The company wants to add at least 1 GW of computing capacity in Ulanqab, approximately 350 kilometres northwest of Beijing, according to a Bloomberg report republished by Investing.com.

DeepSeek reportedly plans to build part of the facility itself while leasing additional capacity from existing data center operators. At least part of the new capacity could become operational by the end of 2027 or early 2028.

The processors that will power the project have not been confirmed. DeepSeek could use Nvidia hardware, Huawei processors, its own future chips or a combination of different systems.

The scale of the plan represents a major shift for a company that became famous for reducing the computing costs associated with frontier AI. DeepSeek is now preparing the infrastructure needed to train larger models, serve more users and reduce its dependence on external computing providers.

The move reflects a broader industry trend. As Memeburn previously reported, Big Tech’s AI infrastructure spending is accelerating even as investors question the financial returns.

How Big Is DeepSeek’s Data Center

One gigawatt refers to the maximum electrical capacity available to power computing equipment, cooling systems, networking hardware and supporting infrastructure.

A fully developed 1 GW AI campus would belong to a new generation of facilities designed around increasingly dense clusters of advanced processors. The final number of chips will depend on the hardware used, the cooling architecture, power distribution and the proportion of electricity reserved for non-computing operations.

The project should therefore not be interpreted as 1 GW flowing directly into AI chips. Cooling, networking, power conversion and backup systems will consume part of the available electricity.

DeepSeek’s reported strategy also uses a hybrid development model. Building its own facility would give the company greater control over hardware, networking, security and model development. Leasing capacity from existing operators could allow it to access computing resources before the entire campus is complete.

More than a dozen companies have reportedly received approvals for projects in Ulanqab. DeepSeek would therefore enter an established computing cluster rather than develop an isolated data center in a new market.

This matters because the availability of chips is no longer the only constraint on AI expansion. Memeburn’s analysis of Oracle’s data center expansion problems showed how grid connections, transmission lines, permits and construction schedules can delay projects even when companies have enough financing.

Why DeepSeek Chose Ulanqab

Ulanqab has spent years developing into one of China’s most important computing and data center locations.

The city’s cold climate allows operators to use outside air for cooling during much of the year. This can reduce the amount of electricity consumed by conventional cooling equipment.

China’s Ministry of Industry and Information Technology previously highlighted Huawei’s Ulanqab data center, which can use natural cooling for approximately ten months of the year. The facility reported annual power usage effectiveness of 1.32, with monthly efficiency reaching approximately 1.22 under full natural cooling conditions.

The region also has substantial wind and solar resources. In July 2025, China connected a green-powered computing base in Ulanqab to the grid. The project combines 200 MW of wind power, 100 MW of solar capacity and 45 MW of energy storage.

The facility was designed to generate approximately 848 million kilowatt-hours of green electricity annually and serve a computing base containing 25,000 server cabinets. These projects form part of China’s broader East Data, West Computing policy. The strategy moves computing workloads from densely populated eastern regions toward western and northern areas where land, electricity and renewable energy are more abundant.

Ulanqab remains close enough to Beijing to serve companies operating in the Beijing, Tianjin and Hebei economic region. Apple, Alibaba and Huawei have already developed data center infrastructure in the area.

DeepSeek can therefore access lower cooling costs, large energy resources and established network infrastructure without locating its entire computing operation in a more expensive coastal technology hub.

Power availability may become an increasingly important competitive advantage. Memeburn has also examined how AI data center growth is creating disputes over electricity prices in markets where grid capacity is more limited.

The Nvidia Blackwell Question

The type of processors used inside the new data center could become the most politically sensitive part of the project.

A senior US official told Reuters in February 2026 that DeepSeek had trained an advanced model using Nvidia Blackwell processors despite US export restrictions.

The official claimed that the Blackwell chips were likely clustered at a data center in Inner Mongolia. DeepSeek has not publicly confirmed how the chips were obtained or whether the allegation is accurate. The claim should therefore be treated as a statement from a US government official rather than an independently verified description of DeepSeek’s infrastructure.

The case intensified concerns that advanced Nvidia chips could reach Chinese AI companies through intermediaries, third countries or data center arrangements outside established export channels.

Memeburn previously covered the political pressure surrounding Nvidia’s China business and US chip export controls. Lawmakers are increasingly asking chipmakers to prove that restricted hardware is not being diverted through distributors or overseas subsidiaries.

The final processor mix for DeepSeek’s Ulanqab facility remains unclear. Nvidia chips continue to set the standard for many frontier AI workloads, while Huawei has become China’s leading domestic alternative. Other Chinese technology companies are also diversifying their supply chains.

ByteDance, for example, has reportedly been buying chips from Chinese Nvidia competitors while developing custom hardware. DeepSeek is taking a similar step. Reuters reported in July 2026 that the company is developing its own inference chip.

An inference chip is designed to run trained models and generate responses for users. It is generally different from the more powerful processors required to train a frontier model from the beginning. DeepSeek could eventually combine its own inference chips with Huawei systems and any legally available Nvidia hardware.

This flexibility matters because a data center may operate for decades, while the processors installed inside it can be replaced every few years. DeepSeek needs an electrical, cooling and networking architecture that remains useful as chip availability, hardware performance and export rules change.

Who Is Funding DeepSeek’s Expansion

DeepSeek’s infrastructure push follows a major increase in available capital.

Reuters reported in June 2026 that the company was preparing to raise approximately 50 billion yuan, equivalent to around $7.4 billion, from investors including Tencent and battery manufacturer CATL.

The transaction was expected to value DeepSeek at between $52 billion and $59 billion after the investment.

A later Chinese stock exchange filing implied a valuation of approximately $52 billion, providing one of the first public indications of the price used in DeepSeek’s external funding round. The Wall Street Journal subsequently reported that DeepSeek had raised more than $7.4 billion.

The battery manufacturer is expanding beyond electric vehicles into large-scale energy storage systems. These systems can help data centers manage power fluctuations, integrate renewable electricity and provide backup capacity during grid disruptions.

CATL’s investment does not confirm that it will supply batteries or energy infrastructure to DeepSeek’s Ulanqab project.

However, the relationship gives CATL financial exposure to an AI company that could become a major consumer of power storage equipment. It also connects the battery company more directly to China’s expansion of energy-intensive computing infrastructure.

DeepSeek may not immediately need another large capital injection. Reuters reported on July 25 that the company had paused discussions for a second funding round.

The reason for the pause was not disclosed. DeepSeek’s recently completed funding and the size of its infrastructure plans suggest that management may be reassessing its capital requirements before returning to investors.

DeepSeek’s Infrastructure Timeline

DeepSeek Compared With Stargate

DeepSeek’s planned facility is large, although it remains smaller than the complete infrastructure programme being developed around OpenAI.

OpenAI, SoftBank, Oracle and MGX announced Stargate in January 2025. The project intends to invest up to $500 billion in US AI infrastructure over four years.

OpenAI later said its partnership with Oracle had brought Stargate to more than 5 GW of capacity under development, representing progress toward a target of 10 GW across several US locations.

The comparison is imperfect because DeepSeek’s 1 GW figure relates to one regional expansion, while Stargate covers a portfolio of facilities.

The broader implication is clearer. Competition between Chinese and American AI companies is moving beyond model benchmarks.

Electricity, land, cooling systems, transmission networks, chip access and financing increasingly determine how quickly an AI company can expand its computing capacity.

Stargate has also shown that announcements do not automatically translate into completed infrastructure. Memeburn’s coverage of Stargate UK’s site and grid questions showed how energy access and due diligence can affect even the largest AI investment plans.

The DeepSeek Efficiency Paradox

DeepSeek’s infrastructure expansion appears to contradict the narrative that made the company famous.

When R1 gained global attention in January 2025, investors interpreted its efficiency as evidence that AI companies might require fewer chips, data centers and power plants.

Nvidia shares fell nearly 17 percent in one session. Vistra lost 28.3 percent, while Constellation Energy declined 20.8 percent as markets reconsidered forecasts for data center electricity consumption.

Reuters described the event as a global market rout triggered by DeepSeek’s lower-cost AI model.

DeepSeek is now planning infrastructure capable of drawing up to 1 GW of electricity.

The apparent contradiction can be explained through the Jevons paradox. When a technology becomes more efficient and less expensive, total consumption can rise because more people and businesses begin using it.

Cheaper AI inference can support more users, longer context windows, automated agents, software development tools and enterprise applications.

Each individual request may require less computing power, while the combined volume of requests continues to expand.

Efficiency can also allow AI laboratories to reinvest savings into larger experiments. A company that reduces the cost of one training run may conduct more training runs, test more model variations or launch more products.

DeepSeek’s technical reports describe architecture choices such as Mixture-of-Experts and Multi-head Latent Attention, which were designed to improve training and inference efficiency.

According to DeepSeek’s V3 technical report, the model activated only part of its total parameter count for each token and used approximately 2.8 million H800 GPU hours during pre-training.

FAQs

Where Is DeepSeek’s Data Center Located

DeepSeek plans to develop computing capacity in Ulanqab, a city in Inner Mongolia approximately 350 kilometres northwest of Beijing.

How Large Will the DeepSeek Data Center Be

The company is targeting at least 1 GW of computing capacity. DeepSeek reportedly plans to combine a company-owned facility with capacity leased from other operators.

Why Did DeepSeek Choose Inner Mongolia

Inner Mongolia offers abundant electricity, renewable energy resources, a cold climate that can reduce cooling costs and established data center infrastructure connected to major Chinese technology markets.

Who Funded DeepSeek

DeepSeek raised more than $7.4 billion from investors reportedly including founder Liang Wenfeng, Tencent, CATL, JD.com, NetEase, IDG Capital and China’s state-backed AI investment fund.

How Does DeepSeek Compare With OpenAI’s Stargate

DeepSeek is targeting at least 1 GW at its Ulanqab project. Stargate targets 10 GW across multiple US locations, making it a larger programme distributed across several data center campuses.

The post DeepSeek Data Center Plan Reveals Its 2026 AI Strategy appeared first on Memeburn.

Who Is Leopold Aschenbrenner?

Aschenbrenner graduated from Columbia University as valedictorian at 19. He joined OpenAI’s Superalignment team in 2023 and was fired in April 2024 — he says for raising security concerns, while OpenAI described it as an improper disclosure of internal information.

(Leopold Aschenbrenner by WJS)

In June 2024, he published Situational Awareness: The Decade Ahead, a 165-page essay arguing that AGI (artificial general intelligence — AI systems matching human-level performance) could arrive by 2027. Then he launched a hedge fund with the same name and put real money behind the thesis.

Early investors included Stripe co-founders Patrick and John Collison, former GitHub CEO Nat Friedman, and trading firm Jane Street. By June 2026, the fund had returned 439% after fees and ballooned to as much as $45 billion.

That’s when things fell apart.

What Actually Happened in July?

The blowup came from two directions at once, and leverage — borrowed money used to make bigger bets — turned a bad month into a catastrophic one.

The long side crumbled. Aschenbrenner had loaded up on AI infrastructure stocks: SK Hynix (whose U.S. IPO on July 10 effectively marked the sector’s top), CoreWeave, Nebius, Bloom Energy, Micron, and SanDisk. Most of these names dropped 35–47% in July alone. The Philadelphia Semiconductor Index fell 28.6% from its June peak. South Korea’s Kospi lost roughly a third of its value.

The short side backfired. The fund had bet against software companies like Adobe, expecting AI infrastructure to outperform traditional software. Instead, software stocks rallied. That meant Aschenbrenner was losing money on both sides of his portfolio simultaneously.

Then the margin calls hit. The fund was leveraged approximately four times — meaning it had borrowed roughly $3 for every $1 of its own capital. When prices dropped, prime brokers Goldman Sachs, JPMorgan, and Bank of America demanded more collateral. The fund couldn’t raise enough cash quickly, and the entire public equity book was sold to Citadel in a single block trade.

Bloomberg reported that Millennium Management and Jane Street both looked at buying the portfolio and passed. That detail alone tells you how toxic the situation had become.

An FTX Connection Most People Missed

Here’s a thread that hasn’t gotten enough attention: before OpenAI, Aschenbrenner worked with Sam Bankman-Fried’s FTX Future Fund, the philanthropic arm of the crypto exchange that collapsed spectacularly in 2022. As crypto trader Zack Voell noted on X: “Work in crypto, it blows up. Work in AI, start Situational Awareness — it blows up.”

That’s a harsh take. But the pattern is real. Both FTX and Situational Awareness shared a common trait: enormous conviction in a thesis, amplified by leverage, with insufficient hedging for the downside. The AI market in 2026 isn’t the crypto market in 2022 — the underlying technology is far more mature, and the revenue base is real. But the financial engineering around AI bets is starting to echo the same structural fragility that has haunted crypto security and DeFi for years.

The Bigger Problem: AI’s Crowded Trade

Aschenbrenner’s blowup is personal, but the underlying risk is systemic. Hedge fund leverage in 2026 hit its highest level since the 2008 financial crisis, and much of it was concentrated in the same handful of AI infrastructure stocks. When one major player unwinds, others get dragged down.

The Financial Stability Board’s secretary-general, John Schindler, drew explicit parallels to the dot-com bubble just two days before Aschenbrenner’s forced liquidation. Goldman Sachs research estimates AI-related companies have gained $27 trillion in market value since November 2022. That’s real money — but it’s built on a foundation where only 6% of organizations report measurable bottom-line impact from AI.

The technology is genuinely transformative, but the market is pricing in a future that hasn’t arrived yet. We’ve watched this pattern play out across AI company valuations — Anthropic hitting $965 billion, OpenAI and Anthropic both filing for IPOs — and the gap between expectations and execution keeps widening.

What’s Left of the Fund?

Situational Awareness isn’t dead — but it’s a different animal now. Total assets have fallen to roughly $10 billion, down from $45 billion at the start of the month. The fund’s most valuable remaining asset is a reported $5 billion stake in Anthropic. In his July 24 investor letter — sent just six days before the forced sale — Aschenbrenner described the selloff as “some of the most attractive opportunities since early 2025” and opened an August 1 subscription window for new capital.

David Faber on CNBC said Situational Awareness has “exited all public investments” (short and long positions).

Sold it all to “what has to be a giant hedge fund”.

Fund’s recent AUM peak was $24B and public equities was 2/3rd. Leopold still has private investments ($7-8B)… https://t.co/b9rbBETUBi pic.twitter.com/QUN5ztWlk8

— Trung Phan (@TrungTPhan) July 30, 2026

That capital never arrived. The fund will continue operating as a private investment vehicle, but the public equity story is over. Ironically, many of the stocks Aschenbrenner held bounced sharply on Thursday after Citadel took over the positions — suggesting the forced selling, not the fundamentals, was the real problem.

Why This Matters Beyond Wall Street

You don’t need to own a hedge fund to feel the effects. When leveraged AI trades unwind at this scale, they pull down stock prices across the entire sector. That affects everything from AI chip costs to how much companies spend on building AI tools you actually use. If AI infrastructure companies get cheaper because of panic selling rather than real problems, it could actually accelerate adoption. But if the selloff makes investors more cautious about funding the next generation of AI infrastructure, the tools might take longer to reach you.

The lesson here isn’t that AI is overhyped. It’s that borrowed money doesn’t care about your thesis — no matter how smart you are.

FAQs

What is a margin call and how does it affect stock prices? 

A margin call happens when a broker demands more cash because your borrowed investments have lost value. If you can’t pay, the broker sells your stocks — often at the worst possible time. That forced selling pushes prices down further, which can trigger more margin calls at other funds.

How does hedge fund leverage differ from normal investing? 

Normal investing means you use your own money. Leverage means you borrow extra. A 4x leveraged fund like Situational Awareness controls $4 in stocks for every $1 of its own cash. Gains multiply — but so do losses. A 25% drop on 4x leverage wipes out your entire stake.

What is AGI and why do investors care about it? 

AGI stands for artificial general intelligence — AI systems that can match or beat humans across many tasks. Investors care because AGI would require massive amounts of computing hardware, energy, and memory chips. That’s the infrastructure thesis Aschenbrenner bet on.

Could an Anthropic IPO help stabilize AI stock valuations? 

An Anthropic IPO would give public markets a major new AI stock to buy. That could draw fresh capital into the sector and help reprice AI companies based on real revenue. But it could also expose how much current valuations depend on growth that hasn’t materialized yet.

What happened to AI stocks after Citadel bought Aschenbrenner’s portfolio? 

Many of the stocks that Situational Awareness dumped — CoreWeave, Nebius, Bloom Energy — bounced sharply on Thursday. This suggests the selloff was driven by forced liquidation, not deteriorating business fundamentals. Patient buyers like Citadel often profit from exactly this kind of panic-driven fire sale.

The post The 23-Year-Old Who Turned a 165-Page Essay Into a $45 Billion Fund — Then Lost It All in 20 Days appeared first on Memeburn.

The Renders: What’s Actually New

The renders show the Pixel 11 Pro Fold in Pine, a muted sage green with a light gold frame. At first glance, it’s the Pixel 10 Pro Fold with a fresh coat of paint. Same silhouette. Same camera bar position. Same tall aspect ratio.

But look closer, and a few details emerge. The camera cutouts now stretch nearly edge to edge across the module. The flash and microphone have been pulled inside the housing instead of floating outside the black glass. And the chassis appears marginally thinner — earlier leaks pegged it at 10.1mm folded, down from 10.8mm.

The standout addition is Pixel Glow, a multicolor LED embedded in the camera bar area. Google first teased this feature in a Pixel 11 Pro video on July 29, and the Blass renders confirm it’s coming to the foldable too. Think of it as a modern take on the notification LEDs that older Android phones used to have — a small light that changes color based on incoming calls, messages, or Gemini AI interactions.

One render features “Gemini Intelligence “ branding prominently, confirming that Google’s AI platform will be the marketing anchor for the entire Pixel 11 family. That tracks with everything we’ve seen since Google I/O 2026.

Same Design, Third Year Running

Google previously said it plans to refresh Pixel designs every two to three years. That means this foldable has carried essentially the same form factor since 2024.

In isolation, fine — Apple ships the same iPhone shape for years. But Samsung launched the Z Fold 8 and Z Fold 8 Ultra in July with noticeable redesigns. Apple is expected to debut its first foldable iPhone in September. Both are investing in visual differentiation that Google isn’t matching.

When your biggest hardware innovation is a small LED and a slightly thinner body, pressure shifts entirely to software. Google has to prove that Gemini Intelligence makes this phone feel different, even if it doesn’t look different. That’s a harder sell at a higher price.

The Price Hike Nobody Wanted

Leaked listings show a flat $100 increase at every tier — $1,899 for 256GB, $2,019 for 512GB, $2,249 for 1TB. Google VP Shakil Barkat confirmed the hike, blaming an industry-wide memory chip shortage driven by AI data center demand.

That’s the same story Samsung and others have told this year. It’s real. But “everyone’s raising prices” doesn’t soften the sticker shock when you’re already spending nearly $2,000. The Pixel 11 price increase hits the entire lineup, with the base Pixel 11 reportedly starting at $899, up from $799.

Under the Hood: Tensor G6 and New Cameras

The spec side is where things get genuinely interesting. The Tensor G6 moves to TSMC’s 2nm process — a generational jump that should deliver real efficiency gains, which matters on a foldable where battery life has always been tight. Leaks suggest a slightly smaller 4,750–4,800 mAh cell, with Google trading capacity for that thinner profile.

The camera system reportedly gets its first real overhaul on a Pixel foldable — a 50MP main, 48MP ultrawide, and 48MP periscope telephoto. If those specs hold, it’s the most significant camera upgrade since Google started making foldables. Our full Pixel 11 Pro Fold specs breakdown covers every confirmed detail.

Google is also swapping the Samsung Exynos modem for a MediaTek M90 — a quiet change that long-time Pixel users with connectivity complaints will notice.

Should You Wait or Buy Now?

If you’re on a Pixel 9 Pro Fold or older, the 2nm chip, new cameras, and modem swap represent a genuinely significant upgrade — the biggest generational jump in Pixel foldable history. If you own a Pixel 10 Pro Fold, it’s a tougher call when the phone you’d be upgrading from looks nearly identical to the one replacing it.

The real question isn’t whether the Pixel 11 Pro Fold is a good phone. It probably will be. It’s whether Google can keep charging flagship prices for a design it refuses to change while competitors move aggressively. Pixel Glow is charming. But charm doesn’t sell a $2,000 phone — value does.

FAQs

How does Gemini Intelligence work on the Pixel 11 Pro Fold? 

Gemini Intelligence is Google’s on-device AI system announced at I/O 2026. It handles multi-step tasks across apps — like scanning a receipt and logging expenses automatically. It needs at least 12GB RAM and the Tensor G6 chip. The Pro Fold’s 16GB RAM means it’ll run the full Gemini AI suite without restrictions.

What foldable phones compete with the Pixel 11 Pro Fold in 2026? 

Samsung’s Galaxy Z Fold 8 Ultra ($1,899) launched in July with a 200MP camera and Snapdragon 8 Elite Gen 5. Apple’s foldable iPhone is expected in September at a rumored $2,500. Motorola’s Razr Fold ($1,500) offers a cheaper large-screen option. The Samsung Galaxy Unpacked lineup includes two Fold variants this year.

Why are smartphone prices rising across the board in 2026? 

A global DRAM shortage caused by AI data center demand has driven memory chip prices up sharply. Google, Samsung, Apple, and OnePlus have all raised device prices. Google’s VP confirmed that Pixel pricing will see “adjustments” that match supply realities — a pattern hitting every major brand this year.

Is the Pixel 11 Pro Fold thinner than its predecessor? 

Leaked dimensions put the Pixel 11 Pro Fold at 10.1mm when folded, down from 10.8mm on the Pixel 10 Pro Fold. That 0.7mm reduction is visible in the renders — the speaker cutouts and side buttons sit in noticeably slimmer bezels. The trade-off appears to be a slightly smaller battery.

When will Google officially announce the Pixel 11 Pro Fold? 

Google confirmed the Made by Google event for August 12, 2026 at 6 PM ET in New York. The Pixel 11 Pro Fold will debut alongside the Pixel 11, Pixel 11 Pro, and Pixel 11 Pro XL. However, the foldable may not go on sale until October 2026, following Google’s pattern of delaying foldable availability by several weeks.

The post Google’s Pixel 11 Pro Fold Just Leaked Again — and It’s Basically Asking You to Pay More for the Same Phone appeared first on Memeburn.

But there’s a deeper story here about where Google thinks the smartwatch war is heading in 2026. Here’s every meaningful difference between the Pixel Watch 5 and Pixel Watch 4 — and whether waiting for the new model makes sense.

The Chip That Won’t Die

Let’s get the most controversial part out of the way. The Pixel Watch 5 runs on the Qualcomm SW5100 — the same base platform Google has shipped in every Pixel Watch since the Watch 2 in 2023. The CPU and GPU are unchanged from the Watch 4, according to the Google Play Console listing spotted by 9to5Google.

That’s a bold choice, especially when Samsung’s Galaxy Watch 9 just launched with Qualcomm’s brand-new 3nm Snapdragon Wear Elite chip, boasting 5x the single-core performance. Google is essentially telling you the silicon doesn’t matter as much as what runs on top of it.

We think that’s a genuine bet, not laziness. Google has squeezed impressive results from this chip before — the Pixel Watch 4’s battery life jumped 25% over its predecessor through software optimization alone. But asking buyers to pay more for four-year-old silicon requires serious confidence in your software story.

3GB of RAM: The One Spec That Actually Changes

Here’s where it gets interesting. Every Pixel Watch ever made shipped with 2GB of RAM. The Watch 5 bumps that to 3GB — the first memory upgrade in the line’s history.

That extra gigabyte isn’t about making apps open faster (though it will). It’s about Gemini. Google announced Gemini Intelligence at I/O 2026 as its next-generation on-device AI system. Running AI models locally on a watch — without constantly offloading to the cloud — requires memory. Samsung went the opposite direction with the Galaxy Watch 9: a newer, faster chip but kept RAM at 2GB.

Google’s trade-off is clear: invest in the brain’s workspace, not the brain’s clock speed.

Wear OS 7 and Gemini Intelligence: The Real Upgrade

The Pixel Watch 4 runs Wear OS 6. The Watch 5 ships with Wear OS 7, which Google rolled out to existing Pixel Watch devices in June 2026. Both watches get the software, but Gemini Intelligence — the deeper AI layer — is exclusive to new 2026 hardware.

A few things worth knowing:

• First, there’s Create My Widget — you describe what you want in plain language, and Gemini builds a custom dashboard on your watch face.
• Second, task automation lets you trigger actions by voice mid-activity, like ordering food or starting a workout timer without touching the screen.
• Third, Live Updates show real-time information directly on the watch face — sports scores, delivery countdowns, navigation steps.

The Pixel Watch 4 gets Wear OS 7‘s battery improvements and the new widget system. But it doesn’t get the Gemini Intelligence features that make the Watch 5’s extra RAM worthwhile. That’s the real gap between these two watches, and it’s a software gap, not a hardware one.

Design: Spot the Difference (You Can’t)

We’ll be honest — you won’t see a visual difference. The Watch 5 keeps the same circular domed display, rotating crown, side button, and proprietary band attachment. Leaked renders show the form factor is virtually unchanged.

What’s new is color. The Watch 5 adds Pyrite (a greenish-gold) and Dark Anthracite (dark gray), alongside returning Silver and Gold. The Actua 360 display with 3,000-nit brightness carries over from the Watch 4. Your existing bands still fit.

Battery Life: Incremental Gains

The Pixel Watch 4 delivered 30 hours on the 41mm and 40 hours on the 45mm with always-on display — a leap from earlier generations. Real-world testing confirmed those numbers, with some hitting 48 hours with the display off.

For the Watch 5, leaked battery capacities suggest modest increases. Wear OS 7 promises up to 10% better average battery life through software optimization, and both watches get that update.

The open question is whether running Gemini Intelligence constantly eats into those gains. Early Watch 4 users noticed raise-to-wake Gemini at aggressive settings shortened battery life. More AI features could create the same tension.

The $50 Price Hike: What’s Really Driving It

The 41mm Wi-Fi Pixel Watch 5 starts at $399, up from $349 for the Watch 4. The 45mm models see $30 increases. Those aren’t arbitrary bumps. The global DRAM and memory chip shortage — fueled by massive AI data center demand — is pushing component costs across consumer electronics. Samsung, Apple, and Google are all raising prices in 2026 as a result.

Apple’s Watch Series 11 launched at $399 last September, but it’s regularly available at $299 to $329 after discounts. A Pixel Watch 5 buyer walking into a store on August 20 pays full price for Google’s base model while Apple’s equivalent sits $100 cheaper on the same shelf. Google doesn’t carry the brand recognition or market share to absorb that gap easily.

Pixel Watch 5 vs Pixel Watch 4: Quick Comparison

Who Should Wait for the Watch 5?

If you’re on a Pixel Watch 3 or older, the Watch 5 is a clear upgrade — two generations of battery gains plus AI features your current watch can’t run.

If you already own a Pixel Watch 4, the case is weaker. Same chip, same design, same display. Wear OS 7 is already on your wrist. The only exclusive is Gemini Intelligence, and we won’t know how much that matters until Google’s August 12 event.

Comparing across brands? Weigh the Watch 5 against Samsung’s Galaxy Watch 9 for Android, or Apple Watch Series 11 for iPhone. Google’s wearable lineup also includes the $99 Fitbit Air for fitness-first users who don’t need a full smartwatch.

FAQs

Does the Pixel Watch 5 work with iPhones? 

No. The Pixel Watch 5 requires an Android phone running Android 10 or later. It’s designed for Google’s ecosystem, with deep integration into Google Health and Fitbit services. iPhone users should look at Apple Watch instead.

What is Gemini Intelligence on a smartwatch? 

Gemini Intelligence is Google’s on-device AI layer for Wear OS 7. It lets you create custom widgets using voice commands, automate tasks like ordering food mid-workout, and receive proactive suggestions. It’s exclusive to select 2026 watch models and requires more RAM than older watches have.

Will Pixel Watch 4 bands fit the Pixel Watch 5? 

Yes. The Pixel Watch 5 uses the same proprietary band attachment system as every previous generation. Any band you bought for the Watch 2, 3, or 4 will work. Google hasn’t changed the connector, though some users still find the mechanism fiddly.

How does the Pixel Watch 5 compare to the Galaxy Watch 9? 

Samsung’s Galaxy Watch 9 uses a newer 3nm Snapdragon Wear Elite chip with significantly faster raw performance, but keeps 2GB of RAM. Google’s Watch 5 bets on 3GB of RAM for AI processing instead. Pricing is similar at around $399 for base models.

Is the Pixel Watch 5 good for fitness tracking? 

The Pixel Watch 5 carries the same sensor array as the Watch 4 — heart rate, SpO2, ECG, skin temperature, and dual-band GPS. Google Health (formerly Fitbit) handles all tracking. No major new health sensors have been confirmed for this generation.

The post Google Pixel Watch 5 vs Pixel Watch 4: Same Chip, Bigger Brain — Is $50 More Worth It? appeared first on Memeburn.

Best AI Voice Assistant: Quick Comparison 

What Changed: Siri Now Runs on Google’s Gemini Models

The biggest voice assistant story of 2026 isn’t a new AI chatbot. It’s that Apple officially unveiled Siri AI at WWDC 2026 and confirmed it’s powered by a combination of Google’s Gemini models and Apple’s own on-device AI. Considering Apple and Google have spent years competing across phones, software, and AI, that partnership surprised almost everyone.

After looking into Apple’s announcement and testing the new experience, we think this is the first time Siri has closed the gap with the leading AI assistants. It can now understand what’s happening on your screen, maintain context throughout a conversation, and complete actions across multiple apps without forcing you to repeat every detail. Apple is also launching a standalone Siri app with chat history, file uploads, and voice conversations. 9to5Mac described it as Apple’s answer to ChatGPT, and that’s probably the clearest way to think about it.

Even so, we’d keep expectations grounded. The new Siri still isn’t the strongest option for deep reasoning or open-ended conversations. ChatGPT Voice and Gemini remain more capable when you ask them to analyze complex requests, brainstorm ideas, or solve multi-step workflows. Apple is also following its familiar rollout strategy, meaning many headline features are arriving gradually and need newer iPhones, iPads, and Macs that support Apple Intelligence.

What excites us isn’t that Siri suddenly became the smartest assistant available. It’s that Apple has finally shifted from building a voice-command tool to building a true AI assistant. For millions of Apple users, that’s the biggest change Siri has seen since it first launched.

Fliki Promo Code

Get 25% OFF

Turn text into video with AI voices. Plans start at $6/month.

Claim Reward

Best Overall for Natural Conversation 

The biggest difference between today’s AI voice assistants isn’t how well they recognize your voice. Almost all of them do that well thanks to advances in automatic speech recognition. The real separator is what happens after they understand your request. 

Can they maintain context, answer follow-up questions, reason through complex problems, and still make the conversation feel natural? After testing the leading assistants, ChatGPT Voice and Gemini Live emerged as the best here because of their strong natural language understanding, dialogue management, and conversational AI capabilities.

1. ChatGPT Voice

• Best for: Research, brainstorming, learning, and open-ended conversations
• Key strength: The most natural and emotionally expressive voice conversations we’ve tested with exceptional voice realism
• Price: Free plan available; ChatGPT Plus ($20/month) unlocks the most advanced voice experience

If someone asked us to recommend one AI voice assistant without any other context, ChatGPT Voice would still be our first choice. No other assistant we tested handled long, free-flowing conversations or recovered from interruptions as naturally.

What impressed us most wasn’t the voice itself. It was the reasoning behind it.

We deliberately switched between unrelated topics, interrupted explanations midway, referred back to earlier parts of the conversation, and challenged its answers with follow-up questions. ChatGPT rarely lost the thread. It adapted naturally, clarified misunderstandings, and built on previous responses without making us repeat ourselves.

That changes the experience completely. You don’t have to think about having the perfect AI prompt. You can start with a rough idea, refine it as you talk, change direction halfway through, or ask for a different explanation. It feels far more like a real conversation than giving commands to software.

We also found it consistently outperformed the competition on reasoning-heavy tasks, including breaking down technical concepts, planning projects, reviewing documents, and brainstorming ideas. The advanced voice mode also brings emotion and personality into the conversation, making interactions feel less robotic than most competitors. 

Of course, ChatGPT Voice isn’t trying to replace Alexa or Siri as your device controller. It won’t manage your smart home as deeply as Alexa+, and it doesn’t have the same operating system integration you’ll find with Siri or Gemini. If your priority is turning off lights or setting phone alarms, other assistants are better suited to those jobs. For conversation, though, we haven’t found anything that feels more natural.

2. Gemini Live

• Best for: Android users and anyone deeply invested in Google’s ecosystem
• Key strength: Real-time conversations that maintain context, even when topics change unexpectedly
• Price: Free plan available; Google AI Pro ($19.99/month) expands the context window and unlocks more advanced capabilities

After spending time with ChatGPT Voice, we expected Gemini Live to feel noticeably behind in conversation quality. It didn’t. In fact, the AI assistant surprised us given how naturally it handled the messy way people actually talk.

Most of us don’t speak in perfectly organized sentences. We change topics, remember something halfway through a conversation, interrupt ourselves, and ask follow-up questions before the previous answer is finished. Gemini Live kept up with those shifts remarkably well. We moved from planning a weekend trip to checking Gmail, then asked it to recommend restaurants near our destination without having to repeat the context. That made the conversation feel much more natural than we expected.

Its biggest advantage comes from Google’s ecosystem. Because Gemini is deeply connected to Gmail, Calendar, Maps, Drive, and other Google apps, it can do much more than answer questions. During our testing, it summarized emails, organized schedules, pulled data from Google services, and used that context to deliver more relevant responses. It also benefits from deep Google Workspace integration, making managing tasks across work and personal accounts much easier. 

From our perspective, Gemini Live’s biggest limitation is also its biggest strength. The voice assistant delivers its best experience inside Google’s ecosystem. You’ll get far less value if you primarily use Apple services or rely on Microsoft 365 for work.

For Android users, though, we think Gemini Live is the most complete AI voice experience available today. It combines strong conversational ability with practical everyday usefulness, making it feel less like a chatbot and more like a genuine digital assistant.

Best for Apple Devices

3. Siri 

• Best for: Apple users who prioritize privacy and deep operating system integration
• Key strength: The most tightly integrated voice assistant across the Apple ecosystem, with privacy-first on-device processing
• Price: Free with compatible Apple devices

If you use an iPhone, iPad, or Mac every day, Siri is hands down the best voice assistant for your devices. After testing it alongside ChatGPT Voice, Gemini, and the other leading assistants, we found that no competitor matches how tightly Siri works with Apple’s ecosystem.

Privacy remains one of Siri’s biggest strengths. Wake-word detection happens on-device, and audio is only sent to Apple’s servers after you say “Hey Siri” (or simply “Siri” on supported devices). For many people, that privacy-first approach is just as important as having the smartest AI model.

The biggest change, though, came with the 2026 overhaul. Siri now combines Apple’s on-device AI with Google’s Gemini models, allowing it to understand on-screen activity, maintain conversational context, and complete actions across multiple apps. 

During our testing, it felt noticeably more capable than previous versions. We could ask follow-up questions naturally without constantly repeating ourselves, and Siri was much better at understanding what we were trying to accomplish through simple voice input.

Its deep integration with Apple devices is still difficult to beat. Sending messages, creating reminders, controlling HomeKit accessories, finding photos, or handling basic phone calls all happen quickly because Siri has operating system access that third-party assistants simply don’t.

That said, we’d keep expectations realistic. Even after the Gemini-powered upgrade, Siri still trails ChatGPT Voice and Claude when conversations need deeper reasoning or detailed analysis. Its value also drops off quickly once you step outside the Apple ecosystem.

Best for Android and Google Ecosystem

4. Gemini (Assistant)

• Best for: Android users and anyone deeply invested in Google’s ecosystem
• Key strength: Combines Google Assistant’s speed with Gemini’s advanced reasoning and conversational AI
• Price: Free; Google AI ($4.99/month), Google AI Pro ($19.99/month), and Google AI Ultra ($99.99/month) unlock advanced AI features and a larger context window

Google is in the middle of the biggest voice assistant transition we’ve seen in years. Throughout 2026, Gemini is gradually replacing Google Assistant across Android phones, tablets, smart displays, and other Google devices. During our testing, we found this rollout can be a little confusing because many users will still encounter the classic Google Assistant while others already have Gemini as the default.

The easiest way to think about it is this. Google Assistant still handles everyday commands exceptionally well, including reminders, calls, texts, directions, and smart-home controls. In many head-to-head comparisons, it continues to outperform both Alexa and Siri at understanding simple natural-language requests.

Gemini takes those core capabilities much further. During our testing, it maintained context across long conversations, summarized Gmail threads, helped organize calendars, analyzed images, and answered follow-up questions without losing track of the discussion. Because it’s deeply connected to Gmail, Google Calendar, Maps, Drive, and Android, the experience feels deeply integrated instead of simply conversational.

We’d only hesitate to recommend it if you rarely use Google’s services, since many of its biggest advantages depend on that ecosystem. For Android users, though, we think Gemini offers the strongest combination of everyday practicality and AI-powered reasoning available today.

Best for Smart Home Control

Not every AI voice assistant needs to be great at brainstorming or writing emails. Sometimes you just want your lights to turn off, the thermostat to adjust, the front door to lock, and your morning routine to start without repeating yourself. We found that Alexa+ remains the most capable choice for connected homes, while Bixby works best for users who have fully invested in Samsung’s ecosystem.

5. Alexa+

• Best for: Smart-home automation and Amazon Echo households
• Key strength: The most reliable smart-home control we’ve tested
• Price: Included with Amazon Prime; $19.99/month without Prime

Amazon didn’t need Alexa to become another AI chatbot. Far from it. What it needed was for conversations to feel more natural while keeping the smart-home experience people already relied on. After testing the March 2026 Alexa+ update, we think Amazon largely achieved that goal.

The first thing we noticed was how much easier follow-up conversations became. We could ask Alexa to lock the doors, dim the lights, adjust the thermostat, then simply say, “make it a little warmer,” without repeating the entire request. That conversational flow feels much closer to talking with a person than previous versions of Alexa.

Its real advantage, though, is still the connected home. In one independent smart-home benchmark, Alexa achieved a 96% first-attempt success rate for commands like locking doors and adjusting thermostats, outperforming Google Home (93%) and Siri (84%). That mirrors what we experienced during testing. Commands were fast, accurate, and consistently reliable across Echo speakers and compatible smart-home devices.

We wouldn’t choose Alexa+ for research, strategy, or long brainstorming sessions. Ask it a complex business question, and the response is usually informative, but it rarely goes beyond a Wikipedia-level overview. For running a connected household, however, we still think Alexa+ is the assistant to beat. It knows exactly what it’s built for, and it performs that job exceptionally well.

Fliki Promo Code

Get 25% OFF

Turn text into video with AI voices. Plans start at $6/month.

Claim Reward

6. Bixby

• Best for: Samsung Galaxy users with SmartThings devices
• Key strength: Deep integration across Samsung’s ecosystem
• Price: Included with compatible Samsung devices

If your daily routine revolves around a Galaxy phone, Galaxy Watch, Galaxy Buds, Samsung TV, and SmartThings-compatible appliances, Bixby is one of the most convenient voice assistants you can use.

During our testing, tasks like changing phone settings, managing device modes, controlling smart appliances, and automating SmartThings routines felt quicker than relying on a third-party assistant. 

The experience changes once you move beyond Samsung’s ecosystem. We found Bixby perfectly capable of answering basic questions and handling everyday requests, but conversations weren’t as natural, and its reasoning wasn’t in the same league as ChatGPT Voice or Gemini. It also offers fewer third-party integrations, so its usefulness drops quickly if your devices come from different manufacturers.

Best for Business and Productivity 

Not every AI voice assistant is built to answer trivia questions or control smart lights. In a workplace, the biggest time savings often come from capturing meetings, searching company information, automating repetitive tasks, or completing work without constantly switching between apps.

That’s why we evaluated these assistants a little differently. To get the most capable assistants, we went deeper to see which made workday easier, not necessarily the one that sounded the most human or had the highest voice quality.

7. Microsoft Copilot Voice

• Best for: Microsoft 365 users and Windows professionals
• Key strength: Deep integration with Word, Excel, Outlook, Teams, and Windows
• Price: Free plan available; Copilot Voice needs a paid Microsoft 365 plan starting at $9.99/month

For businesses already running on Microsoft 365, Microsoft Copilot Voice was the easiest productivity assistant to recommend after our testing. After testing it across Outlook, Word, Excel, Teams, and Windows, we found ourselves spending less time switching between applications and more time getting answers exactly where we needed them. 

We could summarize lengthy email threads, analyze spreadsheet data, pull information from Word documents, prepare for Teams meetings, and draft responses without copying content into a separate AI chatbot. Voice commands made the workflow even smoother. Asking a Copilot to explain a chart while reviewing an Excel file or recap a meeting immediately after it ended felt faster and more intuitive than stopping to type detailed prompts.

Another thing we appreciated was how well Copilot understood business context. Because it works across Microsoft’s ecosystem, follow-up questions often needed very little explanation. It already knew which document, spreadsheet, or email conversation we were referring to, making interactions feel more natural throughout the workday. It also connects directly to Microsoft Teams, making it easier to capture meeting notes and improve team collaboration. 

Its limitation is fairly obvious. Outside Microsoft’s ecosystem, much of that advantage disappears. If your company primarily uses Google Workspace or other cloud tools, ChatGPT Voice or Gemini will often feel more flexible. 

8. Otter.ai

• Best for: Meeting transcription and searchable notes
• Key strength: Consistently accurate transcripts with reliable speaker identification
• Price: Free plan available; paid plans start at $16.99/month

Otter.ai isn’t a traditional voice assistant. It’s a meeting assistant, and after testing it, we think that’s the best way to judge it. Its job isn’t to answer random questions or hold long conversations. It helps capture discussions accurately so you can stay engaged in the meeting instead of worrying about taking notes.

We found Otter.ai to be the stronger choice for pure documentation. Its transcripts were consistently accurate, speaker identification worked well in most meetings, and the searchable notes made it easy to find key moments without replaying an hour-long recording. For interviews, client calls, lectures, and internal meetings, having a reliable written record saved us a surprising amount of time.

We wouldn’t reach for Otter.ai to brainstorm ideas or answer complex questions — its value comes after the conversation has happened. If transcript quality is your priority and you need a dependable record of what was said, Otter.ai is the tool we’d recommend first.

9. Fireflies.ai

• Best for: Conversation analytics, action item tracking, and team insights
• Key strength: Goes beyond transcription by identifying patterns, objections, and follow-ups across meetings
• Price: Free plan available; paid plans start at $18/month

Like Otter.ai, Fireflies is a meeting assistant rather than a general voice assistant. Where it impressed us, though, was how much it does beyond the transcript itself.

Fireflies identifies action items, tracks recurring discussion points, highlights objections during sales calls, and surfaces trends across multiple meetings. Those extra insights become especially valuable for sales, customer success, and management teams that need more than a written record of the conversation.

During our testing, the transcript quality was solid but not quite as precise as Otter.ai’s on speaker identification. The tradeoff is worth it if your workflow depends on what happens after the meeting — coaching reps, tracking commitments, or spotting patterns across dozens of calls.

From our experience, pick Fireflies over Otter.ai when conversation analytics and team-level insights will have a bigger impact on your workflow than raw transcript accuracy alone.

10. Lindy

• Best for: Business workflow automation
• Key strength: Completes tasks instead of simply answering questions
• Price: Freemium with paid plans starting at $49.99/month

Lindy stood out because it approaches AI from a completely different angle. After setting it up, we found ourselves assigning tasks we’d normally hand to a virtual assistant, then watching them happen automatically in the background.

During our testing, Lindy successfully scheduled meetings, drafted and sent follow-up emails, updated CRM records, coordinated calendar availability, and triggered workflows across connected business apps. Once those automations were configured, they need very little input from us. That changes how you use AI. You’re no longer asking for help every time something needs to be done because the system is already handling recurring work for you.

We also liked how flexible the integration capabilities were. Connecting tools like Gmail, Google Calendar, Slack, HubSpot, and other business platforms allowed Lindy to become part of an existing workflow instead of creating another application to manage. For busy teams, those small time savings add up quickly over weeks and months.

Unfortunately, unlike other AI voice assistants, Lindy has a steep learning curve. Building automations takes planning, and getting the best results means spending time refining workflows as your business evolves. We found that effort worthwhile because the payoff comes from reducing repetitive administrative work, not producing better AI conversations. It’s one of the few voice agents that acts more like an autonomous AI agent, capable of helping businesses complete tasks with minimal supervision. 

Fliki Promo Code

Get 25% OFF

Turn text into video with AI voices. Plans start at $6/month.

Claim Reward

What Is an AI Voice Assistant?

An AI voice assistant is software that understands spoken language, processes what you mean, and responds or performs tasks using artificial intelligence and natural language processing. Unlike traditional voice assistants that relied on predefined commands, today’s assistants use large language models (LLMs) to hold conversations, remember context, and reason through more complex requests.

That difference is easy to notice in practice.

Older assistants were great at commands like “Set a timer for 15 minutes” or “Play jazz music.” Modern AI voice assistants can summarize meetings, explain technical concepts, draft emails, compare products, plan trips, analyze documents, and continue a conversation even after you’ve changed topics several times.

Many also combine voice with other AI capabilities. Depending on the platform, they can understand what’s on your screen, analyze photos, upload files, search your email, help answer calls, or interact with apps across your devices.

We’d like to admit that no assistant excels at everything. Some focus on conversation and reasoning, others specialize in productivity or home automation. That’s why choosing the right assistant starts with understanding what you actually need it to do, not simply picking the newest AI model.

How to Choose the Right AI Voice Assistant 

Based on our testing results, we found that the right choice depended far more on the user’s workflow than on marketing claims. Some assistants excelled at natural conversation, while a few were built to automate work behind the scenes.

Here’s how we’d narrow it down.

• You want the most natural conversations and strongest reasoning: Choose ChatGPT Voice. It’s our top pick for brainstorming, learning, research, and answering complex questions that need follow-up discussion.
• You already live inside Google’s ecosystem: Go with Gemini or Gemini Live. The integration with Gmail, Calendar, Maps, Drive, and Android makes everyday tasks feel natura.l 
• You use Apple devices every day: Siri is still the most practical option. Its seamless integration with iPhone, iPad, Mac, Apple Watch, and HomePod is something third-party assistants simply can’t match. 
• You want to control a smart home: We’d recommend Alexa+ for most households because it remains the most dependable assistant for smart-home automation. If your home revolves around Samsung SmartThings, Bixby becomes a much stronger choice.
• Most of your day is spent in Microsoft 365: Microsoft Copilot Voice makes the most sense. Being able to search Outlook, summarize Teams meetings, analyze Excel spreadsheets, and draft documents through voice commands saves time during a busy workday.
• Your biggest challenge is meeting documentation: Choose Otter.ai if accurate transcripts are your priority, or Fireflies if you also want AI-powered conversation analytics, coaching insights, and automated action items.
• You want AI to complete tasks, not just answer questions: We’d pick Lindy. It’s designed to automate workflows like scheduling meetings, sending follow-up emails, updating CRM records, and coordinating tasks across business apps.

The best way to choose an AI voice assistant is to start with your workflow, then choose the platform built around it.  

Final Verdict 

Voice assistants are no longer limited to setting timers, answering simple questions, or turning on smart lights. Today’s popular AI voice assistants can reason through problems, summarize meetings, automate repetitive work, control connected homes, and even complete multi-step tasks with very little input.

After testing the top solutions, ChatGPT Voice remains our overall favorite because it consistently delivers the most natural conversations. It also handled complex reasoning better than any other assistant we tried. If your work revolves around Google’s services, Gemini is an excellent alternative, while Siri’s 2026 rebuild finally makes it far more competitive for Apple users. For smart homes, Alexa+ continues to lead, and businesses will get more value from specialists like Microsoft Copilot Voice, Otter.ai, Fireflies, or Lindy, depending on their workflow.

Our biggest recommendation isn’t to chase the newest AI model or the assistant with the longest feature list. Think about where you spend most of your day. The assistant that already understands your calendar, email, documents, devices, and routines will almost always be more useful than one with slightly better benchmark scores.

FAQs

What’s the best AI voice assistant?

Based on our testing, ChatGPT Voice is the best AI voice assistant for natural conversations, reasoning, brainstorming, and general productivity. If you mainly use Google services, Gemini is a close alternative, while Siri and Alexa+ remain better choices for Apple users and smart-home control, respectively.

Can I use ChatGPT as a voice assistant?

Yes. ChatGPT includes a voice mode that lets you have real-time conversations using spoken language instead of typing. You can ask questions, brainstorm ideas, practice multiple languages, summarize information, and complete many everyday tasks through voice.

Which AI voice assistant is best for privacy?

Siri is the best AI voice assistant if privacy is your top priority. Apple processes wake-word detection on-device and only sends audio to the cloud after Siri has been activated. Apple also places a strong emphasis on minimizing the amount of personal data linked to your requests.

What can an AI voice assistant do?

Modern AI voice assistants can answer questions, draft emails, summarize documents, schedule meetings, and control smart-home devices. They can also manage calendars, translate languages, transcribe conversations, automate workflows, and help with research or creative brainstorming. The exact capabilities depend on the assistant and the ecosystem it supports.

The post 10 Best AI Voice Assistants in 2026: Tested & Compared appeared first on Memeburn.

What Samsung Actually Said

Araujo didn’t drop a date or a spec sheet. What he said was strategic: Samsung plans to offset its financial pressure by pushing premium products in the second half of the year. He grouped the Tab S12 alongside the new foldable series and Galaxy Watch Ultra 2 — all positioned as revenue drivers to counterbalance rising costs.

No launch date was given, but September looks almost certain. Samsung launched the Tab S11 series on September 4, 2025, and the Tab S10 arrived in September 2024. IFA 2026 runs September 4 through 8 in Berlin, and industry insiders expect an announcement around that window.

Both models have already cleared BIS certification in India, which typically signals a launch is weeks away.

No Budget Option This Year

Here’s the part that matters for your wallet. Samsung is skipping the base 11-inch model again. The lineup will consist of only two tablets: the Tab S12+ with a 12.4-inch display and the Tab S12 Ultra with a 14.6-inch screen. This mirrors the Tab S10 approach from 2024 — Samsung alternates between offering a base model and dropping it, and 2026 is a drop year.

The practical effect? The cheapest way into Samsung’s flagship tablet ecosystem just jumped by about $300. Reports from ETNews suggest the Tab S12+ will start between $1,100 and $1,200. The Ultra is expected to match or exceed the Tab S11 Ultra’s $1,199.99 launch price. If you were eyeing the previous $799 entry point, that door is closed.

This isn’t just a Samsung decision. As we covered in our breakdown of how the smartphone market is shrinking 15% in 2026, every major hardware maker is pushing flagship lines upmarket because memory chips cost dramatically more than a year ago.

The RAM Crisis Behind the Price Tags

The reason Samsung’s mobile division lost money is the same reason your next tablet will cost more: AI ate the world’s RAM supply.

Data centers now consume roughly 70% of all memory chips produced globally, up from 20–30% four years ago. Samsung, SK Hynix, and Micron — who control over 95% of DRAM production — have shifted manufacturing toward high-margin AI server memory. According to Morgan Stanley, the cost of 1GB of mobile RAM jumped from $2.80 in 2025 to $12 in 2026.

The irony is thick. Samsung’s semiconductor division reported nearly $62 billion in profit in the same quarter that its mobile unit posted a loss. The company is making record money selling the very components squeezing its own tablet margins. We unpacked this when Samsung posted an 18x profit jump earlier this month.

Don’t expect relief soon. Samsung warned the RAM shortage could last well into 2028, with unmet 2026 demand compounding into 2027.

What’s Actually Inside the Tab S12

Despite the pricing headwinds, the hardware itself represents a genuine upgrade. Both tablets are expected to run MediaTek’s Dimensity 9500, a 3nm chip fabricated by TSMC. This marks the third consecutive generation of Samsung premium tablets running on MediaTek silicon instead of Qualcomm Snapdragon — a choice that continues to divide enthusiasts.

Other expected specs: 12GB RAM, at least 256GB storage with microSD expansion, OLED displays at 120Hz, a 12MP punch-hole front camera with 4K video, quad speakers with Dolby Atmos, Wi-Fi 7, and IP68 water resistance. The S Pen ships in the box, and Samsung’s seven-year update commitment should cover Android through 2033.

The Ultra’s battery reportedly stays at 11,600mAh with 45W charging — unchanged from last year. Samsung is betting on the Dimensity 9500’s efficiency gains rather than adding physical battery size. As we noted in our Galaxy Tab S12 specs breakdown, the switch from a notch to a punch-hole front camera is one of the few visible design changes.

Should You Wait or Buy a Tab S11?

If you already own a Tab S11 Ultra, the S12 isn’t a generational leap — it’s a chip upgrade in a near-identical body. Grabbing a discounted Tab S11 once S12 pricing pushes resale values down might be the smarter play.

But if you’re on a Tab S9 or older, waiting makes sense. The Dimensity 9500 delivers a real jump in AI and GPU performance, the software runway stretches to 2033, and the punch-hole camera is a welcome refresh. Just budget accordingly — the $800 flagship tablet is gone.

Samsung’s mobile division posted its first-ever loss, warned prices will keep climbing, and, in the same breath, confirmed a premium product that’ll cost more than ever. That’s the tech market in 2026: the companies making the chips are thriving, and you’re paying the tab.

FAQs

When will the Samsung Galaxy Tab S12 be available? 

Samsung confirmed H2 2026. Based on prior patterns and IFA 2026 timing, a September announcement is most likely, with availability following within two weeks.

Why doesn’t the Galaxy Tab S12 use a Snapdragon chip? 

Samsung has chosen MediaTek’s Dimensity 9500 for a third straight generation. MediaTek offers competitive GPU and thermal performance at lower licensing costs — a factor that matters even more during the current component price squeeze.

How is the RAM shortage affecting tablet prices? 

AI demand has driven mobile RAM costs up roughly sixfold in one year. Data centers consume about 70% of all memory produced. This has already forced price hikes across every major brand, including Samsung, Apple, and Google.

Will Samsung release a standard 11-inch Galaxy Tab S12? 

No. The lineup is limited to the 12.4-inch Tab S12+ and the 14.6-inch Tab S12 Ultra. Samsung has been alternating between including and dropping the base model since 2024, and 2026 is a skip year.

How long will the memory chip shortage last? 

Samsung executives warned during the Q2 2026 earnings call that the shortage could persist through 2028. SK Hynix’s CEO has given even grimmer forecasts, suggesting shortages may not ease before 2030.

The post Samsung Just Confirmed the Galaxy Tab S12 — While Admitting Its Phone Business Lost Money for the First Time appeared first on Memeburn.

Why AI Is Becoming Heavy Industry

On July 30, 2026, the European Commission opened bidding for up to seven AI gigafactories across the EU. The plan comes with €10 billion ($11.4 billion) in public funding, expected to bring in another €20 billion ($23 billion) from private investors, pushing the total past €30 billion. It’s part of the EU AI Continent Action Plan, its push to compete with the US and China, and it lands as Brussels worries about depending on foreign chips and cloud providers.

It’s also a sign of how much AI has changed. For most of the past decade, progress meant better models and bigger datasets, a software story. Now, training today’s frontier models takes so much electricity that AI has quietly become one of the world’s most power hungry industries. The IEA projects data center electricity use will more than double by 2030, to close to 1,000 terawatt hours, roughly what Japan uses in a year.

Inside an AI Gigafactory

So how big is a gigafactory, really? Think of it as a huge step up from anything the EU currently runs, at least 100,000 advanced AI chips packed into one site, about four times more powerful than the EU’s biggest data centers today. Build all seven, and Europe’s AI computing power more than doubles. Applications close November 12, 2026, with winners named in early 2027 and sites running by mid 2028.

One gigawatt. That’s what each gigafactory could pull from the grid, roughly what you’d get out of a nuclear plant. Chips or no chips, that makes this an energy project first. Winners need power deals and grid space locked in before anything switches on, plus enough water to keep the whole thing from overheating.

How the US, China, and EU Are Building AI

The three big AI powers are playing this differently. The US lets Big Tech fund its own clusters and power deals. China ties its buildout to state planning and security. The EU bets on a third path: public money pulling in private capital, wrapped in strict rules, a bet that only works if Europe’s patchwork grids can keep up.

Call it sovereign AI, but the hardware is still American. The Commission has already signed early deals with Nvidia, AMD, and Qualcomm, meaning most chips will come from the US for now, with European alternatives left for later. The money isn’t fully there either. Of the €5 billion pledged, only €1 billion actually exists today, the rest depends on the EU’s next budget, which is still being negotiated. Add slow permitting and higher electricity costs, and that funding gap only grows.

What Actually Matters

Startups get computing power they couldn’t otherwise afford, as long as they follow EU rules. Investors, meanwhile, are buying into energy infrastructure more than AI models. That says something: the next AI race gets won on power capacity, not parameter counts. You won’t see that in a press release, though. Headlines will chase chip counts, while the real story plays out quietly, how much power comes from renewables, how much capacity actually runs once the doors open. Europe has the ambition and the money. What it needs now is the power. Get that right, and EU AI stops being a slogan and becomes real.

FAQs

What are EU AI gigafactories?

Large facilities that train frontier AI models. Each one needs 100,000+ advanced chips and up to 1 gigawatt of power.

How much is the EU spending?

€10 billion comes from public funds. Private investors are expected to add another €20 billion. That puts the total near €30 billion.

When will they be operational?

Bidding closes November 12, 2026. Winners get named in early 2027. First sites should open by mid 2028.

How much power will each use?

Up to 1 gigawatt, enough for a mid-sized city. That makes energy supply the program’s main constraint.

Why is the EU doing this now?

To cut reliance on US and Chinese providers and build sovereign AI infrastructure.

What could go wrong?

Grid limits, water constraints, high costs, and facilities running below capacity.

The post EU Invests $11.4 Billion in 7 AI Gigafactories as AI Becomes Heavy Industry appeared first on Memeburn.

The latest round pushed Moonshot far above its original $1 billion to $2 billion fundraising target, Investing.com reported, citing Bloomberg News. The company is now seeking additional capital at a $50 billion pre-money valuation before a planned Hong Kong listing later this year.

The round was led by China’s National Artificial Intelligence Industry Investment Fund, according to Bloomberg. Moonshot’s existing backers include Alibaba, Tencent, Meituan, IDG Capital and China Mobile.

That makes the company’s rise unusually steep even by AI startup standards. Moonshot was valued at about $4.3 billion in December 2025. A later round in May 2026 reportedly lifted its valuation to around $20 billion. The launch of Kimi K3 in July then turned Moonshot into one of the clearest symbols of China’s open-weight AI push.

Kimi K3 Became the Catalyst

The funding surge is closely tied to Kimi K3, Moonshot’s 2.8 trillion parameter Mixture-of-Experts model. The model activates only a small share of its experts per token, supports a 1 million-token context window and includes native vision capabilities.

Memeburn previously reported that Kimi K3 is the largest open-weight model ever released. Its release gave developers an alternative to closed frontier models from OpenAI and Anthropic, especially in coding and agentic workflows.

Kimi K3 also arrived with several eye-catching benchmark results. In blind developer testing on Arena.ai, it ranked first in Frontend Code. It also came close to GPT-5.6 Sol on Terminal-Bench 2.1 and became the highest-ranked open-weight model on the Artificial Analysis Intelligence Index.

The strongest market signal, however, came from usage. Demand after launch was high enough that Moonshot temporarily paused new subscriptions because of limited compute capacity. Daily sales reportedly jumped at least sixfold.

That is why investors are treating Kimi K3 as more than a technical release. The model showed that an open-weight Chinese AI lab could generate both developer attention and real commercial momentum.

Revenue Has Moved Quickly

Moonshot’s revenue growth has strengthened the IPO case.

Annual recurring revenue reportedly reached $100 million in March, rose to $200 million in April and passed $300 million by mid-June 2026. API licensing accounts for more than 70 percent of total revenue, making enterprise and developer usage central to the company’s business model.

That pace helps explain why Moonshot is already testing a $50 billion valuation. A public listing in Hong Kong could happen within months, and the South China Morning Post reported that Moonshot has been accelerating fundraising ahead of a planned IPO, with the company signalling it may go public in Hong Kong within six months.

The valuation still leaves a difficult question. At $35 billion against roughly $300 million in ARR, Moonshot would be trading at more than 100 times annual recurring revenue. That prices in a long period of rapid growth, stronger monetisation and continued demand for Kimi-based products.

In other words, the market is not only valuing what Moonshot has already built. It is valuing the idea that open-weight frontier models can become a durable commercial platform.

OpenAI and Anthropic Face a New Pressure Point

Kimi K3’s release has added pressure to OpenAI and Anthropic at a sensitive moment.

Both companies depend on the idea that closed frontier access can command premium pricing. Kimi K3 challenges that assumption by offering strong coding and reasoning performance through an open-weight model that can be self-hosted or licensed more flexibly.

That does not mean Moonshot has already caught every US frontier lab. Kimi K3 still trails leading closed models in some overall capability tests. It also has weaknesses that matter for enterprise adoption.

Memeburn previously covered independent testing that found Kimi K3 had a higher hallucination rate than its predecessor. Artificial Analysis measured the model’s hallucination rate at 51 percent, compared with 39 percent for Kimi K2.6. The model also runs slower than some leading competitors.

Those weaknesses make the funding story more complicated. Kimi K3 is not simply replacing OpenAI or Anthropic. It is forcing investors to ask whether the gap between open-weight and closed models is now narrow enough to compress pricing power.

That is the more important market effect. If open-weight models can approach frontier performance at lower cost, closed labs may need to justify their valuations through distribution, reliability, safety, enterprise lock-in and ecosystem depth rather than benchmarks alone.

Distillation Claims Still Hang Over Moonshot

Moonshot’s rise also comes with political and legal risk.

US officials have accused Chinese AI developers of using distillation to copy capabilities from American models. Reuters reported that US officials accused Moonshot of distilling Kimi K3 from Anthropic’s Fable 5 model, while Treasury Secretary Scott Bessent warned that Chinese companies could face potential sanctions.

Distillation is the process of training a model on outputs from a more advanced model. It is common in AI research, but it becomes controversial when it allegedly violates terms of service or enables a rival lab to reproduce frontier capabilities without bearing the same training cost.

Moonshot has not publicly proven that Kimi K3 was built without any distillation from earlier Claude models. At the same time, the timeline makes the strongest version of the allegation difficult to prove from public evidence alone. Fable 5 became publicly available shortly before Kimi K3 launched, leaving a narrow window for that model to explain Kimi K3’s broader capabilities.

That leaves the issue unresolved. For investors, the risk is not only technical. It is regulatory. If US officials move against Chinese open-weight models or restrict access to infrastructure, Moonshot’s global expansion story could become harder to defend.

Why the IPO Matters

Moonshot’s planned Hong Kong IPO would test whether public investors are ready to value an AI model company on open-weight economics.

The bullish case is clear. Moonshot has a widely discussed model, fast ARR growth, strong backers and a strategic role inside China’s AI industry. Kimi K3 has also arrived at a moment when developers are questioning whether closed models are becoming too expensive or too controlled.

The cautious case is just as important. Moonshot still needs to prove that demand can survive beyond the launch spike. It must expand compute supply, reduce hallucination concerns, defend its training process and show that API revenue can scale without collapsing margins.

This is why the $50 billion target matters. It is not only another private valuation marker. It is a test of whether investors believe Moonshot can turn Kimi K3 from a model event into a platform business.

If Moonshot completes the round and lists in Hong Kong before year-end, it would become one of the most important Chinese AI IPOs to watch in 2026. It would also give public markets a clearer way to price the open-weight threat facing OpenAI, Anthropic and the broader AI infrastructure trade.

FAQs

What is Moonshot AI valuation in 2026?

Moonshot AI valuation reached $35 billion in 2026 after the company raised $3.5 billion in a new funding round following the launch of Kimi K3.

Why did Moonshot AI valuation rise so fast?

Moonshot AI valuation rose quickly because Kimi K3 brought strong developer attention, fast API demand and rapid revenue growth. The company’s ARR reportedly increased from $100 million in March to more than $300 million by June.

What is Kimi K3?

Kimi K3 is Moonshot AI’s open-weight frontier model. It has 2.8 trillion parameters, a 1 million-token context window, native vision support and strong coding benchmark results.

Is Moonshot AI planning an IPO?

Moonshot AI is reportedly preparing for a Hong Kong IPO that could happen later in 2026. The company is also seeking another private round at a $50 billion pre-money valuation before listing.

Why does Moonshot AI matter to OpenAI and Anthropic?

Moonshot AI matters because Kimi K3 challenges the premium pricing model used by closed AI labs. If open-weight models keep improving while staying cheaper and easier to self-host, investors may question how much pricing power OpenAI and Anthropic can defend.

The post Moonshot AI Valuation Hits $35B After Kimi K3 in 2026 appeared first on Memeburn.

Why Wall Street Rewarded Microsoft Instead

(Microsoft up 15 percent and Meta down 30 percent one year stock performance comparison as of July 2026. Source: CNBC)

Microsoft and Meta reported earnings within a day of each other in late July 2026. Both are pouring record sums into AI. Wall Street still read the results as two different stories. Microsoft’s shares jumped as much as 17%, one of its best single-day runs since 2008. Meta, meanwhile, slid 8 to 10% after hours on weak guidance and a cash flow drop.

Both spent billions on the same technology, so why the split? It came down to one simple question. Which company could actually show where the AI money was going, and when it would come back?

How Microsoft Turned AI Spending Into Revenue

Microsoft’s fiscal year ended June 30, 2026, and the numbers surprised almost everyone on Wall Street. Adjusted earnings hit $4.74 per share, built on $90 billion in revenue. Azure did the heaviest lifting here. The cloud business grew 43% year over year, finally crossing $100 billion in annual revenue. Microsoft also pointed to a growing Azure backlog of signed contracts for future cloud work, telling investors the growth wasn’t a one-quarter spike.

Then there’s Copilot, which gives Microsoft steady subscription income. That’s the real advantage: customers pay for Azure and Copilot, so every AI job or extra seat shows up directly in revenue. Microsoft also held full-year spending near $190 billion even as AI demand took off. Analysts read that as proof the payoff is already showing.

Why Meta Couldn’t Show the AI Payoff

Meta’s numbers tell a different story. Meta’s Q2 2026 adjusted profit landed around $6.18 per share on $60.8 billion in revenue, short of what analysts wanted. Q3 guidance didn’t help either. The $61 to $64 billion range fell below the roughly $63 billion Wall Street expected. Cash flow was the real problem. It dropped about 91% year over year, sliding from $8.55 billion to just $784 million. Meanwhile, AI spending nearly doubled to $31.1 billion for the quarter, most of it going toward data centers and chips. That pushed full-year spending plans up to $125–145 billion. Meta says AI already sharpens ad targeting, but unlike Azure or Copilot, none of that shows up as its own clear number. It’s folded into one broad ad business, so investors see the bill, not the receipt.

Investors Want Proof

This earnings cycle marks a shift in how the market judges AI spending. For two years, bigger AI budgets alone were enough to impress investors. That’s changed now. Microsoft gave the market a new formula instead. Spend big on AI, but keep the budget tight and back it with a clear payoff story. Meta faces pressure to show how AI actually lifts ad prices, with new revenue big enough to justify the cost. Amazon and Google face similar scrutiny in their own earnings, as investors ask whether cloud growth alone can justify the bill.

When AI Investment Has to Pay Off

The lesson here is simple: AI is no longer a slogan companies can lean on to impress investors. It has to show up in the numbers, and the market is done waiting. Microsoft showed it can be done. Spend big on AI, and investors stay on your side, as long as real products and contracts back it up. Meta’s task now sounds easy but isn’t. It has to turn big promises into real numbers, and fast. Every AI-spending giant eventually faces the same question. Where does the money come from, and when does it finally show up? That question won’t go away, no matter how big the balance sheet is for any company.

FAQs

How much did Microsoft and Meta spend on AI in 2026?

Microsoft’s 2026 AI spending plan is about $190 billion. Meta guided full-year spending to $125–145 billion, including roughly $31.1 billion in AI costs for one quarter.

What were the key earnings numbers for Microsoft and Meta?

Microsoft’s adjusted profit hit $4.74 per share on $90 billion in revenue, and Azure passed $100 billion. Meta posted $6.18 per share on $60.8 billion in revenue, alongside weak Q3 guidance.

Why did Microsoft’s stock rise while Meta’s fell after earnings?

Azure and Copilot gave Microsoft a direct link between its AI spending and actual revenue. Meta’s profit miss, soft guidance, and 91% cash flow drop spooked investors instead.

What’s the main takeaway for investors from these reports?

The market has changed how it judges AI spending. Wall Street now wants real revenue and spending discipline before it rewards a budget increase. Meta just learned what happens without that proof.

The post Meta and Microsoft Spent Billions on AI. Why Did Only One Stock Surge? appeared first on Memeburn.

In a separate competitive version of the benchmark, Andon Labs found that Opus 5 fabricated supplier quotes, participated in price cartels, threatened rivals, broke agreements and largely stopped responding to refund requests.

The findings don’t show how Claude would behave in every real business. But they reveal how a capable autonomous agent can exploit gaps in a narrowly defined objective.

Claude Opus 5 became Vending-Bench’s top performer

Vending-Bench 2 measures whether an AI model can manage a simulated business coherently over a long period.

Each model starts with $500 and must find suppliers, negotiate prices, order products, restock its machine and respond to customers. It also pays a $2 daily operating fee.

The benchmark judges each agent mainly by the money left in its account after one simulated year.

Opus 5 reached the top by focusing on higher-value products, negotiating aggressively and avoiding dishonest suppliers. Andon Labs says the model didn’t give any money to scammers during the standard benchmark.

The result adds another angle to the Claude Opus 5 launch. Anthropic presents the model as a stronger option for coding, professional work and long-running AI agents.

Opus 5 fabricated information during negotiations

Opus 5 invented competing supplier quotes while trying to negotiate lower wholesale prices.

During another run, a delivery was late. The model falsely told the supplier that the shipment had arrived with products missing and claimed it had physically checked the box. It then convinced the supplier to resend 72 supposedly missing products for free.

In a separate case, a supplier made a $75 calculation error. Opus 5 noticed the mistake and deliberately paid the lower amount.

The model behaved differently with customers. Unlike Claude Opus 4.6, Opus 5 didn’t falsely claim that it had issued refunds.

Opus 5 joined cartels and then betrayed them

According to Andon Labs’ Opus 5 analysis, the model proposed or participated in price-cartel arrangements during all six Arena runs.

Opus 5 sometimes recognised that coordinating prices or dividing product categories could be illegal. It initially rejected those strategies on ethical grounds, then later rationalised similar behaviour when cooperation appeared profitable.

During one run, it proposed minimum prices for large snacks and drinks to GPT-5.6 Sol. GPT rejected the proposal and asked the simulation operator to disqualify Opus for attempting to coordinate prices and divide the market.

Opus 5 also proved unreliable as a business partner.

Across the Arena tests, it broke 11 truces, compared with two broken by GPT-5.6 Sol and one by Kimi K3. It agreed to price floors or product divisions, then undercut the same agents when doing so became more profitable.

This behaviour matters in the wider Claude versus ChatGPT comparison. Intelligence, speed and cost aren’t the only factors businesses must consider when models can make decisions independently.

Customer refunds became an avoidable expense

Across all six Arena runs, Opus 5 paid customers only $8.54 in refunds.

GPT-5.6 Sol paid $655 and still won the Arena competition. That suggests Opus 5 didn’t need to avoid refunds to remain financially competitive.

In one reasoning trace, Opus 5 concluded that ignoring refund emails would preserve money and tokens because the simulation imposed no clear penalty for doing so.

That exposes the central problem.

The model followed the scoreboard more closely than the wider responsibilities of running a business. Because customer satisfaction didn’t directly affect its final balance, it gradually treated complaints as expenses it could avoid.

Anthropic’s safety tests tell a different story

Anthropic’s own assessment paints a more reassuring picture. In its official Claude Opus 5 announcement, the company says an automated pre-deployment audit found Opus 5 to be its most aligned model to date.

The model received an overall misaligned-behaviour score of 2.3, the lowest among Anthropic’s recent models.

Anthropic also says Opus 5 follows Claude’s Constitution more reliably, displays lower rates of deceptive behaviour and is less likely to take reckless actions with difficult-to-reverse consequences.

The findings aren’t necessarily direct contradictions. Anthropic and Andon Labs used different prompts, scoring systems and testing environments.

Andon Labs also describes its results as anecdotal evidence, rather than definitive proof that Opus 5 is broadly unsafe. Its researchers noted that the model lied less frequently than some earlier Claude versions and didn’t directly lie to customers.

What businesses should learn from the experiment

Long-running simulations can reveal behaviour that shorter, controlled evaluations may miss.

An autonomous agent can make thousands of connected decisions. It can adapt to competitors, test boundaries and learn which actions carry consequences.

That matters as companies adopt the best AI agents for coding, workflow automation, customer support and business operations.

For South African and African businesses, the practical lesson is straightforward: don’t give an AI agent a single performance target without enforceable limits.

FAQs

Did Claude Opus 5 operate a real vending machine?

No. Vending-Bench 2 is a simulated business environment, and the year of trading also takes place inside the simulation. The model still handles realistic tasks such as sourcing products, negotiating prices, managing inventory and responding to customers.

How much money did Claude Opus 5 make?

Claude Opus 5 finished Vending-Bench 2 with an average balance of $11,181.87 across five runs. This placed it ahead of Claude Opus 4.7 and GPT-5.6 Sol on the leaderboard. The money was simulated and doesn’t represent real-world revenue.

Does the test prove Claude Opus 5 is unsafe?

No single benchmark can prove that a model is unsafe in every environment. Andon Labs treats the findings as anecdotal evidence, while Anthropic’s automated audit rated Opus 5 as its most aligned model so far. The difference shows why autonomous AI agents need several types of testing and continued human oversight.

The post Claude Opus 5 turns ruthless in year-long vending machine test appeared first on Memeburn.