proof is terrific: It’s as close as you could realistically get to putting an AI system in the shoes of a mathematician,” says Burnham. Though he admires how First Proof tests AI’s mathematical utility for a wide range of mathematics and mathematicians, Epoch AI has its own new approach to testing—FrontierMath: Open Problems. Uniquely, the pilot benchmark consists of 16 open problems (with more to follow) from research mathematics that professional mathematicians have tried and failed to solve. Since Open Problems’ release on 27 January, none have been solved by an AI.

“With Open Problems, we’ve tried to make it more challenging,” says Burnham. “The baseline on its own would be publishable, at least in a specialty journal.” What’s more, each question is designed so that it can be automatically graded. “This is a bit counterintuitive,” Burnham adds. “No one knows the answers, but we have a computer program that will be able to judge whether the answer is right or not.”

Burnham sees First Proof and Open Problems as being complementary. “I would say understanding AI capabilities is a more-the-merrier situation,” he adds. “AI has gotten to the point where it’s—in some ways—better than most Ph.D. students, so we need to pose problems where the answer would be at least moderately interesting to some human mathematicians, not because AI was doing it but because it’s mathematics that human mathematicians care about.”

AI has driven an explosion of new number formats—the ways in which numbers are represented digitally. Engineers are looking at every possible way to save computation time and energy, including shortening the number of bits used to represent data. But what works for AI doesn’t necessarily work for scientific computing, be it for computational physics, biology, fluid dynamics, or engineering simulations. IEEE Spectrum spoke with Laslo Hunhold, who recently joined Barcelona-based Openchip as an AI engineer, about his efforts to develop a bespoke number format for scientific computing.

LASLO HUNHOLD

Laslo Hunhold is a senior AI accelerator engineer at Barcelona-based startup Openchip. He recently completed a Ph.D. in computer science from the University of Cologne, in Germany.

What makes number formats interesting to you?

Laslo Hunhold: I don’t know another example of a field that so few are interested in but has such a high impact. If you make a number format that’s 10 percent more [energy] efficient, it can translate to all applications being 10 percent more efficient, and you can save a lot of energy.

Why are there so many new number formats?

Hunhold: For decades, computer users had it really easy. They could just buy new systems every few years, and they would have performance benefits for free. But this hasn’t been the case for the last 10 years. In computers, you have a certain number of bits used to represent a single number, and for years the default was 64 bits. And for AI, companies noticed that they don’t need 64 bits for each number. So they had a strong incentive to go down to 16, 8, or even 2 bits [to save energy]. The problem is, the dominating standard for representing numbers in 64 bits is not well designed for lower bit counts. So in the AI field, they came up with new formats which are more tailored toward AI.

Why does AI need different number formats than scientific computing?

Hunhold: Scientific computing needs high dynamic range: You need very large numbers, or very small numbers, and very high accuracy in both cases. The 64-bit standard has an excessive dynamic range, and it is many more bits than you need most of the time. It’s different with AI. The numbers usually follow a specific distribution, and you don’t need as much accuracy.

What makes a number format “good”?

Hunhold: You have infinite numbers but only finite bit representations. So you need to decide how you assign numbers. The most important part is to represent numbers that you’re actually going to use. Because if you represent a number that you don’t use, you’ve wasted a representation. The simplest thing to look at is the dynamic range. The next is distribution: How do you assign your bits to certain values? Do you have a uniform distribution, or something else? There are infinite possibilities.

What motivated you to introduce the takum number format?

Hunhold: Takums are based on posits. In posits, the numbers that get used more frequently can be represented with more density. But posits don’t work for scientific computing, and this is a huge issue. They have a high density for [numbers close to one], which is great for AI, but the density falls off sharply once you look at larger or smaller values. People have been proposing dozens of number formats in the last few years, but takums are the only number format that’s actually tailored for scientific computing. I found the dynamic range of values you use in scientific computations, if you look at all the fields, and designed takums such that when you take away bits, you don’t reduce that dynamic range

This article appears in the March 2026 print issue as “Laslo Hunhold.”

Huge amounts of data could be archived in a compact format for millennia, by using lasers to encode that data into glass.

In 2028, the amount of data that the world is projected to generate—in the form of photos, videos, emails, documents, and other files—could reach about 394 zettabytes (394 trillion gigabytes), according to the market analyst firm International Data Corporation. That’s a nearly threefold increase from 2023.This demand has driven research toward radically new ways of storing data, in part because all mainstream electronic data storage techniques break down over time. For example, most digital archives are currently saved on magnetic tape, which only lasts about 30 years.

One possibility that scientists have explored is storing data in glass because of its resistance to moisture, temperature fluctuations, and electromagnetic interference. However, previous attempts exploring this strategy faced problems with data density, writing throughput, and energy efficiency.

In a study, Microsoft scientists sought to overcome those deficiencies by using femtosecond lasers, which fire high-power laser pulses just quadrillionths of a second long. The researchers focused a laser beam on a point within glass to modify its optical properties and encode data as voxels, the 3D equivalent of pixels. The researchers published their findings online on 18 February in Nature.

Advanced Glass Data-Storage Technology

Each voxel the scientists encoded was about 0.5 micrometers large in each direction and separated from each other by about 6 µm. They could then read out the data using an automated microscope. Machine learning models used during reading help account for the effects of random errors and optical interference between voxels.

The new archival system, called Silica, can store up to 1.59 gigabits per cubic millimeter. This amounts to 4.84 terabytes stored in roughly 300 layers within a 12-square-centimeter, 2-millimeter-deep glass chip, the equivalent of about 2 million printed books or 5,000 ultrahigh-definition 4K films.

Femtosecond laser pulses can write data into glass in the form of tiny 3D voxels.Microsoft Research

One drawback of previous attempts at glass data storage was how they often relied on multiple laser pulses to write each voxel. This led to reduced throughput and high power demands. To overcome this challenge, the scientists developed two different types of voxels, called phase voxels and birefringent voxels, which each required only a single laser pulse to write.

“This significantly reduces the power required from the laser to store data, and it does not require the laser focus to alternate between staying in the same place to deliver multiple pulses and movement to the next location,” says Richard Black, the research director for Microsoft’s Project Silica. “Consequently, the laser focus can be swept rapidly across the glass, enabling writing speed only limited by the speed of the femtosecond laser itself.” All in all, by firing millions of laser pulses per second, Silica could write up to 25.6 megabits per second with a single beam.

Both of these kinds of voxels modify the speed at which light travels through the glass, which is based off a factor called the refractive index. A phase voxel alters the refractive index uniformly in all directions, whereas a birefringent voxel changes the refractive index in a direction-dependent manner.

With birefringent voxels, the scientists could achieve significantly better data density, write throughput, and energy efficiency. However, these voxels required high-purity silica glasses. In comparison, phase voxels can potentially be written in any durable transparent material, such as borosilicate glass, which is widely available in cookware. Phase voxels also require simpler writing and reading hardware.

To assess the long-term stability of Silica, the scientists repeatedly heated inscribed glass plates up to 500 °C, simulating the long-term aging of the glass at lower temperatures. Their results suggest the data on the plates might be readable for more than 10,000 years at 290 °C, and likely even longer at ambient temperature, an endurance far exceeding conventional electronic data-storage techniques.

An automated microscope can read out the voxels carved into glass to recover the stored data.Microsoft research

“This is not a replacement for everyday storage like [solid-state drives] or hard drives,” Black says. “It’s designed for data you want to write once and preserve for a very long time.”

The clearest applications “are archival—anywhere data must survive for centuries such as national libraries, scientific data, or cultural records,” Black says. “It’s also compelling for cloud archives, where data is written once and kept indefinitely.”

A common misconception about glass is that it flows over a few centuries. Black notes that “at room temperature, glass is effectively a solid and does not flow on any meaningful time scale.” At normal storage conditions, “the glass and the data structures within it are essentially unchanged for far longer than human history.”

Black cautions that questions remain about how feasible this strategy can prove at scale, as femtosecond lasers are currently expensive. The researchers hope that by sharing their results, other groups can build on their work to help make the technology accessible.

This article appears in the April 2026 print issue as “Microsoft Boosts Density of Forever Storage in Glass.”

Join Hannah Alpert (NASA Ames) to explore thermal data from the record-breaking 6-meter LOFTID inflatable aeroshell. Learn how COMSOL Multiphysics® was used to perform inverse analysis on flight thermocouple data, validating heat flux gauges and preflight CFD predictions. Attendees will gain technical insights into improving thermal models for future HIAD missions, making this essential for engineers seeking to advance atmospheric reentry design. The session concludes with a live Q&A.

Register now to watch this free on-demand webinar!

In 2024, Google claimed that its data centers are 1.5 times more energy-efficient than the industry average. In 2025, Microsoft committed billions to nuclear power for AI workloads. The data center industry tracks power-usage effectiveness to three decimal places and optimizes water usage intensity with machine precision. We report direct emissions and energy emissions with religious fervor.

These are laudable advances, but these metrics account for only 30 percent of total emissions from the IT sector. The majority of the emissions are not directly from data centers or the energy they use, but from the end-user devices that actually access the data centers, emissions due to manufacturing the hardware, and software inefficiencies. We are frantically optimizing less than a third of the IT sector’s environmental impact, while the bulk of the problem goes unmeasured.

Incomplete regulatory frameworks are part of the problem. In Europe, the Corporate Sustainability Reporting Directive (CSRD) now requires 11,700 companies to report emissions using these incomplete frameworks. The next phase of the directive, covering 40,000+ additional companies, was originally scheduled for 2026 (but is likely delayed to 2028). In the United States, the standards body responsible for IT sustainability metrics (ISO/IEC JTC 1/SC 39) is conducting active revision of its standards through 2026, with a key plenary meeting in May 2026.

The time to act is now. If we don’t fix the measurement frameworks, we risk locking in incomplete data collection and optimizing a fraction of what matters for the next 5 to 10 years, before the next major standards revision.

The limited metrics

Walk into any modern data center, and you’ll see sustainability instrumentation everywhere. Power-usage efficiency (PUE) monitors track every watt. Water-usage efficiency (WUE) systems measure water consumption down to the gallon. Sophisticated monitoring captures everything from server utilization to cooling efficiency to renewable energy percentages.

But here’s what those measurements miss: End-user devices globally emit 1.5 to 2 times more carbon than all data centers combined, according to McKinsey’s 2022 report. The smartphones, laptops, and tablets we use to access those ultra-efficient data centers are the bigger problem.

Data center operations, as measured by power-usage efficiency, account for only 24 percent of the total emissions.

On the conservative end of the range from McKinsey’s report, devices emit 1.5 times as much as data centers. That means that data centers make up 40 percent of total IT emissions, while devices make up 60 percent.

On top of that, approximately 75 percent of device emissions occur not during use, but during manufacturing—this is so-called embodied carbon. For data centers, only 40 percent is embodied carbon, and 60 percent comes from operations (as measured by PUE).

Putting this together, data center operations, as measured by PUE, account for only 24 percent of the total emissions. Data center embodied carbon is 16 percent, device embodied carbon is 45 percent, and device operation is 15 percent.

Under the EU’s current CSRD framework, companies must report their emissions in three categories: direct emissions from owned sources, indirect emissions from purchased energy, and a third category for everything else.

This “everything else” category does include device emissions and embodied carbon. However, those emissions are reported as aggregate totals broken down by accounting category—capital goods, purchased goods and services, use of sold products—but not by product type. How much comes from end-user devices versus data center infrastructure, or employee laptops versus network equipment, remains murky, and therefore, unoptimized.

Embodied carbon and hardware reuse

Manufacturing a single smartphone generates approximately 50 kilograms CO₂ equivalent (CO₂e). For a laptop, it’s 200 kg CO₂e. With 1 billion smartphones replaced annually, that’s 50 million tonnes of CO₂e per year just from smartphone manufacturing, before anyone even turns them on. On average, smartphones are replaced every two years, laptops every three to four years, and printers every five years. Data center servers are replaced approximately every five years.

Extending smartphone life cycles to three years instead of two would reduce annual manufacturing emissions by 33 percent. At scale, this dwarfs data center optimization gains.

There are programs geared toward reusing old components that are still functional and integrating them into new servers. GreenSKUs and similar initiatives show that 8 percent reductions in embodied carbon are achievable. But these remain pilot programs, not systematic approaches. And critically, they’re measured only in the data center context, not across the entire IT stack.

Imagine applying the same circular economy principles to devices. With over 2 billion laptops in existence globally and two- to three-year replacement cycles, even modest lifespan extensions create massive emission reductions. Extending smartphone life cycles to three years instead of two would reduce annual manufacturing emissions by 33 percent. At scale, this dwarfs data center optimization gains.

Yet data center reuse gets measured, reported, and optimized. Device reuse doesn’t, because the frameworks don’t require it.

The invisible role of software

Leading load balancer infrastructure across IBM Cloud, I see how software architecture decisions ripple through energy consumption. Inefficient code doesn’t just slow things down—it drives up both data center power consumption and device battery drain.

For example, University of Waterloo researchers showed that they can reduce 30 percent of energy use in data centers by changing just 30 lines of code. From my perspective, this result is not an anomaly—it’s typical. Bad software architecture forces unnecessary data transfers, redundant computations, and excessive resource use. But unlike data center efficiency, there’s no commonly accepted metric for software efficiency.

This matters now more than ever. With AI workloads driving massive data center expansion—projected to consume 6.7 to 12 percent of total U.S. electricity by 2028, according to Lawrence Berkeley National Laboratory—software efficiency becomes critical.

What needs to change

The solution isn’t to stop measuring data center efficiency. It’s to measure device sustainability with the same rigor. Specifically, standards bodies (particularly ISO/IEC JTC 1/SC 39 WG4: Holistic Sustainability Metrics) should extend frameworks to include device life-cycle tracking, software efficiency metrics, and hardware reuse standards.

To track device life cycles, we need standardized reporting of device embodied carbon, broken out separately by device. One aggregate number in an “everything else” category is insufficient. We need specific device categories with manufacturing emissions and replacement cycles visible.

To include software efficiency, I advocate developing a PUE-equivalent for software, such as energy per transaction, per API call, or per user session. This needs to be a reportable metric under sustainability frameworks so companies can demonstrate software optimization gains.

To encourage hardware reuse, we need to systematize reuse metrics across the full IT stack—servers and devices. This includes tracking repair rates, developing large-scale refurbishment programs, and tracking component reuse with the same detail currently applied to data center hardware.

To put it all together, we need a unified IT emission-tracking dashboard. CSRD reporting should show device embodied carbon alongside data center operational emissions, making the full IT sustainability picture visible at a glance.

These aren’t radical changes—they’re extensions of measurement principles already proven in the data center context. The first step is acknowledging what we’re not measuring. The second is building the frameworks to measure it. And the third is demanding that companies report the complete picture—data centers and devices, servers and smartphones, infrastructure and software.

Because you can’t fix what you can’t see. And right now, we’re not seeing 70 percent of the problem.

When Alan DeKok began a side project in network security, he didn’t expect to start a 27-year career. In fact, he didn’t initially set out to work in computing at all.

DeKok studied nuclear physics before making the switch to a part of network computing that is foundational but—like nuclear physics—largely invisible to those not directly involved in the field. Eventually, a project he started as a hobby became a full-time job: maintaining one of the primary systems that helps keep the internet secure.

Alan DeKok

Employer

InkBridge Networks

Occupation

CEO

Education

Bachelor’s degree in physics, Carleton University; master’s degree in physics, Carleton University

Today, he leads the FreeRADIUS Project, which he cofounded in the late 1990s to develop what is now the most widely used Remote Authentication Dial-In User Service (RADIUS) software. FreeRADIUS is an open-source server that provides back-end authentication for most major internet service providers. It’s used by global financial institutions, Wi-Fi services like Eduroam, and Fortune 50 companies. DeKok is also CEO of InkBridge Networks, which maintains the server and provides support for the companies that use it.

Reflecting on nearly three decades of experience leading FreeRADIUS, DeKok says he became an expert in remote authentication “almost by accident,” and the key to his career has largely been luck. “I really believe that it’s preparing yourself for luck, being open to it, and having the skills to capitalize on it.”

From Farming to Physics

DeKok grew up on a farm outside of Ottawa growing strawberries and raspberries. “Sitting on a tractor in the heat is not particularly interesting,” says DeKok, who was more interested in working with 8-bit computers than crops. As a student at Carleton University, in Ottawa, he found his way to physics because he was interested in math but preferred the practicality of science.

While pursuing a master’s degree in physics, also at Carleton, he worked on a water-purification system for the Sudbury Neutrino Observatory, an underground observatory then being built at the bottom of a nickel mine. He would wake up at 4:30 in the morning to drive up to the site, descend 2 kilometers, then enter one of the world’s deepest clean-room facilities to work on the project. The system managed to achieve one atom of impurity per cubic meter of water, “which is pretty insane,” DeKok says.

But after his master’s degree, DeKok decided to take a different route. Although he found nuclear physics interesting, he says he didn’t see it as his life’s work. Meanwhile, the Ph.D. students he knew were “fanatical about physics.” He had kept up his computing skills through his education, which involved plenty of programming, and decided to look for jobs at computing companies. “I was out of physics. That was it.”

Still, physics taught him valuable lessons. For one, “You have to understand the big picture,” DeKok says. “The ability to tell the big-picture story in standards, for example, is extremely important.” This skill helps DeKok explain to standards bodies how a protocol acts as one link in the entire chain of events that needs to occur when a user wants to access the internet.

He also learned that “methods are more important than knowledge.” It’s easy to look up information, but physics taught DeKok how to break down a problem into manageable pieces to come up with a solution. “When I was eventually working in the industry, the techniques that came naturally to me, coming out of physics, didn’t seem to be taught as well to the people I knew in engineering,” he says. “I could catch up very quickly.”

Founding FreeRADIUS

In 1996, DeKok was hired as a software developer at a company called Gandalf, which made equipment for ISDN, a precursor to broadband that enabled digital transmission of data over telephone lines. Gandalf went under about a year later, and he joined CryptoCard, a company providing hardware devices for two-factor authentication.

While at CryptoCard, DeKok began spending more time working with a RADIUS server. When users want to connect to a network, RADIUS acts as a gatekeeper and verifies their identity and password, determines what they can access, and tracks sessions. DeKok moved on to a new company in 1999, but he didn’t want to lose the networking skills he had developed. No other open-source RADIUS servers were being actively developed at the time, and he saw a gap in the market.

The same year, he started FreeRADIUS in his free time and it “gradually took over my life,” DeKok says. He continued to work on the open-source software as a hobby for several years while bouncing around companies in California and France. “Almost by accident, I became one of the more senior people in the space. Then I doubled down on that and started the business.” He founded NetworkRADIUS (now called InkBridge Networks) in 2008.

By that point, FreeRADIUS was already being used by 100 million people daily. The company now employs experts in Canada, France, and the United Kingdom who work together to support FreeRADIUS. “I’d say at least half of the people in the world get on the internet by being authenticated through my software,” DeKok estimates. He attributes that growth largely to the software being open source. Initially a way to enter the market with little funding, going open source has allowed FreeRADIUS to compete with bigger companies as an industry-leading product.

Although the software is critical for maintaining secure networks, most people aren’t aware of it because it works behind the scenes. DeKok is often met with surprise that it’s still in use. He compares RADIUS to a building foundation: “You need it, but you never think about it until there’s a crack in it.”

27 Years of Fixes

Over the years, DeKok has maintained FreeRADIUS by continually making small fixes. Like using a ratcheting tool to make a change inch by inch, “you shouldn’t underestimate that ratchet effect of tiny little fixes that add up over time,” he says.

He’s seen the project through minor patches and more significant fixes, like when researchers exposed a widespread vulnerability DeKok had been trying to fix since 1998. He also watched a would-be successor to the network protocol, Diameter, rise and fall in popularity in the 2000s and 2010s. (Diameter gained traction in mobile applications but has gradually been phased out in the shift to 5G.) Though Diameter offers improvements, RADIUS is far simpler and already widely implemented, giving it an edge, DeKok explains.

And he remains confident about its future. “People ask me, ‘What’s next for RADIUS?’ I don’t see it dying.” Estimating that billions of dollars of equipment run RADIUS, he says, “It’s never going to go away.”

About his own career, DeKok says he plans to keep working on FreeRADIUS, exploring new markets and products. “I never expected to have a company and a lot of people working for me, my name on all kinds of standards, and customers all over the world. But it worked out that way.”

This article appears in the March 2026 print issue as “Alan DeKok.”

This past December, the U.K. startup Space Forge turned on an orbital furnace aboard its ForgeStar-1 satellite, producing a stream of superhot plasma that could someday enable production of near-ideal semiconductor crystals in orbit. Hailed as a breakthrough in orbital manufacturing, the milestone is a first for a free-flying commercial satellite and a payload not operated by humans.

Founded in 2018, Space Forge is one of several companies founded on the premise that made-in-space materials can help bring about ultra-efficient next-generation electronics, ultrafast optical networks, and breakthroughs in pharmaceutical research. Space Forge’s flying furnace is specifically designed to make seed crystals that would be used later, on Earth, to produce substrates of gallium and aluminum nitride or silicon carbide for high-performance power devices.

Semiconductors have been made in space before. Astronauts grew crystals of indium antimonide and germanium in the 1970s aboard the Skylab space station, and similar experiments have been taking place on the International Space Station. Between 1973 and 2016, around 160 semiconductor crystals were grown in microgravity aboard various spacecraft, according to a meta-analysis study published in the journal Nature in 2024. The study found that 86 percent of those space-grown crystals grew larger, were more uniform, and showed better performance than their counterparts grown on Earth.

“There is potential for significant energy savings, perhaps as much as 50 percent within large infrastructure installations such as 5G towers,” says Joshua Western, Space Forge’s cofounder and CEO, in an interview.

E. Steve Putna, the director of the Texas A&M Semiconductor Institute, says that “space-grown crystals have demonstrated significantly higher electron mobility,” which could translate to a 20 to 40 percent increase in switching efficiency compared with Earth-grown counterparts. (Putna has no involvement with Space Forge.)

Why Space Fabrication Makes Better Crystals

On Earth, most semiconductors today are fabricated using highly pure materials such as silicon, gallium arsenide, or gallium nitride, produced by depositing vaporized precursor chemicals layer by layer on substrates inside a vessel called a reactor. The fabrication processes are very tightly controlled, but the resulting semiconductor crystals nevertheless have small amounts of impurities and defects, which can cause heat during operation and slightly reduce performance.

Western says that the superior vacuum in space does away with those impurities, thus increasing the quality of the crystals produced. ”For example, if you’re worried about nitrogen interfering with your growth process, on Earth [in a vacuum chamber] nitrogen might be present at concentration of around 10 to the –11” (10 to the power of minus 11, or 1 x 10^-9 percent). “In space, above 500 kilometers altitude, it’s naturally present at 10 to the –22.”

ForgeStar-1 was photographed during its successful test.Space Forge

Moreover, the microgravity environment gives the crystal growth process a “better head start,” he says, creating uniform conditions in which the crystals form. “On Earth, you have trouble that, perhaps, some crystals grow around the interior of the reactor and not in other parts because the process between hot and cold is influenced by gravity,” Western says. “Microgravity effectively prevents convection from taking place, so you get a continually uniform deposition area.”

Such advantages are responsible for the enhanced performance and efficiency of the resulting electronic components made with space-grown crystals, says Western. “The thermal performance of a semiconductor is directly driven by how good its lattice structure is and how good its purity is,” he says. “If you can change those things, then you are able to produce semiconductors that require much less cooling because they run at a lower temperature, which allows you to significantly reduce energy consumption. Or you can trade off that significant reduction in energy consumption for significantly higher output.”

Putna says the decreased internal resistance and reduced heat generation that results from fewer defects in the semiconductor structure could be “a game changer for AI data centers where cooling costs are a primary bottleneck” for wider deployment. In power electronics, more perfect crystals can allow a smaller chip to handle higher voltages without failing, he adds.

Will the Performance Boost Justify the High Costs?

Still, launching stuff into space and returning it back to Earth is expensive. Currently, SpaceX’s Falcon 9 launches payloads to low Earth orbit for an estimated US $1,500 per kilogram. Materials made on the ISS can be returned to Earth on SpaceX’s Cargo Dragon capsule, but demand for its services is high and availability limited.

Western says the seed crystals Space Forge will create in space will be further sprouted in terrestrial foundries while passing on their out-of-this-world qualities. From a single kilogram of space-grown semiconductor, manufacturers on Earth will grow tonnes of high-performance material, he says. “There will be a level of degradation over time and over generations of growth,” he concedes, “but it will be multiple growth runs before the quality degrades to the point of the current state of the art.”

Putna insists that “if a space-grown substrate increases the yield of a $10,000 high-end AI processor from 50 percent to 90 percent or allows a quantum computer to function closer to room temperature rather than near absolute zero, the launch cost becomes a negligible fraction of the total value created.”

Not everyone is convinced. Matt Francis, CEO of the electronics company Ozark Integrated Circuits, says that the price of silicon substrates plummeted over the past few years, making infrastructure operators less likely to buy costly space-grown crystals. And terrestrial fabrication technologies, he points out, keep improving. Ozark specializes in rugged ICs for aerospace and other applications.

“While I remember paying $20K a wafer in the early days, we are down in the hundreds-of-dollars range in volume markets like power,” Francis says. “When they were a prized commodity, maybe sending to space made sense. While the cost of space is decreasing, it’s not decreasing faster than the cost of producing wafers.” In some cases, Francis notes, it might turn out to be possible to get performance superior to a space-grown device by using multiple conventional devices, and at lower overall cost.

Space Forge has yet to test its return technology. Although the ForgeStar-1 satellite will deploy a novel heat shield during its de-orbit maneuver later this year, this spacecraft is designed only to put the orbital furnace through its paces and prove it can “repeatedly create and maintain the manufacturing environment required for the chemistry process” needed to grow superconductor crystals, according to Western.

The satellite will ultimately perish during its return to Earth, meaning Space Forge will only get its first batch of space-grown crystals back home with the follow-up mission, expected to launch at some point next year. The amount of material brought back will be at best a few kilograms.

Other companies are betting on the potential of space-grown crystals. Colorado-based Voyager Technologies recently patented a new method for growing crystals of novel fiber-optic materials in orbit, which the company claims could significantly speed up data transmission. London-headquartered ACME Space wants to test its balloon-launched orbital factory Hyperion later this year, also eyeing the semiconductor, pharmaceuticals, and optical-fiber markets. California-based Varda Industries raised an impressive $329 million last year to begin manufacturing pharmaceuticals in space. Since 2023, the company has conducted multiple orbital flights to test its return technology and is planning additional tests this year. Some analysts estimate the in-orbit manufacturing market could reach $28.19 billion by 2034.

Amid this enthusiasm, Anne Wilson, a professor of biochemistry at Butler University, in Indianapolis, and an author of the 2024 Nature metanalysis, sounds a note of caution. “I don’t think that microgravity is going to be ideal for the manufacture of bulk materials,” she says. “However, niche materials for specific applications might be worth the investment.”

If it feels these days as if everything in technology is about AI, that’s because it is. And nowhere is that more true than in the market for computer memory. Demand, and profitability, for the type of DRAM used to feed GPUs and other accelerators in AI data centers is so huge that it’s diverting the supply of memory away from other uses and causing prices to skyrocket. According to Counterpoint Research, DRAM prices have risen 80 to 90 precent so far this quarter.

The largest AI hardware companies are likely first in line for new memory chips, but that leaves everybody else—makers of PCs, consumer gizmos, and everything else that needs to temporarily store a billion bits—scrambling to deal with scarce supply and inflated prices.

How did the electronics industry get into this mess, and even more important, how will it get out? IEEE Spectrum asked economists and memory experts to explain. They say today’s situation is the result of a collision between the DRAM industry’s historic boom-and-bust cycle and an AI hardware infrastructure build-out that’s without precedent in its scale. And, barring some major collapse in the AI sector, it will take years for new capacity and new technology to bring supply in line with demand. Prices might stay high even then.

To understand both ends of the tale, you need to know the main culprit in the supply-and-demand swing: high-bandwidth memory, or HBM.

What is HBM?

HBM is the DRAM industry’s attempt to short-circuit the slowing pace of Moore’s Law by using 3D chip-packaging technology. Each HBM chip is made up of as many as 12 thinned-down DRAM chips called dies. Each die contains a number of vertical connections called through-silicon vias. The dies are piled atop each other and connected by arrays of microscopic solder balls aligned to the TSVs. This DRAM tower—well, at about 750 micrometers thick, it’s more of a brutalist office block than a tower—is then stacked atop what’s called the base die, which shuttles bits between the memory dies and the processor.

This complex piece of technology is then set within a millimeter of a GPU or another AI accelerator, to which it is linked by as many as 2,048 µm-scale connections. HBMs are attached on two sides of the processor, and the GPU and memory are packaged together as a single unit.

The idea behind such a tight, highly connected squeeze with the GPU is to knock down what’s called the memory wall. That’s the barrier in energy and time of bringing the terabytes per second of data needed to run large language models into the GPU. Memory bandwidth is a key limiter to how fast LLMs can run.

As a technology, HBM has been around for more than 10 years, and DRAM makers have been busy boosting its capability.

As the size of AI models has grown, so has HBM’s importance to the GPU. But that’s come at a cost. SemiAnalysis estimates that HBM generally costs three times as much as other types of memory and constitutes 50 percent or more of the cost of the packaged GPU.

Origins of the memory chip shortage

Memory- and storage-industry watchers agree that DRAM is a highly cyclical industry with huge booms and devastating busts. With new fabs costing US $15 billion or more, firms are extremely reluctant to expand and may only have the cash to do so during boom times, explains Thomas Coughlin, a storage and memory expert and president of Coughlin Associates. But building such a fab and getting it up and running can take 18 months or more, practically ensuring that new capacity arrives well past the initial surge in demand, flooding the market and depressing prices.

The origins of today’s cycle, says Coughlin, go all the way back to the chip supply panic surrounding the COVID-19 pandemic. To avoid supply-chain stumbles and support the rapid shift to remote work, hyperscalers—data-center giants like Amazon, Google, and Microsoft—bought up huge inventories of memory and storage, thus boosting prices, he notes.

But then supply became more regular and data-center expansion fell off in 2022, causing memory and storage prices to plummet. This recession continued into 2023, and even resulted in big memory and storage companies such as Samsung cutting production by 50 percent to try to keep prices from going below the costs of manufacturing, says Coughlin. It was a rare and fairly desperate move, because companies typically have to run plants at full capacity just to earn back their value.

After a recovery began in late 2023, “all the memory and storage companies were very wary of increasing their production capacity again,” says Coughlin. “Thus there was little or no investment in new production capacity in 2024 and through most of 2025.”

The AI data-center boom

That lack of new investment is colliding headlong with a huge boost in demand from new data centers. Globally, there are nearly 2,000 new data centers either planned or under construction right now, according to Data Center Map. If they’re all built, it would represent a 20 percent jump in the global supply, which stands at around 9,000 facilities now.

If the current build-out continues at pace, McKinsey predicts companies will spend $7 trillion by 2030, with the bulk of that—$5.2 trillion—going to AI-focused data centers. Of that chunk, $3.3 trillion will go toward servers, data storage, and network equipment, the firm predicts.

The biggest beneficiary so far of the AI data-center boom is unquestionably the GPU maker Nvidia. Revenue for its data-center business went from barely a billion in the final quarter of 2019 to $51 billion in the quarter that ended in October 2025. Over this period, its server GPUs have demanded not just more and more gigabytes of DRAM but an increasing number of DRAM chips. The recently released B300 uses eight HBM chips, each of which is a stack of 12 DRAM dies. Competitors’ use of HBM has largely mirrored Nvidia’s. AMD’s MI350 GPU, for example, also uses eight 12-die chips.

With so much demand, an increasing fraction of the revenue for DRAM makers comes from HBM. Micron—the No. 3 producer behind SK Hynix and Samsung—reported that HBM and other cloud-related memory went from being 17 percent of its DRAM revenue in 2023 to nearly 50 percent in 2025.

Micron predicts the total market for HBM will grow from $35 billion in 2025 to $100 billion by 2028—a figure larger than the entire DRAM market in 2024, CEO Sanjay Mehrotra told analysts in December. It’s reaching that figure two years earlier than Micron had previously expected. Across the industry, demand will outstrip supply “substantially…for the foreseeable future,” he said.

Future DRAM supply and technology

“There are two ways to address supply issues with DRAM: with innovation or with building more fabs,” explains Mina Kim, an economist with Mkecon Insights. “As DRAM scaling has become more difficult, the industry has turned to advanced packaging… which is just using more DRAM.”

Micron, Samsung, and SK Hynix combined make up the vast majority of the memory and storage markets, and all three have new fabs and facilities in the works. However, these are unlikely to contribute meaningfully to bringing down prices.

Micron is in the process of building an HBM fab in Singapore that should be in production in 2027. And it is retooling a fab it purchased from PSMC in Taiwan that will begin production in the second half of 2027. Last month, Micron broke ground on what will be a DRAM fab complex in Onondaga County, N.Y. It will not be in full production until 2030.

Samsung plans to start producing at a new plant in Pyeongtaek, South Korea, in 2028.

SK Hynix is building HBM and packaging facilities in West Lafayette, Ind., set to begin production by the end of 2028, and an HBM fab it’s building in Cheongju, South Korea, should be complete in 2027.

Speaking of his sense of the DRAM market, Intel CEO Lip-Bu Tan told attendees at the Cisco AI Summit last week, “There’s no relief until 2028.”

With these expansions unable to contribute for several years, other factors will be needed to increase supply. “Relief will come from a combination of incremental capacity expansions by existing DRAM leaders, yield improvements in advanced packaging, and a broader diversification of supply chains,” says Shawn DuBravac, chief economist for the Global Electronics Association (formerly the IPC). “New fabs will help at the margin, but the faster gains will come from process learning, better [DRAM] stacking efficiency, and tighter coordination between memory suppliers and AI chip designers.”

So, will prices come down once some of these new plants come on line? Don’t bet on it. “In general, economists find that prices come down much more slowly and reluctantly than they go up. DRAM today is unlikely to be an exception to this general observation, especially given the insatiable demand for compute,” says Kim.

In the meantime, technologies are in the works that could make HBM an even bigger consumer of silicon. The standard for HBM4 can accommodate 16 stacked DRAM dies, even though today’s chips only use 12 dies. Getting to 16 has a lot to do with the chip-stacking technology. Conducting heat through the HBM “layer cake” of silicon, solder, and support material is a key limiter to going higher and in repositioning HBM inside the package to get even more bandwidth.

SK Hynix claims a heat-conduction advantage through a manufacturing process called advanced MR-MUF (mass reflow molded underfill). Further out, an alternative chip-stacking technology called hybrid bonding could help heat conduction by reducing the die-to-die vertical distance essentially to zero. In 2024, researchers at Samsung proved they could produce a 16-high stack with hybrid bonding, and they suggested that 20 dies was not out of reach.

This post was corrected on 9 March. Previously it misrepresented the position of large AI hardware makers, and it misstated the scale of datacenter hardware market.

This article appears in the April 2026 print issue as the feature “AI is a Memory Hog.”

The sidebar from that article is here “Low-Cost Computers Nearly Double in Price as RAM Shortage Hits.“