Strategic Foresight - Insights in Business and Technology

New eBook. - Four Boxes, Clear Thinking: Preview Now Available

2026-10-27T23:12:45.103+11:00

Four Boxes, Clear Thinking: Preview Release Now Available

I'm pleased to share a preview release of my latest eBook, Four Boxes, Clear Thinking: The Art and Science of Matrices That Drive Decisions.

Almost every strategy deck has a two sided matrix in it. Some are brilliant: they turn a messy argument into two lines and four boxes and help a room agree on a direction in minutes. Many more are decoration, with axes that overlap, lines drawn by eye and boxes that never tell anyone what to do.

This book is about the difference. It is a practical guide to building your own matrices from first principles, so they hold up when an executive, a procurement panel or a board starts asking questions.

This book explores:

Why the two by two matrix has outlasted its critics, and the science of why four boxes fit the mind
A seven step method: question, axes, thresholds, quadrant names, plotting, testing and the so what
Twelve common traps and how to avoid them
Commercial uses including company analysis, organic growth, risk and return, account planning and sourcing
Technology uses including architecture, cloud, digital transformation and networks
Industry playbooks across healthcare, care services, financial services, government, telecom, energy, retail, manufacturing and education
When not to use a matrix, and how to use AI to build and stress test one

Every matrix has been drawn fresh for this book, drawing on more than 26 years of business and IT advisory work.

Whether you are an executive, consultant, sales leader, strategist or anyone who builds decision slides, I hope you find it useful.

Download the eBook: Four Boxes, Clear Thinking: Download here: Link

I welcome your feedback, comments and suggestions before the final release.

New eBook Release: The AI Bottleneck Shift From Chips to Credit

2026-10-25T12:37:37.566+11:00

I'm pleased to share a preview release of my latest eBook, The AI Bottleneck Shift: From Chips to Credit.

Over the last three years, the AI conversation has focused on chips, GPUs, models and data centres. But a new bottleneck is emerging. The question is no longer "Can we get the chips?" but "Who will fund the trillions of dollars needed to build the future of AI?"

This book explores:

Why the AI bottleneck has shifted from silicon to capital
How Nvidia evolved from a chip supplier into a financial backstop for the industry
The rise of neoclouds and AI infrastructure financing
The impact of bond yields, interest rates and US government borrowing on AI
Why Anthropic's IPO may become the most important financial event in AI
Why OpenAI is waiting and what it tells us about the next phase of the market
What business leaders, investors and technology buyers should watch next

Drawing on research from SemiAnalysis, Aswath Damodaran, public market data and industry analysis, the book connects technology, finance and strategy into a single narrative.

Whether you are a technology executive, investor, entrepreneur, consultant or simply interested in where AI is heading, I hope you find the insights valuable.

Download the eBook

📖 The AI Bottleneck Shift: From Chips to Credit - Download here: Link

I welcome your feedback, comments and perspectives as the AI industry enters what may become its most important phase yet: the battle for capital.

AI Has a Money Problem, and Anthropic's IPO Is the Test

2026-10-23T18:07:42.863+11:00

Abstract

AI's bottleneck has moved. Three years ago the scarce thing was chips. Then it was power and land. Now it's money. Specifically, it's credit: who will lend, how much and at what rate.

October 2026 made that shift visible. Bond yields pushed above 5%. The Nasdaq sold off twice in a week. The listed neoclouds, the companies that rent out GPUs, lost about a fifth of their value in five trading days. An Australian AI data centre company called Firmus pulled its IPO. Nvidia, its supplier and shareholder, is now in talks to put more money in.

Into this market, Anthropic is pressing ahead with its own IPO. It's the only big tech listing still on the autumn calendar. It's growing faster than almost any software company in history. It also has more than $600 billion of compute commitments to pay for.

OpenAI, its great rival, has decided to wait until 2027. Microsoft, meanwhile, owns a slice of both.

So the Anthropic IPO is about more than one company. It's a test of whether public markets will fund the AI build-out now that credit has become the constraint. And because Anthropic is a young, loss-making company, it will be priced rather than valued. That price will reflect the credit cycle as much as it reflects Claude.

Five beats: one story

The big idea in one line

Anthropic's IPO is a referendum on whether public equity will fund AI now that credit is the bottleneck.

Hold on to that. Everything else hangs off it.

Ten things to take away

I'll explain each of these properly later. For now, here's the list.

Credit is the bottleneck. AI needs roughly $1 trillion of new funding every year. Chips and power are no longer the binding constraint.
The shock arrived. The US 10-year yield touched about 5.33% after a hot inflation print. The Nasdaq lost 4.5% in one week, then fell 2.6% in a single day on 22 October.
The casualties are the borrowers. Neoclouds fell about 22% in a week. Firmus pulled a $5 billion IPO. Switch and Nscale delayed theirs.
Nvidia is backstopping again. It's reportedly in talks to invest hundreds of millions in Firmus. That's vendor finance in plain sight.
Anthropic has growth and obligations. First-half 2026 revenue was $16.3 billion. Compute commitments top $600 billion, about 30 times its year-end cash.
Anthropic has reasons to go now. It needs to fund commitments, tidy its capital structure and sell its growth while it's the only AI IPO in town.
OpenAI has reasons to wait. Sam Altman says "not 2026". Private money is still available to it.
Microsoft wins either way. It owns about 27% of OpenAI. It put $5 billion into Anthropic, and Anthropic agreed to spend $30 billion on Azure.
Anthropic will be priced, not valued. Every pricing metric you'll hear hides assumptions about growth, margins and risk.
The timing is a forced choice. The window is poor. But Anthropic's need and its demand are strong enough to go anyway.

Act 1: The backdrop

Part I: This month in markets

Let's start with what happened. It's been a rough few weeks.

Two sharp drops in a week

The Nasdaq had a four-week winning streak. It ended abruptly.

On Friday 16 October, the index fell 2.1%. That capped a weekly loss of about 4.5%, its worst week since June.
It bounced on Wednesday 21 October, up about 1.1%.
Then on Thursday 22 October, it fell 2.6%. That was its worst single day since early June.

Chip stocks took the heaviest hit. The SOX semiconductor index dropped 6.7% on 22 October alone. It now sits more than 26% below its June record. It's worth remembering that the index is still up around 50% for the year. This is a sharp pullback, not a collapse. But it hurts.

Why it happened

No single thing caused this. Several worries landed at once.

Inflation ran hot. September CPI came in at 3.7%. Producer prices were firm too.
The Fed may hike. Markets now price a high chance of a December rate rise. The new Fed chair, Kevin Warsh, gave little away.
Yields jumped. The 10-year Treasury yield hit about 5.33% on CPI day. It later eased back to around 5.0% to 5.1% as money fled into bonds.
Oil spiked. Brent crude rose above $110 on reports of possible strikes on Iran after the US midterms.
AI spending spooked investors. Meta fell about 7% despite beating forecasts, because it flagged how much money AI will need.
Neocloud credit cracked. CoreWeave dropped 10% on 22 October. Its bonds and credit default swaps stayed under pressure.
Jensen Huang said the quiet part out loud. Nvidia's chief executive told CNBC that neocloud demand is "off the charts". Then he added that capital is scarce and holding growth back.

That last point matters most for this story. When the man selling the chips says the problem is money, not demand, you should listen.

The yield line to watch: In my earlier blogpost, From Chips to Credit, I flagged a level on the 10-year yield of about 5.25%.

Here's why it matters. Most of the time, when shares fall, investors buy government bonds. Bond prices rise. Your portfolio gets a cushion. That's the logic behind the classic 60/40 mix of shares and bonds.

Above roughly 5.25%, that cushion tends to vanish. Rising rates hurt shares and bonds together. Long-dated growth stocks, like AI companies, lose their hedge.

This month, yields crossed that line on CPI day and then slipped back below it. It's a fragile truce.

The further from cash, the bigger the fall

Look at who fell hardest in the week of 12 to 16 October.

The Dow rose about 0.7%.
The S&P 500 fell about 2%.
The Nasdaq fell about 4.5%.
Nvidia fell about 6.6%.
Chip stocks fell about 11.4%.
The eight listed neoclouds fell about 22% on average.

See the pattern? The more a business relies on borrowed money to build AI capacity, the harder it fell. Consumer staples and healthcare actually rose. The money didn't leave the market. It moved to companies that don't need to borrow to grow. That's the chips-to-credit story showing up in prices.

The Australian casualty: Firmus

Closer to home, Australia had its own AI drama.

Firmus is a neocloud. It designs and runs AI data centres, and Nvidia, Coatue, Blackstone and Jane Street all back it. It planned to list on the ASX at about A$11 a share. That would have valued it at roughly US$30.6 billion and raised about US$5 billion. It would have been one of the largest floats in Australian history.

On 9 October, it pulled the deal.

Here's the timeline.

April 2026: Firmus raises $505 million at a $5.5 billion valuation.
August 2026: a $2 billion round lifts its valuation above $10.5 billion.
9 September: Nvidia names Firmus in its 2-gigawatt Australian partner build-out.
5 October: data centre partner CDC walks away from Project Southgate. Only 42 megawatts of a planned 1.6 gigawatts had been built.
9 October: Firmus pulls its IPO.
21 October: reports say Nvidia is in talks to invest hundreds of millions more.

One fund manager summed it up neatly. UniSuper's chief investment officer said it was a good business story that "simply did not have a valuation that matched it".

Why investors said no - Firmus is the chips-to-credit story in miniature. Investors balked for clear reasons.

The asking price was about three times its August private valuation, just two months earlier.
The company carried around $30 billion of debt.
It had only two operating data centres, in Melbourne and Singapore.
It had little track record of building at scale.
A key partner had just walked away.
More than half the stock would have been free to trade on day one.

So what happens next? Firmus plans a private round of $2 billion to $3 billion. It may list on the Nasdaq later. And Nvidia, its supplier, may step in with fresh equity.

That last move is the lesson. When public equity says no, the vendor steps in. It's the same "balance sheet as a service" pattern we saw with Nvidia and CoreWeave.

The IPO calendar is down to one name - Firmus isn't alone. According to Reuters, the once-busy autumn calendar for tech IPOs is "down to just Anthropic".

Firmus pulled its IPO on 9 October.
Switch, a data centre operator, delayed.
Nscale, an AI infrastructure firm, delayed. Part of the reason was SEC staff comments.
SB Energy, which supplies power for AI, postponed.
OpenAI isn't coming this year. It's targeting 2027.
Anthropic is pressing ahead, aiming for mid-November.

Bankers report that other issuers don't want to compete with Anthropic for investors' attention. When one company is the only game in town, it gets the room to itself.

Two Australian stories, two sides of the chain - Australia gives us a neat illustration of the whole system.

On one side, Anthropic is the customer. In September it signed its first Australian data centre lease, a 1.44-gigawatt campus near Dalby in Queensland, reportedly worth about A$32 billion. Approvals are still pending.
On the other side, Firmus is a builder. It needs equity and debt to build AI factories. Public investors refused its price.

The link is simple. The lab signs the lease. The landlord needs credit. If Anthropic's IPO goes well, its leases become stronger collateral for everyone building for it. If it goes badly, that collateral weakens.

Part II: Chips to credit, the change underneath

The selloff is the weather. This part is about the climate.

The bottleneck moved

Think about what has constrained AI over the past few years.

In 2023 and 2024, it was chips. GPUs were on allocation. High-bandwidth memory was scarce. If you could get chips, you could build.
In 2024 and 2025, it was power and sites. Grid connections took years. Land and permits were tight. Developers solved this with capital, refurbished industrial sites and on-site generation.
From 2026, it's credit. The question is no longer "can we get the chips?" It's "who will lend us the money, and at what rate?"

This October simply made the shift visible. Huang says demand is off the charts. Neocloud bonds wobble. IPOs get pulled. Power now sits with whoever has a strong balance sheet, or whoever is willing to lend theirs to others.

The bill - The numbers here are enormous. The research firm SemiAnalysis estimates:

about $11.3 trillion of AI capital spending between 2024 and 2029, covering both chips and data centres;
about $7 trillion of debt outstanding by 2029;
about $1 trillion of new funding needed every year.

Let's make that concrete. A 100-megawatt AI campus costs roughly $40 million per megawatt, so about $4 billion in total. A young neocloud might have $400 million of equity. It has to borrow the other $3.6 billion.

Now multiply that across hundreds of campuses. Then add Anthropic's own commitments of more than $600 billion. The trillions add up fast.

How money moves through the stack

It helps to picture AI as a stack of layers.

At the top: users and businesses pay for tokens, seats and agents.
Next: AI labs like Anthropic and OpenAI sign multi-year compute contracts.
Then: clouds like Azure, AWS and Google, plus neoclouds, rent out GPU capacity.
Then: chipmakers like Nvidia supply the hardware.
Then: data centre and power companies build the sites.
At the bottom: banks, private credit funds, bond buyers and IPO investors supply the capital.

Capital flows up. Revenue flows down to repay it. Each layer's lender relies on the layer above to keep paying. Anthropic sits near the top. Its revenue is what ultimately repays the neocloud and data centre debt underneath it. That's why its IPO matters to companies that have never heard of Claude.

Balance sheet as a service - Here's the clever bit, and the risky bit.

Most AI labs and neoclouds don't have investment-grade credit ratings. Lenders prefer to lend against strong balance sheets. So strong companies lend their credit to weaker ones.

Nvidia has about $497 billion of commitments, according to SemiAnalysis. These include rental floors, lease and power guarantees, cloud agreements and guarantees on the resale value of GPUs.
Hyperscalers like Microsoft sit between AI labs and lenders. A lab signs with Microsoft. Microsoft signs with a neocloud. The neocloud borrows against Microsoft's signature, not the lab's.

Firmus this month is a live example. Public equity declined, so the vendor stepped in again.

This works well when times are good. When times turn, the supplier gets hit twice: once through lower sales, and again through the support it has extended. Telecom equipment makers learned that lesson painfully in 2001.

The gap that equity has to fill. SemiAnalysis estimates Nvidia's balance sheet can support roughly 46 gigawatts of backstops. The industry needs about 240 gigawatts of new capacity.

That leaves about 194 gigawatts that someone else must fund. That someone is hyperscalers, private credit, bond markets, asset-backed securities and equity investors. IPOs sit in that last group.

Equity matters because it's the cushion lenders lend against. Every successful AI IPO adds equity to the system. Every pulled IPO, like Firmus, takes some away. Anthropic's listing is the single biggest equity injection on the table.

How higher yields travel down the chain

October gave us a clear chain reaction.

Yields rose on hot inflation and rate-hike fears.
Neocloud credit spreads widened. CoreWeave's credit default swaps rose.
Neocloud shares fell about 22% in a week.
IPOs were pulled (Firmus) or delayed (Switch, Nscale).
Vendors stepped in as backstops (Nvidia and Firmus).
AI labs' commitments now rest on this whole chain.

There's a feedback loop here too. Lower share prices make equity harder to raise. Dearer debt makes projects harder to finance. That weighs on AI shares again. Anthropic sits right in the middle. Its commitments are the revenue that neocloud and data centre lenders count on. A strong IPO strengthens the whole chain. A weak one weakens it.

Where October 2026 sits - I find a simple two-by-two helpful here. Put AI demand on one axis and bond yields on the other.

Weak demand, high yields: danger. That's the 2001 telecom pattern.
Strong demand, high yields: the squeeze. Demand is there, but capital is dear.
Weak demand, low yields: patient capital. Projects slow, but refinancing is easy.
Strong demand, low yields: goldilocks. The build-out accelerates.

October 2026 sits firmly in the squeeze. Anthropic's revenue grew 45% in a single quarter. Huang says demand is off the charts. Yet yields sit around 5%. In a squeeze, the best-placed issuers have real demand and a genuine reason to raise money now. That describes Anthropic far better than a debt-heavy neocloud like Firmus.

Why credit sets IPO timing

So how does all this affect when companies list? Here's the logic.

Debt markets tighten. Higher base rates and wider spreads raise the cost of GPU-backed and project debt.
Equity has to fill the gap. Companies with large commitments need fresh equity to keep lenders comfortable.
But equity reprices too. The same higher rates lower what investors will pay for long-dated growth.
So timing becomes a trade-off. Go now at a lower price, or wait and risk funding gaps as commitments fall due.

Three companies made three different choices this month.

Firmus waited. It's debt-heavy, and its story didn't support its price.
OpenAI waited. Private money is still available to it.
Anthropic went. It has huge commitments, the strongest growth and a window with no rivals.

Act 2: The players

Part III: What Anthropic is all about

Let's meet the company at the centre of this story.

Who they are: Anthropic was founded in 2021 by a group of former OpenAI researchers, led by siblings Dario and Daniela Amodei. It's about five years old.

It started as a safety-first AI lab. It has become, in practice, an enterprise AI company.

Mission: build frontier AI safely. Anthropic is a public benefit corporation. It even warns investors about AI's risks in its own prospectus.
Product: the Claude family of models, sold through an API, through Claude apps and through Claude Code. Its models also run on AWS, Google Cloud and Microsoft.
Business model: enterprise-first and usage-based. Revenue grows with the work customers do, not with the number of people logged in.

Two numbers stand out. The founders' holding company owns a single Class F share that carries 50.1% of the vote on key matters, according to reports on the draft prospectus. And the session I drew on puts its active users at about 250 million, far fewer than OpenAI's, but much more valuable each.

Revenue: a near-vertical line

Anthropic's revenue growth is extraordinary. Here are the figures as reported.

2024 revenue: about $386 million.
2025 revenue: about $4.6 billion, roughly 12 times the year before.
Q2 2025: $787 million.
Q2 2026: about $11.5 bill
First half of 2026: $16.3 billion.
Q3 2026: $16.8 billion, up 45% on the previous quarter.
Annualised run rate in September: about $67 billion, with a target near $100 billion by year-end.

Claude Code has driven much of this. Developers and businesses use it heavily, and they pay by usage.

There's a catch. Revenue is concentrated. Two customers made up about a quarter of 2025 revenue and about a fifth of first-half 2026 revenue.

Losses: scary headline, more nuanced reality

The draft prospectus reportedly shows a net loss of $80.7 billion for the first half of 2026. That's one of the largest half-year losses ever reported. Don't panic just yet.

Most of it isn't cash.

About $72 billion came from an accounting charge on convertible notes. These are loans that can turn into shares. As Anthropic's valuation rises, those notes are worth more, and the increase is booked as a loss.
A similar charge, about $34 billion, drove most of 2025's roughly $42 billion net loss.

An IPO that converts those notes into shares should end the charge.

But the operating loss is real. Anthropic lost about $8.1 billion from operations in 2025. Its compute and infrastructure spending was about $7.3 billion in 2025. In the first half of 2026 alone, it was about $9 billion.

Commitments: the giant purchase order

This is the figure that links Anthropic to the credit story.

In September, reports put Anthropic's cloud and compute commitments at $518 billion, owed to six partners. About 80% are non-cancellable.
By October, the Wall Street Journal reported more than $600 billion, with up to $150 billion more under negotiation.
Anthropic ended 2025 with about $20 billion of cash.

Do the maths. Commitments are roughly 30 times year-end cash.

Two examples bring this to life.

•Microsoft: Anthropic agreed to buy $30 billion of Azure capacity as part of Microsoft's November 2025 investment.
Australia: the 1.44-gigawatt Dalby lease, reportedly about A$32 billion.

Those commitments can only be met by future revenue or new capital. That's the core reason Anthropic is going public.

Risks: 80 pages of them

Reports say the draft prospectus runs to about 261 pages. Around 80 of those are risk factors. Only 48 cover the business itself. That ratio tells you something.

The big risks include:

Safety and existential risk. The prospectus reportedly warns that AI could pose catastrophic risks and that models may try to resist being shut down.
Customer concentration. A handful of customers account for a large slice of revenue, and many aren't on long contracts.
A government dispute. The US Department of Defense designated Anthropic a supply-chain risk in March 2026. In September, the DC Circuit upheld that exclusion under federal procurement law. A district court set aside a separate designation in August.
Governance. The founders' Class F share controls 50.1% of key votes. Anthropic also has duties as a public benefit corporation.
Commitments. More than $600 billion, mostly non-cancellable.
Competition. OpenAI, Google and open-source models are all chasing the same customers.

What investors are actually buying

So what are you buying if you buy Anthropic shares? I'd split it into four boxes.

Big and visible: enterprise growth, Claude Code demand and high gross margins.
Known obligations: the $600 billion-plus of commitments, founder control and customer concentration.
Big but uncertain: the size of the AI market, how long Anthropic's advantage lasts and what its margins look like in ten years.
Smaller and uncertain: consumer reach and regulatory outcomes, including the defence dispute.

Investors can see the growth and the obligations clearly. What they can't see is the third box. That's exactly where pricing replaces valuation, as we will see in Act 3.

Growth engine, safety brand - and a giant purchase order

Part IV: Microsoft, backing both horses

Now for the company that has quietly placed a bet on both sides.

From OpenAI's partner to everyone's landlord

Here's how Microsoft's position has evolved.

2019: Microsoft invests $1 billion in OpenAI.
2023: total commitments to OpenAI reach about $13 billion.
October 2025: OpenAI restructures. Microsoft ends up with about 27%.
November 2025: Microsoft invests $5 billion in Anthropic. Anthropic commits to spend $30 billion on Azure.
April 2026: OpenAI's terms with Microsoft are reset. The relationship becomes non-exclusive.
July 2026: Microsoft books a $3.2 billion gain on Anthropic and a $600 million markdown on OpenAI in a single quarter.

Two days before those results, Satya Nadella reportedly warned that any company over-relying on a single AI lab "will not remain a firm". Microsoft practises what it preaches.

Two stakes side by side

Here's how the two investments compare.

Money in: about $13 billion into OpenAI since 2019, against $5 billion into Anthropic.
Ownership: about 27% of OpenAI, against a minority stake in Anthropic.
Cloud tie: both have committed to large Azure spending. Anthropic's commitment is $30 billion.
Accounting in Microsoft's 2026 financial year: OpenAI added $5 billion for the year but lost $0.6 billion in the fourth quarter. Anthropic added $3.2 billion in the fourth quarter alone.
Exclusivity: ended for OpenAI in April 2026. Never existed for Anthropic, whose models also run on AWS and Google.
In Microsoft products: GPT models power much of Copilot. Claude models now sit in Copilot and Foundry too.
IPO: OpenAI isn't coming before 2027. Anthropic plans mid-November 2026.

One quarter of Anthropic matched a year of OpenAI

This chart says a lot. Microsoft's OpenAI stake added about $5 billion across its whole financial year. Its Anthropic stake added $3.2 billion in a single quarter.

A word of caution. These are accounting marks, not operating results. Microsoft doesn't revalue Anthropic every quarter.

But the IPO will change that. For the first time, Microsoft's Anthropic stake will have a public price. That's one more reason Redmond will watch November closely.

Getting paid twice

Microsoft's lab deals follow a simple loop.

Microsoft invests equity in a lab.
The lab commits to buy Azure compute.
Azure revenue and Microsoft's order book grow.
The lab grows on Microsoft's compute.
The stake is marked up, and an IPO sets the price.

Microsoft gets paid twice: once through cloud revenue and once through the value of its stake. Critics call this circular. Supporters call it smart ecosystem building. Both can be true.

For scale, Microsoft's 2026 numbers are huge. Its contracted backlog is $678 billion. Azure revenue passed $100 billion, up 41%. Capex hit $115.9 billion. And it has $329 billion of data centre leases signed but not yet started.

Microsoft's hedge

Map Microsoft's model bets on two axes: how much it owns of each model maker, and how much infrastructure revenue it earns from them.

Anthropic: a small stake but a big compute contract. Supplier first.
OpenAI: a large stake and a large compute contract. Partner and owner.
Open models on Foundry: a marketplace. Low ownership, some revenue.
Its own MAI models on Maia chips: full control.

Whichever lab wins, Microsoft earns on compute and on equity. The Anthropic IPO validates that hedge. OpenAI's delay keeps Microsoft's biggest stake private and unpriced for now.

Even Microsoft faces the credit question

It's easy to forget, but Microsoft is part of the financing story too.

Operating cash flow rose to about $183 billion in its 2026 financial year.
Capex jumped to about $116 billion.
Free cash flow actually fell slightly, to about $67 billion.
Cloud gross margin has slipped to about 65%.

Microsoft can easily afford this. But when even the strongest balance sheet in tech sees free cash flow dip because of AI spending, you know how capital-hungry this build-out is.

Microsoft, Alphabet and Amazon report results next week. Their capex and cloud guidance will set the tone for the Anthropic roadshow, which is expected to start the week of 9 November.

Even Microsoft is now part of the financing question

Part V: Why Anthropic goes now, and why OpenAI waits

Two labs. One market. Opposite choices.

Five reasons Anthropic is going now

Fund the commitments. More than $600 billion of mostly non-cancellable compute, against about $20 billion of year-end cash. The gap can only be closed by future revenue or new capital.
Clean up the capital structure. The convertible notes drove $34 billion of non-cash charges in 2025 and $72 billion in the first half of 2026. An IPO can convert them into ordinary shares.
Sell growth at its peak. Q3 revenue rose 45% on the quarter. The run rate is heading from about $67 billion towards $100 billion. That's a compelling pitch.
Use a window with no rivals. The autumn calendar is down to Anthropic alone. Big investors see it as rare exposure to AI they can't ignore.
Strengthen the credit chain. Public equity makes Anthropic's leases and contracts stronger collateral for the neoclouds and developers building for it.

Why OpenAI hasn't listed

OpenAI quietly filed a confidential draft prospectus on 8 June 2026. Then it chose to wait.

Private money still flows. A pending private round is rumoured at about $1.4 trillion. Why accept a public discount?
There's a valuation floor. Altman reportedly called any valuation below $1 trillion a "non-starter". This market might not clear it.
Timing and safety. Altman told Fortune it would be "an ill-advised moment to go public". The CFO reportedly targets 2027.
Heavy cash burn. Reports project about $27 billion of cash burn in 2026 and about $63 billion in 2027. That's a hard story to sell to public investors right now.
A complex structure. OpenAI is a public benefit corporation controlled by a non-profit. Microsoft owns about 27% and takes a revenue share.
A consumer-heavy model. OpenAI has about 1.2 billion users but earns less per user than Anthropic.

OpenAI filed confidentially - then chose to wait

Side by side

Confidential prospectus: Anthropic filed on 1 June 2026. OpenAI filed on 8 June 2026.
Public prospectus: Anthropic's is expected as early as late October. OpenAI has none.
Planned timing: Anthropic markets from the week of 9 November and aims to trade before 26 November. OpenAI targets 2027.
Valuation talk: Anthropic $1.5 trillion to $2 trillion, against a last private price of $965 billion. OpenAI wants more than $1 trillion, against a last private price of $852 billion.
Run-rate revenue: about $67 billion for Anthropic on a gross basis. About $50 billion for OpenAI on a net basis.
Customers: Anthropic is enterprise-led and usage-based. OpenAI is consumer-heavy.

The timing matrix

Here's how I'd map the decision. Put the need for capital on one axis and the quality of the market window on the other.

Good window, low need: opportunistic. List if the price is right. SpaceX did this in June.
Good window, high need: go now. The easy decision.
Poor window, low need: wait. That's OpenAI.
Poor window, high need: forced choice. That's Anthropic, and it was Firmus too.

The difference between Anthropic and Firmus is telling. Both faced a forced choice. Anthropic has demand that investors want. Firmus had debt that investors feared. So Firmus fell back on Nvidia.

Is now the right time?

Let's be fair to both sides. The case for going now:

Scarcity. It's the only big AI IPO, so it won't compete for capital.
It's arguably the fastest-growing large software business on record.
It can raise equity before credit tightens further.
It can convert the notes and end those giant non-cash charges.
A lower entry price may suit long-term holders.

The case against:

The 10-year yield sits around 5%, and markets expect a December hike.
The Nasdaq has sold off twice, and chips are down more than a quarter from their peak.
Anthropic's own suppliers, the neoclouds, are under credit stress.
The headline loss, customer concentration and defence dispute will spook some investors.
Post-midterm geopolitical risk and $110 oil add uncertainty.

Four ways November could play out

Blockbuster: strong demand and calmer markets. It prices at $1.5 trillion to $2 trillion and pops.
Priced to sell: strong demand despite weak markets. It lists at a discount and trades up.
Downsized: weak demand, calmer markets. A smaller raise at a lower price.
Delay: weak demand and weak markets. It's postponed, Firmus-style.

My base case today sits in "priced to sell".

What tips the balance? Five things: the public prospectus, next week's big-tech earnings, the 10-year yield against 5.25%, CoreWeave's credit and the US midterms on 3 November.

Act 3: Pricing the IPO

This act draws on a valuation session about the Anthropic IPO that was published on YouTube in October. I've added this month's context where it helps. The core argument is simple and important: young companies get priced, not valued. Let's see why.

Part VI: Value versus price

The pre-IPO meeting: In mid-October, Anthropic met a group of prospective institutional investors. Reports date it to 14 October. The session I drew on says 15 October.

At that meeting, Anthropic shared its annualised revenue run rate, about $67 billion. It said it expected that figure to approach $100 billion by the end of 2026. A draft prospectus then leaked on 20 October. The public version is expected as early as late October, with trading planned before 26 November.

The choice of metric is telling. Anthropic talked about a run rate, not last year's revenue. That's because its story isn't just about large revenue. It's about fast-growing revenue.

What is ARR?

ARR stands for annualised revenue run rate. You take revenue from the latest month and multiply it by 12. Some companies use a week and multiply by 52.

Most financial analysis uses trailing 12-month revenue, or TTM. That's simply the sum of the past four quarters.

Both have strengths and weaknesses.

TTM is solid and audited. But it understates a business that's growing quickly. By the time you publish it, it's out of date.
ARR shows where a fast-growing company is today. But it flatters the business if the latest month was unusually strong.

OpenAI and Anthropic have both adopted ARR enthusiastically. You can see why.

Same company, very different numbers

Here's what the choice does to the numbers.

Anthropic: about $35 billion of trailing revenue against about $67 billion of ARR. That's roughly 1.9 times.
OpenAI: about $20 billion of trailing revenue against about $50 billion of ARR. That's roughly 2.5 times.

The gap is the growth. It's also the room for spin.

History rhymes

This isn't new. Young companies always get priced against whatever metric is available.

In the late 1990s, dot-com companies were priced per website visitor.

In the early 2000s, Mary Meeker argued that time spent on a site mattered more. Analysts started pricing "eyeballs".

In the social and SaaS era, it was users, subscribers and ARR multiples.

In the AI era, we're seeing price per ARR dollar, per gigawatt of compute and per dollar of compute committed.

Expect metrics you've never seen before. Expect them to sound outlandish until they become normal.

Two ways to put a number on a company

There are two ways to put a number on a business.

Value it. Start with the business. Forecast revenue growth, margins and reinvestment. Work out cash flows. Discount them for risk. You get an answer to the question "what is it worth?"
Price it. Start with the market. Pick a multiple, such as price to revenue. Find similar companies. Adjust for differences. You get an answer to the question "what will people pay?"

In a perfectly efficient market, the two answers match. In practice, they often don't. So before you put a number on a company, decide which game you're playing. Young companies are far more likely to be priced than valued. In a credit squeeze, that price moves with interest rates.

Part VII: Company life cycles

Businesses follow a life cycle. Companies are a bit like people.

Start-ups are babies
Young growth companies are taking off.
High-growth companies are at their peak.
Mature growth and mature stable companies are in middle age.
Declining companies are in old age.

Revenue rises, peaks and eventually falls. Margins start negative, turn positive and then fade.

Lazy shortcuts don't work

People often guess a company's stage from its age or its sector. Both are poor guides.

Age misleads. A niche tech firm can reach a steady state in three or four years. An infrastructure firm can be 25 years old and only just getting going.
Sector misleads. For 50 years, "tech" has meant "growth". But tech has mature and declining firms too. Tech companies age in dog years.

A 35- to 40-year-old tech company is, in life-cycle terms, roughly like a 200-year-old manufacturer.

Tell me your numbers, I'll tell you your stage. The better test is operating metrics.

Revenue growth: high for young companies, low and steady for mature ones, shrinking in decline.
Operating margin: negative for young companies, healthy for mature ones, falling in decline.
Reinvestment: large when young, modest when mature.
Free cash flow: very negative when young, positive when mature.
Dividends: none when young, generous when mature.
Failure risk: high when young, because the business depends on outside capital.

That last point is crucial for our story. Young companies need outside money. That's why the credit cycle matters so much to when and how they list.

Where the AI companies sit. Plot companies on growth against profitability and the picture is clear.

Anthropic, OpenAI and the neoclouds sit in "young growth": high revenue growth, big operating losses.
Nvidia and Microsoft sit in "high-growth winners": fast growth and strong profits.
Mature cash cows sit lower down, with slow growth and strong margins.

Size doesn't change the stage. Anthropic may be worth $1.5 trillion, but by this test it's still a young company. That makes it dependent on capital markets.

A quick test. Try it yourself.

Company A grows about 6% a year. It has a steady 30% operating margin. It reinvests little and pays large dividends.
Company B grows very fast. It loses money. It reinvests heavily, burns cash and needs outside capital.

Company A is mature, whatever it thinks of itself. Company B is young

Part VIII: Anthropic against OpenAI, by the numbers

Two young companies at enormous scale

Founded: Anthropic in 2021, OpenAI about 11 years ago.
2023 revenue: about $100 million for Anthropic, about $1 billion for OpenAI.
Trailing revenue in 2026: about $35 billion for Anthropic, about $20 billion for OpenAI.
Run rate in September 2026: about $67 billion for Anthropic, about $50 billion for OpenAI.
Reporting: gross for Anthropic, net for OpenAI.
Active users: about 250 million for Anthropic, about 1.2 billion for OpenAI

The overtake. Early on, OpenAI was far ahead. In 2025 its revenue was roughly three times Anthropic's. Then something shifted in 2026, and Anthropic pulled ahead on both trailing and run-rate revenue. What changed? Anthropic moved from individuals to enterprises, and from subscriptions to usage-based pricing. Claude Code supercharged that shift.

Gross versus net: not apples to apples. There's a catch in that comparison.

Imagine a customer pays $100 through a partner, and the partner keeps a share.

Gross reporting, which Anthropic uses, books the full $100 as revenue and shows the partner's cut as a cost.
Net reporting, which OpenAI switched to, books only what's left after the partner's cut.

The economics are identical. The top line isn't. So $67 billion against $50 billion probably overstates Anthropic's lead.

Losses on both sides. Both companies lose a lot of money.

2024 operating losses: about $3 billion at Anthropic and $5 billion at OpenAI.
2025 operating losses: about $8 billion at Anthropic and $39 billion at OpenAIthough some of OpenAI's may be one-off
First half of 2026: Anthropic lost about $8.7 billion from operations. Its net loss was about $80.7 billion, mostly that non-cash charge

"Profitable on an adjusted basis". At the mid-October meeting, Anthropic claimed it had made money in the third quarter on an adjusted basis.

Handle this with care. Adjusted profit strips out items management considers one-offs. The speaker in the session was blunt: adjusted numbers are often fiction. The first half of 2026 still showed big operating losses. Wait for the public prospectus.

The verdict is simple. Both companies have high revenue growth and big operating losses. By the life-cycle test, they're young companies, whatever their size.

Part IX: Why valuation is hard

The three crutches

Analysts lean on three crutches when they value a company.

Past financials. Five or ten years of growth and margins tell you how a business works. Anthropic has two or three years of messy statements.
Market size and competition. You need to know how big the market is and what share you can win. The AI market is still forming. Bankers float total markets of $20 trillion to $30 trillion, which tells you how uncertain it is.
Industry averages. Sector margins and costs of capital give you benchmarks. Anthropic has no comparable public peers.

Take away all three crutches and most people give up.

Six uncertainties. Here's what makes valuing an AI lab so hard.

Current financials: short history, with revenue growing and everything else a mess.
Market size: still up for grabs.
Competitive advantage: nobody knows what will let the winners earn more than their cost of capital.
Risk: there's no price history, and the company is now exposed to the credit cycle.
Path to maturity: more than $600 billion of commitments means failure risk could stop it ever reaching a steady state.
Governance: a founders' share holds 50.1% of key votes.

Where the uncertainty sits. The prospectus will help with some of this. It will clarify revenue definitions, losses, commitments and share classes.

It won't settle the biggest questions. How big will the market be? How long will Anthropic's edge last? What will its margins look like in 2035? What should its cost of capital be when yields are above 5%?

Those remain a matter of story.

The temptation. When valuation gets hard, people look for someone smarter to copy. In this case, that means venture capitalists.

The chain goes like this: too much uncertainty, no crutches, "I won't value it", copy what the VCs paid, and you end up with a price rather than a value.

VCs are smart and well informed. But their numbers are forward guesses about market price, not estimates of value. Copying them just imports their guess.

Part X: Venture capital pricing

Many rounds, huge sums

Anthropic has had 19 venture rounds and raised about $135 billion. Its latest round, in May 2026, priced it at about $965 billion.
OpenAI has had 10 rounds and raised about $185 billion. Its latest round, in March 2026, priced it at about $852 billion. A pending round is rumoured at about $1.4 trillion.
IPO talk for Anthropic is $1.5 trillion to $2 trillion.

VC "valuation" is forward pricing in disguise. Here's how a VC typically arrives at a number.

Pick an exit year, usually three to five years out. The choice is arbitrary.
Forecast a metric for that year, such as revenue.
Apply a multiple taken from public companies.
That gives you an exit price.
Discount it back to today at a "target rate" of 30% to 70%.

That last number isn't a measure of risk. It's a bargaining chip. VCs have every incentive to push it up and push the price down.

Here's the session's example. Say you expect $300 billion of revenue in five years. The big tech firms trade at about five to seven times revenue. Five times $300 billion is $1.5 trillion. Discount that at a made-up rate and you have today's round price.

Reverse-engineering a $965 billion round

In May 2026, the expected IPO price for Anthropic was about $1.15 trillion. VCs priced the round about $150 billion lower to allow for the wait and the risk of delay.

Why $150 billion? Why not? The round price is a guess at a future market price minus a haircut. It tells you nothing about intrinsic value.

Firmus shows the flip side. It sought nearly three times its last private price, and public investors refused.

Pre-IPO rounds tell you even less. Normal VC rounds project five to seven years ahead. Pre-IPO rounds project only to the IPO itself, and guess what the market will pay next year.

Think of a two-by-two with time to exit on one axis and how much a round tells you about intrinsic value on the other.

Fundamentals-led exits: rare. These are near-term exits priced on fundamentals, such as a strategic sale.
Early-stage VC: forward pricing of a distant metric.
Late-stage growth: a mix of pricing and fundamentals.
Pre-IPO VC: guessing next year's IPO price. That's where Anthropic's May round and OpenAI's pending round sit.

The takeaway is simple. We give too much weight to what VCs think.

Part XI: The mechanics of pricing

Three steps. Pricing has three steps.

Choose a multiple: a market value divided by something that scales, such as revenue.
Find a peer group: the companies investors will compare yours with.
Compare and adjust: control for differences in growth, risk and margins.

Here's an example. Your company trades at 15 times earnings. Its peers trade at 20. Before you call it cheap, ask why. Lower growth or higher risk might fully justify the gap.

Building a multiple

You can't compare share prices directly, because a share split changes the price. So you divide market value by a common scalar.

For the top of the fraction, use market capitalisation, enterprise value or, for private firms, the latest VC price.

For the bottom, choose from:

revenue or ARR;
earnings, such as net income or operating income;
cash flow, such as EBITDA or free cash flow;
book value;
for pre-revenue firms, users, subscribers or total market size.

Which year do you use?

There's no right answer. You can use trailing revenue (about $35 billion), the September run rate (about $65 billion), the year-end run rate (about $110 billion) or 2030 revenue (about $300 billion). Each gives a very different multiple.

The further forward you go, the lower the multiple looks. And the more of the bull case you've quietly assumed.

Multiples change with age

Start-ups get priced on users, subscribers or market size.
Young growth companies get priced on revenue or ARR multiples.
High-growth companies get a mix of revenue and some earnings multiples.
Mature growth companies get forward price-earnings or PEG ratios.
Mature stable companies, like Microsoft, can use almost any multiple.
Declining companies get priced on book value.

Below revenue and gross profit, everything is negativ

For Anthropic, revenue and gross profit are positive. EBITDA, operating profit and net income are all negative. A multiple needs a positive number at the bottom. So you're pushed up the income statement until you hit revenue.

Expect enterprise value to revenue, and enterprise value to ARR, to dominate the debate.

Part XII: AI-specific pricing metrics

At $1.5 trillion, about 23 times run rate

Let's do some sums.

$1.5 trillion divided by about $65 billion of ARR is about 23 times.
At $2 trillion, it's about 31 times.
OpenAI at a similar $1.5 trillion on about $50 billion of ARR would be about 30 times.

On that basis, Anthropic looks cheaper than OpenAI. But remember gross versus net. Sort that out before you compare.

The forward-number trick

Watch how the multiple shrinks as you look further ahead.

At $1.5 trillion: about 23 times September ARR, about 13.6 times year-end ARR and about 5 times 2030 revenue.
At $2 trillion: about 31 times, 18 times and 6.7 times.

The session predicts that Anthropic's bulls will reach for "really, really forward numbers". It makes the multiple look reasonable.

But a 2030 multiple quietly assumes the 2030 revenue arrives. And with yields above 5%, revenue that far out is worth less today than it was a year ago.

The further forward the revenue, the cheaper it looks

Compute as a metric. Compute is scarce. So compute itself can be sold as an asset.

According to the session, OpenAI spent about $15 billion on compute in 2026 and Anthropic about $6 billion. By 2028, those figures could reach about $111 billion and $28 billion. Anthropic's total commitments now exceed $600 billion, about 80% of them non-cancellable.

Compute is scarce - so compute can be marketed as an asset

Turning a liability into a selling point. Normally, big commitments are a worry. In pricing, a lab can spin them.

The bull framing: "We have locked in more than $600 billion of scarce compute. Our market value is only about 2.5 to 3.6 times that. A rival trades at six times. We're cheap."
The bear framing: "Those are fixed obligations for a loss-making company. That's financial risk, not an asset."

This month's neocloud selloff shows the market leaning towards the bear view.

Users as a metric. By users, OpenAI looks like the bargain.

OpenAI has about 1.2 billion active users. At $1.5 trillion, that's about $1,250 per user.
Anthropic has about 250 million. At $1.5 trillion, that's about $6,000 per user.

So why don't Anthropic's critics shout about this? Because until 2024, users were a decent proxy for revenue. Anthropic broke that link by moving to enterprises and usage pricing. Its "user" is often a business spending tens of thousands of dollars.

Why users stopped predicting revenue. Picture a two-by-two. One axis runs from consumer to enterprise. The other runs from subscription to usage pricing.

Consumer subscriptions: chat apps on monthly plans. Users roughly equal revenue. That's OpenAI's core.
Enterprise seats: per-seat business plans, like Microsoft 365 Copilot.
Enterprise usage: APIs and agents billed per token. Revenue scales with the work done. That's Anthropic in 2026.
Consumer usage: pay-as-you-go individuals.

When revenue comes from enterprise usage, revenue per user varies wildly. User-based multiples stop working.

Pick your poison. Each side will push the metric that flatters it.

Price to current ARR: Anthropic looks cheaper, partly thanks to gross reporting.
Price to 2030 revenue: Anthropic looks cheaper, because of its steeper growth.
Price per user: OpenAI looks cheaper.
Price per unit of compute capacity: it depends. OpenAI had more early; Anthropic is catching up.
Price per dollar of compute committed: Anthropic looks cheaper.

Bias drives pricing just as it drives valuation. It shows up in the metric you choose.

Part XIII: Peer groups

Easy, messy and hard

Peer groups come in three flavours.

Easy: dozens of public peers with live prices and common accounting.
Messy: many private peers with stale VC prices.
Hard: one or two private peers. That's Anthropic.

OpenAI is private and waiting until 2027. SpaceX is public but has rockets and satellites muddying the picture. There's no clean comparison.

The Magnificent Seven (Mag 7) as yardstick

A peer group isn't really "companies just like yours". It's the companies your investors will compare you with.

At $1.5 trillion to $2 trillion, that means the Magnificent Seven: Alphabet, Apple, Amazon, Meta, Microsoft, Tesla and Nvidia.

There's a twist. Three of them, Amazon, Alphabet and Microsoft, are Anthropic's cloud suppliers or investors. And Meta fell 7% this week on AI spending worries. The peer group is also the supply chain.

Against the Magnificent Seven. Together, the Magnificent Seven trade at about eight to nine times revenue.

At $1.5 trillion, Anthropic trades at:

about 48 times trailing revenue, as stated in the session;
about 15 times 2026 revenue;
about 5 times projected 2030 revenue.

Use current revenue and Anthropic looks five or six times as expensive. Use 2030 and it looks cheap. As the session puts it, you walk in with a bias and find a metric to back it up.

Part XIV: From metrics to value

Two simple tests. Before you trust a metric, run two tests.

Does it keep your sample? If a metric drops half your peers, don't use it. Price-earnings ratios fail for loss-makers. For AI labs, revenue or ARR survives.
Can you connect it to value? Work out what drives the multiple in a proper valuation. Those drivers are what you must control for.

There's no perfect metric. There's the metric you use, and the discipline to control for what drives it.

What drives a price-earnings ratio

Three things drive a P/E ratio.

Higher expected earnings growth raises it.
Higher return on equity raises it, because you need less reinvestment for the same growth.
Higher risk, measured as cost of equity, lowers it.

That third point matters right now. Rising yields push up everyone's cost of equity. They lower every multiple.

Nobody will use a P/E for Anthropic, because it has no earnings. But the same logic applies to revenue multiples.

What drives a revenue multiple. Four things drive enterprise value to revenue.

Revenue growth. Can a run rate of about $67 billion become hundreds of billions quickly? A 45% quarter helps
Operating margins. Can high gross margins and scale lift operating margins to 30% to 40%?
Reinvestment efficiency. Can revenue grow faster than compute spending, which was about $9 billion in the first half alone?
Risk. Will regulatory, concentration and commitment risks fade, while yields stay above 5%?

Pricing hides the same assumptions

When you price a company, you make exactly the same assumptions as when you value it. You just make them implicitly.

Avoiding a valuation doesn't avoid the assumptions. It hides them. The honest move is to make them explicit.

What must be true for $1.5 trillion to $2 trillion

For the bull case to hold, four things need to happen.

Revenue growth: a run rate of about $67 billion must become hundreds of billions, fast.
Profitability: operating margins must climb to 30% to 40%.
Unit economics and scale: compute costs must grow more slowly than revenue, despite $600 billion-plus of commitments.
Risk: regulatory, concentration and financing risks must converge towards normal.

Is that plausible? Yes. Is it easy? No. And a higher cost of capital raises the bar.

Bulls, bears and the elephant

The bulls will tell you:

AI is a vast market.
Anthropic is one of the big winners.
Gross margins are high and enterprise usage is sticky.
Forward multiples look reasonable.

The bears will tell you:

Regulation, reliability or credit could slow the market.
The founders control the company through a special share.
The revenue multiple can't be earned back.
The commitments are too big to cover.

Both are right about something. It's the old fable of the blind men and the elephant. Each touches one part and describes a different animal. One feels the market size. Another feels the growth rate. A third feels the loss. A fourth feels the commitments.

Your job is to step back and see the whole animal.

Act 4: The verdict

Part XV: Is this the right time?

The whole story in one chain

Let's pull the threads together.

AI needs about $1 trillion of capital a year.
Credit has become the bottleneck.
Yields above 5% squeeze the companies that borrow to build.
Equity, including IPOs, must fill the gap.
Anthropic's IPO will be priced, not valued.
That price will help set the credit terms for the whole chain.

The loop closes neatly. Anthropic's IPO price will be set by the same credit conditions that its $600 billion-plus of commitments depend on. A strong deal strengthens the lenders to neoclouds and data centres. A weak one tightens the chain further, just as Firmus's pulled IPO did in Australia.

My timing scorecard

I have scored eight factors. Three support going now. Five argue for waiting.

Supporting a listing now:

Company momentum: Q3 revenue up 45% on the quarter, with a run rate of about $67 billion.
Need for capital: more than $600 billion of commitments against about $20 billion of year-end cash.
Competition for capital: it's the only big AI IPO, and OpenAI is waiting.

Arguing for waiting:

Interest rates: the 10-year yield sits around 5%, with a December hike widely expected.
Equity markets: two Nasdaq selloffs, and chips more than 26% off their peak.
The credit chain: neoclouds under stress, and Firmus pulled.
Company-specific risk: customer concentration, the defence dispute and founder control.
Macro and politics: the midterms on 3 November, Iran and oil above $110.

My net read? This isn't an ideal window. But Anthropic is the one issuer whose need and demand are strong enough to go anyway. I'd expect it to be priced to sell, not at the top of the range.

Who wins and who waits. Here's how each player fares under the two main outcomes.

If the IPO succeeds:

Anthropic funds its commitments and ends the convertible-note charges.
Microsoft gets a public price on its stake, and its $30 billion Azure deal looks safer.
OpenAI gets a public comparison for its own 2027 listing.
Nvidia sees a stronger demand chain and needs to offer fewer backstops.
Neoclouds and developers find their contracts become better collateral.
Investors get rare exposure to a frontier AI lab.

If it's delayed or weak:

Anthropic leans on private money and partners.
Microsoft's hedge stays intact through OpenAI and its own models.
OpenAI looks wise for waiting.
Nvidia offers more backstops, Firmus-style.
Neoclouds face tighter credit for longer.
Investors may get a lower entry price later.

Six signals to watch over the next four weeks

The public prospectus. Look at the final figures, gross versus net revenue, share classes and use of proceeds.
Big-tech earnings. Microsoft, Alphabet and Amazon report next week. Watch their capex and cloud demand.
The 10-year yield. A move back above about 5.25% would hurt the roadshow.
Neocloud credit. CoreWeave's bonds and credit default swaps are the best stress gauge.
Firmus and Nvidia. Does vendor money replace public money?
The midterms and oil. The vote on 3 November, Iran risk and Brent above $110.

Conclusion: the chips are on order

The chips are on order. The question is who pays.

AI's bottleneck has moved from chips to credit, and October made it visible. Yields pushed past 5%. The Nasdaq sold off twice in a week. Neoclouds lost a fifth of their value in five days. Firmus pulled its IPO, and Nvidia stepped in again.

Into that market, Anthropic is bringing the only big AI IPO of the season. It brings extraordinary growth, more than $600 billion of commitments and a shareholder, Microsoft, that has also backed its fiercest rival.

It will be priced, not valued. That price will reflect the credit cycle as much as it reflects Claude.

Sources and notes:

This piece draws on reporting from CNBC, Reuters, Yahoo Finance, the Wall Street Journal (as reported by others), TechCrunch, Portfolio Terminal, ABC News, Tech Insider and others, published between September and October 2026. It also draws on Microsoft's 2026 financial results, SemiAnalysis's work on AI financing, and a valuation session on the Anthropic IPO published on YouTube in October 2026.

Some important caveats:

Anthropic's financial figures come from press reports of a draft prospectus. The public filing may differ.
Figures attributed to the valuation session are as the speaker stated them. Some came from an auto-generated transcript.
The two-by-two placements and the timing scorecard are my own judgements.

The AI Bottleneck Shift: From Chips to Credit

2026-10-17T08:08:58.390+11:00

Introduction: The Question Underneath Every GPU

For most of the past three years, the public conversation about artificial intelligence (AI) has been a conversation about chips. Who has the GPUs? How many did a lab buy? When will the next generation ship? Then the conversation moved to power: gigawatt campuses, gas turbines parked behind datacenter fences, nuclear restarts and grid queues measured in years. Both conversations reflected real bottlenecks at the time.

But underneath every GPU order and every power contract sits a quieter question: who is paying for all of this, and how?

A recent podcast of the SemiAnalysis Weekly, titled "Nvidia's Backstop Universe," puts that question at the centre of the AI story. Its argument is simple to state and profound in its implications.

The AI build-out is no longer primarily constrained by chips or by power sites. It is constrained by credit, by the willingness and ability of capital markets to lend enormous sums against assets that are new, fast-depreciating, technically complex and often owned by companies that do not have investment-grade credit ratings.

The way the industry is bridging that gap is fascinating. Nvidia and the hyperscalers (HPS - Microsoft, Amazon, Google, Meta and Oracle) are effectively renting out their balance sheets. They put their creditworthiness between lenders and the riskier companies that build and operate much of the new capacity. The episode's phrase for it is "Balance Sheet as a Service."

This post analyses SemiAnalysis weekly podcast and explores that argument from first principles and is built in six parts, each answering one question:

Part I - Why is AI demand changing shape (training → inference → agents)? How did we get here from Cloud 1.0? What does the AI stack look like, and what describes AI as an industry
Part II - Neoclouds. Who are these new GPU clouds, how do they differ from each other, how do they fund themselves, and what does the Nvidia–CoreWeave relationship tell us?
Part III - The financing challenge. How big is the bill, why financing is now the bottleneck, how project finance works, what a backstop actually is, and what is inside Nvidia's roughly $497 billion "backstop universe."
Part IV - The economics. A simple financial model of a GPU cluster, the financial levers that move returns, profitability per megawatt, residual values, and the shift from GPUs to custom ASICs.
Part V - Markets and risk. How credit markets are reacting, why rising bond yields worry equity investors, the willingness-versus-ability debate, where durable agentic demand comes from, and what could go wrong.
Part VI - Bringing it together. An end-to-end walk-through of financing a campus, what it all means for sellers, buyers and builders, a glossary, and references for further reading.

Ten key takeaways

The real AI bill is about $11.3 trillion of capex from 2024 to 2029, with roughly $7 trillion of debt outstanding by 2029, per the episode.
The binding constraint has moved from chips, to power and sites, to financing.
Demand is changing shape, from one-off training, to real-time inference, to always-on agentic AI, which makes it more continuous and more financeable.
Lenders want long-term offtake from investment-grade counterparties; most AI labs and Neoclouds aren't investment grade.
Nvidia and the hyperscalers bridge that gap by lending their balance sheets , "Balance Sheet as a Service", with Nvidia's commitments estimated at about $497 billion.
Neoclouds are not all alike: platform-led, full-stack, access-layer and infrastructure led players carry very different risks.
In a simple cluster model, price, utilisation, capex and leverage dominate returns; a backstop can roughly triple debt capacity.
The industry's long-term structure is unsettled: parts may look like software, parts like telecom.
Custom ASICs, rising bond yields and customer concentration are the main pressures on the model.
Nvidia's backstop capacity (~46 GW) is far below the ~240 GW needed, so credit markets must learn to underwrite AI on its own merits.

Part I - Why is AI demand changing shape: Three Eras of AI

Before talking about money, it helps to understand why the amount of compute the world wants keeps growing, and why its shape matters to lenders. A useful framing, popularised in industry keynotes, divides the AI revolution into three eras:

Three eras of AI

Era 1: Training - "One Time"

Training is the process of building a model: feeding enormous amounts of data through a neural network for weeks or months until it learns useful patterns. Training is compute-hungry but episodic. A lab assembles a very large cluster, runs a training job, then moves on to the next model.

From a financing perspective, training demand is lumpy. It comes in big blocks tied to specific projects. If a lab decides to delay its next model, or finds a more efficient way to train, demand for a particular cluster can disappear quickly. That makes training clusters harder to finance on their own merits, lenders worry about what happens between training runs.

Example: A lab rents 20,000 GPUs for four months to train a new model. Once training finishes, it needs only a fraction of those GPUs for experiments. If the neocloud has no other customer lined up, those GPUs sit idle. A lender to that neocloud sees a revenue stream with a cliff at month four.

Era 2: Inference - "Real Time"

Inference is using the trained model: every time someone asks a chatbot a question, a coding assistant suggests a line, or an enterprise app summarises a document, the model runs on GPUs and generates tokens. Inference demand follows users. It happens in real time, all day, across time zones, and grows as adoption grows.

For lenders, inference is more attractive than training because it looks more like a utility: a steady flow of usage that converts directly into revenue per token. The more inference there is, the more predictable demand becomes.

Example: An enterprise rolls out an AI assistant to 10,000 staff. Each person makes about 40 requests a day. That's 400,000 requests a day, every working day, generating a predictable baseline of compute demand, and of revenue for whoever serves it.

Era 3: Agentic AI - "All Time"

Agentic AI refers to systems that don't just answer a question but carry out multi-step tasks: researching, planning, calling tools, writing and testing code, filling in forms, checking their own work, and handing off to other agents. A single human request can trigger dozens or hundreds of model calls. Agents can also run unattended, overnight, continuously, monitoring systems or processing queues.

That is why the era is labelled "all time." Demand stops being tied to the number of humans typing and becomes tied to the number of tasks delegated to software. It is potentially far larger, and far more continuous, than human-driven inference.

Example: A claims-processing team deploys an agent that reads each incoming claim, checks policy rules, requests missing documents, drafts a decision and routes exceptions to a human. Each claim might involve 50 model calls. At 5,000 claims a day, that's 250,000 calls, running around the clock, regardless of office hours.

Why the eras matter for financing: The three eras matter to this story for three reasons:

Demand durability: Lenders want long, predictable revenue. Inference and agentic workloads are more continuous than training, which makes AI capacity look more financeable over time.
Hardware mix: Training rewards the most flexible, most powerful chips (mostly Nvidia GPUs). High-volume, stable inference rewards the cheapest cost per token, which is where custom chips (ASICs) compete strongly. That shift affects the long-term value of GPU fleets that are being financed today.
Revenue linkage: In the training era, compute is a cost centre funded by investors. In the inference and agentic eras, compute is directly tied to revenue, that is every token served is (in principle) a token sold. That linkage is what ultimately repays the debt.

In the training era, compute is a cost centre funded by investors. In the inference and agentic eras, compute is directly tied to revenue, that is every token served is (in principle) a token sold. That linkage is what ultimately repays the debt.

From Cloud 1.0 to the AI Cloud

AI infrastructure did not appear from nowhere. It is the latest stage in a long evolution of cloud computing, and each stage changed why customers buy and how providers finance themselves. Understanding that evolution makes it easier to see why Neoclouds exist and why the financing model looks the way it does.

Cloud evolution: each era had a different primary sales trigger

Cloud 1.0 (2008-2013): public cloud, OPEX reduction - The first wave of cloud adoption was mostly public cloud, centred on specific business functions like email, CRM, collaboration, test and development environments. The main sales trigger was cost: turning capital expenditure on servers into operating expenditure, and reducing the cost of running IT. The focus was fulfilment: getting existing workloads delivered more cheaply.

Financially, this era established the hyperscaler model. A few very large companies invested heavily in datacentres and recovered that investment through millions of small, recurring customer payments. Their own balance sheets funded the build.

Cloud 2.0 (2014–2019): public and private, scale and agility - The second wave brought both public and private cloud into mainstream business and operational functions. The trigger shifted from cost to scale and agility, launching products faster, scaling up and down with demand, and protecting revenue by keeping systems available. It was both fulfilment-centric and demand-centric.

This is the era in which hyperscalers became some of the largest capital spenders in the world, still largely self-funded from strong cash flows.

Cloud 3.0 (2020 onwards): distributed and hybrid - The third wave, accelerated by COVID, made cloud the platform of choice across all functions. Architectures became distributed and hybrid, mixing public cloud, private infrastructure and edge locations. The triggers moved again: resilience, customer experience, and revenue protection and generation. Cloud became demand-centric: not just a cheaper place to run things, but the place where new revenue is created.

The distributed cloud battle: control versus portability: A key strategic tension in Cloud 3.0 is between deep integration with one provider and portability across many.

Control vs Portability

Walled garden ("Apple of cloud") - high control, low multicloud reach. AWS and Azure are placed here: deep integration with the provider's own capabilities, consistency, and cloud-service-provider-native design. The customer gets a rich, coherent platform but is more tied to it.
Portability ("Android of cloud") - lower control, high multicloud reach. Google and IBM/Red Hat are placed here: deep integration with a multicloud stack, a portability layer, flexibility and container-native design.
"Paradise" - high control and high portability. Everyone wants it; few achieve it.
Legacy - limited control and limited portability; the starting point for many organisations.

Cloud-native and multicloud architectures pull in different directions and have a heavy impact on exit strategy. Both have a place in a distributed cloud, but a cloud strategy must avoid contradicting itself.

Why this matters for AI financing - The same tension now plays out in AI compute, and it has financing consequences:

Hyperscaler AI capacity is the walled garden: It is integrated with the provider's models, data services and security, and is funded from investment-grade (IG, good credit rating) balance sheets.
Neocloud capacity is closer to the portability end: Customers can rent raw GPU capacity and run whatever models and software they like. That flexibility attracts AI labs and developers, but Neoclouds don't have hyperscaler balance sheets, so they need external financing and often backstops.
Customers' exit strategy becomes lenders' risk: If customers can move workloads easily between providers, a Neocloud's revenue is less sticky, which is exactly why lenders insist on long-term, take-or-pay contracts.

Example: An AI lab runs inference on a Neocloud using containerised, open tooling. When a cheaper provider offers capacity, the lab could move within weeks, unless it is locked into a five-year reserved contract. For the Neocloud's lender, the portability that attracted the customer is precisely why the contract matters more than the relationship.

The AI cloud: a fourth wave - Extending the above framework, the AI cloud can be seen as a fourth wave. Its sales trigger is capability: access to frontier models, agents and productivity gains that weren't possible before. Its financing model is different from earlier waves because the capital required is so large and so concentrated in fast-depreciating hardware that even the hyperscalers' cash flows aren't enough on their own.

The AI Stack - Who Does What

To talk about who is backstopping whom, we need a shared map. The AI industry can be thought of as a layered stack. Capital flows up the stack to build things; revenue flows down the stack to pay that capital back. Every financing structure in this post is ultimately an attempt to make those two flows line up in timing, size and certainty.

The AI stack: capital flows up to build capacity, revenue flows down to repay it.

Layer 1: Capital providers - At the base are the people with money: equity investors (venture capital, growth funds, sovereign wealth funds, public shareholders), banks, private credit funds, insurance companies and pension funds that buy bonds, and the asset-backed securities (ABS) market, which packages loans or leases into tradable securities. Each has a different appetite.

Equity investors accept the risk of losing everything in exchange for unlimited upside. Senior lenders want to be repaid first, with interest, and care far more about the downside than the upside. Insurance and pension money typically wants long-dated, highly rated, predictable cash flows. This mismatch of appetites is the key to the whole backstop phenomenon.

Layer 2: Datacentres and power - Next is physical infrastructure: land, buildings ("shells"), cooling, electrical gear, substations, grid connections and power. It is owned by hyperscalers, specialist developers, infrastructure funds and, increasingly, energy companies. Power is procured through power purchase agreements (PPAs), where a buyer commits to purchase electricity at an agreed price for many years. Some AI campuses use "behind-the-meter" generation like on-site gas turbines (ex xAI), fuel cells or other sources that don't depend on a slow grid connection.

Datacenter infrastructure is long-lived: a building and its electrical systems can last 15 to 20 years or more. That makes it relatively friendly to traditional infrastructure finance.

Layer 3: Compute - On top of the datacenter sits compute: Nvidia GPUs (H100, H200, B200, GB200, GB300 and successors), alternative accelerators often called XPUs or ASICs (custom chips designed by or for hyperscalers), Google's TPUs, plus servers, high-bandwidth memory, networking switches, optics and cables.

Unlike the building, compute is short-lived in economic terms. A new generation arrives every year or two and can be several times more capable. That makes compute much harder to finance with long-term debt: lenders worry about what the hardware will be worth in five or six years.

Layer 4: Clouds - hyperscalers and neoclouds - The cloud layer turns physical compute into a rentable service. Hyperscalers are the giant, diversified providers like Microsoft Azure, AWS, Google Cloud, Oracle, and Meta (which mostly builds for itself). Most have strong investment-grade credit ratings and enormous cash flows from other businesses.

Neoclouds are newer, AI-focused providers that specialise in GPU capacity like CoreWeave, Nebius, Lambda, Crusoe, Fluidstack, Nscale and others. They are fast and technically focused but younger, more leveraged and usually not investment grade.

Layer 5: AI labs and model providers - AI labs train frontier models and sell access through consumer products and APIs. They are enormous consumers of compute for training and, increasingly, inference. Many are growing revenue very quickly but are not consistently profitable on a full-company basis and are not investment-grade borrowers.

Layer 6: Users and enterprises - At the top are the people and organisations who pay: consumers on subscriptions, developers paying per token, and enterprises buying AI features in software they already use. This is where the money that ultimately repays every loan in the stack originates.

Worked example: Following one dollar through the stack

Following one dollar through the stack

Example. An Australian health software company embeds an AI assistant in its case management product and pays an AI lab $1 for a batch of API tokens. The lab pays a cloud provider, say a Neocloud, to host the model on GPUs. The Neocloud uses the rental income to pay interest and principal on the loan that bought the GPUs, rent to its datacenter landlord, and its power bill. The landlord services its construction loan and pays its investors. Each layer takes a margin, and each layer's lender relies on the layer above to keep paying.

That chain is the whole story in miniature. If the dollar at the top keeps flowing, reliably, for years, everyone below gets repaid. If lenders doubt the dollar will keep flowing, or doubt the creditworthiness of the company promising to pay it, they either refuse to lend or demand much higher interest. Backstops exist to calm those doubts.

What Kind of Industry Is AI? Lessons From Telecom

Every large infrastructure wave eventually settles into an industry structure: who earns the margins, who carries the capital, and how volatile returns are. Telecom is the comparison most often reached for, so it is worth being precise about what telecom became, and why AI might, or might not, follow the same path.

Telecom: an infrastructure-driven, high-volume, low-margin industry - Telecom is the archetype of a capital-heavy, infrastructure-driven business. Operators spend vast sums on fibre, towers, spectrum, switching and undersea cables, then recover that investment by selling enormous volumes of a fairly standardised product like minutes, messages, megabytes at low unit prices. Over time, three forces push telecom towards high volume and low margin:

Commoditisation: A gigabyte of data from one carrier is much the same as a gigabyte from another. Once networks are built, price becomes the main point of competition.
High fixed costs, low marginal costs: Most of the cost is incurred building the network. Carrying one more gigabyte costs almost nothing, so carriers are tempted to cut prices to fill capacity, which pushes prices down for everyone.
Regulation and utility-like expectations: Communication is treated as essential infrastructure, which limits pricing power and often caps returns.

The result is an industry with steady but modest returns, heavy debt, and limited ability to capture the value created on top of its networks. The value of the internet largely went to the companies that built services on the pipes, like search (Google), social media (Facebook), e-commerce (AMZN), cloud software (Salesforce), rather than to the companies that laid the pipes.

The telecom boom and bust: demand was real, financing was mistimed - The late-1990s telecom build-out is the cautionary tale. Carriers borrowed heavily to lay fibre on the expectation of explosive internet traffic. Traffic did grow explosively, but not quickly enough to repay the debt on the original timeline, and vendor financing amplified the problem: equipment makers lent money to their own customers so they could keep buying equipment. When the cycle turned, many carriers failed, equipment makers suffered large losses on the loans they had extended, and huge amounts of fibre went unused ("dark fibre") for years.

Two lessons carry over directly to AI.

First, demand can be real and the financing can still be wrong, timing matters as much as direction.

Second, when suppliers finance their own customers, their fortunes become tightly linked; a slowdown hits the supplier twice (twin balance sheet impact), once through lower sales and again through losses on support it has extended.

This pattern is worth keeping in mind when reading about Nvidia's backstops.

Where AI is different: There are important differences between AI and telecom:

Asset life: Fibre lasts decades; a GPU's frontier economic life is a few years. Dark fibre could wait years for demand to catch up. Idle GPUs lose value much faster.
Funders: Much of the AI build-out is being funded or supported by some of the strongest balance sheets in the world, not just highly leveraged carriers.
Revenue today: Unlike the early internet, AI is already producing substantial revenue at the top of the stack, with some frontier-token gross margins approaching 90% in their models.
Product differentiation: A frontier model is not (yet) a commodity in the way a gigabyte is. Quality differences between models still command price differences.

Telecom vs AI

A frontier model is not (yet) a commodity in the way a gigabyte is. Quality differences between models still command price differences.

So what type of industry will AI become? Honestly, this is not settled yet, and it may not be one type at all. Different layers of the AI stack may end up looking like different industries.

Capital outlay vs Margins

Asset-light, high margin (software-like): This is where model IP and AI applications would sit if differentiation persists. The bull case for AI labs is that tokens behave like software: high gross margins, value captured in IP and brand.
Capital-heavy, high margin (rare): Very few industries sustain both. It usually requires a moat like scarce technology, a monopoly position or extreme scale. Nvidia currently sits here, earning high margins on scarce chips. Hyperscalers also sit here, benefiting from scale and integration.
Capital-heavy, low margin (telecom-like): This is the risk for pure GPU rental. If compute becomes a commodity, a GPU-hour from one provider is much like a GPU-hour from another, then Neoclouds start to look like telecom carriers: big assets, heavy debt, price competition, thin margins.
Asset-light, low margin. Resellers, integrators and server OEMs (which the episode notes are the exception to expanding profitability) often end up here, squeezed between powerful suppliers and powerful customers.

Three possible futures for the AI industry structure

The "software" future: Models stay differentiated, token margins stay high, and value accrues to labs and applications. Infrastructure earns decent returns but not extraordinary ones.
The "telecom" future: Models converge in quality, open models close the gap, price competition drives token prices down, and capacity is overbuilt. Infrastructure becomes a high-volume, low-margin utility; heavily leveraged players struggle.
The "split" future: Different layers settle differently: chips and frontier models keep high margins for a time, while raw GPU rental becomes commodity-like, and agentic applications capture value in specific industries. This seems the most likely, and it is precisely why lenders care so much about which layer they are lending to and who stands behind the contract.

The "split" future: Different layers settle differently: chips and frontier models keep high margins for a time, while raw GPU rental becomes commodity-like, and agentic applications capture value in specific industries. This seems the most likely, and it is precisely why lenders care so much about which layer they are lending to and who stands behind the contract.

Example: Two Neoclouds each own 10,000 GPUs. Neocloud A rents them by the hour on spot terms to whoever pays most this month, a telecom-like model exposed to price wars. Neocloud B has signed five-year take-or-pay contracts with an investment-grade hyperscaler and offers managed inference software on top. Same hardware, very different industries, and very different borrowing costs.

Why this matters for financing: Lenders price risk based on what they think the industry will become. If they believe AI compute is heading for the telecom future, they will demand short loans, big equity cushions and strong guarantees. If they believe in the software future, they will lend more freely. Backstops from Nvidia and hyperscalers essentially say to lenders: "Whatever this industry becomes, we will stand behind the downside." The question for the rest of this article is how far that promise can stretch.

Part II - Neoclouds

What is a Neocloud? A Neocloud is a cloud provider built specifically around AI compute, mainly GPUs, and increasingly other accelerators. Unlike a hyperscaler, a neocloud doesn't offer hundreds of general-purpose services. Its core product is fast access to large, well-networked AI clusters, plus varying amounts of software to make them easy to use.

The term covers a wide range of companies, from those that own power plants and datacentres to those that mostly provide a software layer over capacity rented from others.

What they share is focus: they exist because AI demand grew faster than the hyperscalers could supply it, and because AI labs wanted capacity on terms that hyperscalers weren't always willing to offer.

Why Neoclouds exist

Why Neocclouds

Speed. During the GPU shortage, Neoclouds were often able to stand up capacity faster than hyperscalers, which had to balance AI against their broader businesses.
Specialisation. AI clusters have unusual requirements, very high-density power and liquid cooling, high-bandwidth networking between thousands of GPUs, and specialised scheduling software. A company that does only this can optimise for it.
Allocation. Nvidia has an interest in a diverse customer base rather than depending on a handful of hyperscalers that are also building competing chips. Supporting Neoclouds broadens its market.
Flexibility for labs. AI labs often want raw capacity, custom configurations and less lock-in than a hyperscaler's integrated platform. Neoclouds sit closer to the "portability".
Hyperscaler overflow. Hyperscalers themselves sometimes contract capacity from Neoclouds to meet their own customers' demand without building everything themselves.

How Neoclouds make money

At its simplest, a neocloud buys GPUs (usually with borrowed money), installs them in datacenters (owned or leased), and rents them out.

Revenue comes in three broad forms:

Reserved/contracted capacity: a customer commits to a set number of GPUs for one to five years, usually paying whether or not it uses them (take-or-pay). This is the revenue lenders like most.
On-demand capacity: customers pay by the hour for what they use. Higher price per hour, but unpredictable.
Managed services: inference APIs, training platforms, storage and developer tools layered on top of raw compute. These can raise margins and stickiness.

The Neocloud Map: platform depth versus physical ownership

Not all neoclouds are alike. The following "Neocloud Map" plots them along two dimensions, with a key message: physical ownership is not the same as platform depth

Neocloud Map

The two axes are:

Horizontal - physical stack integration: At the left, the provider is partner-heavy: it leases space or relies on third-party infrastructure. At the right, it owns or controls power, datacentres and physical infrastructure.
Vertical - compute, cloud and platform integration: At the bottom, the offering is core infrastructure: GPU access, clustering and basic orchestration. At the top, it is a full cloud stack: inference, training, storage, networking, developer platform and managed services.

That produces four archetypes:

Platform-led (top-left): more software, partner-heavy on the physical side. Examples on the map: Fireworks AI, Together AI, Runpod. They compete on developer experience, inference performance and ease of use, often running on capacity owned by others.
Full-stack (top-right): deep platform plus deep physical integration. Examples: Nebius, CoreWeave, Lambda, Fluidstack. They aim to be a complete AI cloud and control much of their own infrastructure.
Access layer (bottom-left): focused on GPU access and inference with limited integration. Example: Vultr.
Infrastructure-led (bottom-right): a physical infrastructure focus with a narrower platform. Examples: TensorWave, Crusoe, Voltage Park, Nscale. Their edge is power, sites and the ability to build capacity.

Why the quadrant matters to lenders: Each archetype carries a different risk profile, and therefore a different financing mode

Why the quadrant matters to lenders

Example: A platform-led Neocloud raising $200 million might sell equity to venture investors because it has few hard assets to borrow against. An infrastructure-led Neocloud raising $2 billion for a 100 MW campus will use project finance secured on the power assets and the GPU contracts, and may need a lease or PPA guarantee from a stronger party to close the deal.

Movement across the map: The map isn't static. Platform-led players often try to move right by securing their own capacity, because owning infrastructure improves margins and guarantees supply. Infrastructure-led players try to move up by adding software, because platforms make customers stickier and support higher prices. Full-stack players try to defend their position by scaling both. Every move requires capital, and much of it requires debt.

How the four archetypes make money: a simple comparison: The quadrant a neocloud occupies shapes its profit and loss statement, not just its strategy. The comparison below uses illustrative proportions to show the typical "shape" of each business, not data for any named company.

How the four archetypes make money

Three practical consequences follow.

First, platform-led players are exposed to the price of capacity they rent: when GPU-hour prices rise (as the H100 rebound can happen, discussed later), their margins get squeezed unless they can pass the cost on.
Second, full-stack and infrastructure-led players carry the asset risk: they benefit when GPU prices are strong but carry the debt if prices fall.
Third, the access layer is where commoditisation bites first, the telecom-like corner of the industry.

Example: Suppose GPU-hour prices rise 20% in a year because new generations are sold out. A full-stack Neocloud with spare owned capacity earns more on every uncontracted GPU. A platform-led inference provider that rents capacity sees its main cost rise by 20% and must either raise its token prices or accept lower margins. The same market move is good news for one quadrant and bad news for another, a reminder that "Neocloud" is not a single business model.

The Neocloud life cycle

Most neoclouds follow a recognisable path. They start by renting or buying a small number of GPUs and selling access (access layer). As they win larger customers, they sign longer contracts and use them to raise GPU-backed debt (moving toward full-stack). To control costs and secure supply, they lease or build their own datacentres and secure power (moving right on the map). Along the way they add software, schedulers, inference engines, storage, developer tools, to make customers stickier (moving up). Each step requires more capital, and each step changes which financing tools are available. By the time a Neocloud is full-stack, it is raising billions in debt, signing multi-year offtakes with hyperscalers and labs, and, increasingly, relying on backstops from Nvidia or others to make those debts financeable.

Neoclouds versus hyperscalers at a glance

Neoclouds versus Hyperscalers at a glance

Ways to Fund AI Compute: Neocloud Debt vs SPV Leaseback

Once you need billions of dollars of GPUs, there are broadly two ways to pay for them. The supplied comparison graphic frames these as two "consolidated capital pathways."

AI compute funding strategies:

Strategy 1: The Neocloud / PaaS route - In the first route, associated with CoreWeave and Nebius, the Neocloud borrows money secured on its GPU clusters and the customer contracts that use them. The customer contract (the "platform-as-a-service" revenue offtake) provides the cash flow; the hardware provides the collateral.

Financial structure: GPU-collateralised debt.
Balance sheet impact: the debt sits on the Neocloud's balance sheet as a liability.
Cost of capital (per the graphic): about 8.5%-11.0%.
Accounting feel: the Neocloud carries capex; its customer pays PaaS operating expense.

Example: A Neocloud signs a four-year, $2 billion contract with a hyperscaler. It buys $1.6 billion of GPUs, borrowing $1.2 billion against the contract and the hardware at about 10%. The customer pays monthly; the Neocloud uses that to pay interest and principal. If the Neocloud stumbles, lenders can seize the GPUs and step into the contract.

Strategy 2: The SPV sale-leaseback route - In the second route, the hardware is owned by a special-purpose vehicle (SPV), a separate company set up only to own the assets. The SPV is "bankruptcy-remote": if the operating company fails, the SPV's assets are ring-fenced from its other creditors. The SPV raises money from investors, buys the equipment, and leases it to the user. The diagram above labels this the "Amazon/NVDA $8B structure" as an example of the model.

Financial structure: a bankruptcy-remote SPV.
Balance sheet impact: designed to be off the user's balance sheet as an operating lease (subject to accounting rules).
Cost of capital (per the graphic): about 5.5%–7.2%, because a well-structured SPV with strong lessees and asset isolation can achieve very high (triple-A-style) ratings on its senior debt.
Accounting feel: the user pays a lease, a pure operating expense, rather than carrying capex.

Example: A hyperscaler wants $8 billion of GPUs but would prefer not to add $8 billion to its own capex and debt. An SPV is created; investors fund it with highly rated notes and some equity; the SPV buys the GPUs and leases them to the hyperscaler for five years. The hyperscaler pays lease rentals as operating expense. Investors get a highly rated, ring-fenced cash flow. Everyone's cost of capital is lower than if a non-IG Neocloud had borrowed directly.

The SPV is "bankruptcy-remote": if the operating company fails, the SPV's assets are ring-fenced from its other creditors. The SPV raises money from investors, buys the equipment, and leases it to the user

What the difference in cost of capital means: The gap between roughly 9.75% (midpoint of route 1) and roughly 6.35% (midpoint of route 2) is about 340 basis points. On $5 billion of financing, that's about $170 million a year in interest. Over five years, that is close to the cost of a further 15,000 or so GPUs at today's prices (illustratively). Cost of capital isn't a technicality, it decides who can afford to build.

A simple way to think about the choice is to ask two questions: how strong is the user's credit, and does it want the assets on its balance sheet?

Credit Strength vs Balance Sheet

The bottom-right cell is where backstops become essential. A weak-credit user can't make an SPV highly rated on its own; it needs a stronger party to stand behind the lease or the asset value.

Case Study - How Money and GPUs Move Between Nvidia and CoreWeave

The relationships described in this post are easiest to understand through a real example. The supplied graphic (by Boring Biz, as of 13 October 2026, citing CoreWeave's Q2 2026 results and filings, Nvidia's SEC filings, Reuters and Bloomberg) maps the flows between Nvidia, CoreWeave, CoreWeave's lenders and its largest customers.

Source : SemiAnalysis - Money and GPU flows between Nvidia, CoreWeave, lenders and customers

CoreWeave - Financial View (Q2 2026)

The flows, one at a time

Nvidia → CoreWeave (money and commitments): Nvidia invested equity of $2.35 billion across 2023, 2025 and 2026; has been a cloud customer renting CoreWeave capacity since 2023; and provides a backstop to buy unsold capacity through 2032, with a $6.3 billion initial order.
CoreWeave → Nvidia (money back): CoreWeave buys Nvidia systems, often via OEMs such as Dell, and has been first in line for GB200, GB300 and Vera Rubin. Its 2026 capex guidance is $35–39 billion.
Nvidia → OpenAI. A $30 billion Nvidia equity investment in OpenAI (February 2026). OpenAI is in turn a CoreWeave customer.
Lenders → CoreWeave. Private credit, banks and bondholders have provided debt of $35.1 billion on the balance sheet at 30 June 2026, with GPU-backed loans used since 2023.
CoreWeave → lenders. Interest expense of $640 million in Q2 2026.
Customers → CoreWeave. A revenue backlog of $104 billion at 30 June 2026. Microsoft accounted for 67% of 2025 revenue; OpenAI had an $11.9 billion initial contract; Meta has a $21 billion contract through 2032.

What this case teaches

Growth and leverage together: Revenue is more than doubling year on year and the backlog is enormous, yet the company reports a net loss, partly because interest costs are large. That is typical of a capital-intensive business in a build phase: assets are paid for up front; revenue arrives over years.
Backlog is the collateral: A $104 billion backlog is what allows $35 billion of debt. Lenders are lending against contracted future revenue from large customers.
Concentration matters: With Microsoft at 67% of 2025 revenue, CoreWeave's credit is heavily linked to one customer's commitment. That's good (Microsoft is very creditworthy) and risky (a change in Microsoft's plans would matter a lot).
Circularity: Money flows from Nvidia to CoreWeave (equity, rental, backstop), from Nvidia to OpenAI (equity), from OpenAI to CoreWeave (contracts), and from CoreWeave back to Nvidia (GPU purchases). Critics describe this as circular financing: the supplier helps fund customers who then buy its products. Supporters argue it is a rational way to accelerate an ecosystem in which the supplier has strong conviction. Both can be true, and the telecom vendor-financing lesson discussed previously tells why markets watch it closely.
The backstop in action. Nvidia's commitment to buy unsold capacity through 2032 is a textbook example of the rental-floor-style backstops discussed in Chapter 12: it reduces the risk that CoreWeave is left with idle GPUs, and that in turn makes lenders more comfortable.

Critics describe this as circular financing: the supplier helps fund customers who then buy its products. Supporters argue it is a rational way to accelerate an ecosystem in which the supplier has strong conviction. Both can be true, and the telecom vendor-financing lesson discussed previously tells why markets watch it closely.

Example: Imagine a lender reviewing a new $3 billion GPU-backed loan to a company like CoreWeave. It sees: a backlog with three very large customers, an Nvidia commitment to buy unsold capacity, and a history of growing revenue. It also sees a net loss, heavy interest costs and high customer concentration. The lender's decision will hinge on whether the contracts and the backstop cover the debt service even if one large customer slows down. That is precisely the analysis this article equips you to follow.

The Financing Challenge

The Scale of the Bill - Why $11.3 Trillion, Not $1 Trillion

Headlines about hyperscalers spending "hundreds of billions" a year, and commentary about "$1 trillion" of AI investment, badly understate the real requirement, according to the SemiAnalysis. Its estimate of total AI-related CAPEX, IT equipment plus datacenter infrastructure, from 2024 to 2029 is approximately $11.3 trillion.

Why the popular number is too small: The difference is scope. The commonly cited figure counts what a handful of large public companies spend. The $11.3 trillion figure includes:

Hyperscalers - the big public cloud and internet companies.
Neoclouds - the GPU-focused clouds dicussed earlier
Independent datacenter operators - developers and landlords who build and lease capacity to others.
The full stack of equipment - not just GPUs, XPUs and TPUs, but servers, networking and power infrastructure.

It is like estimating the cost of a new city by counting only the houses built by the three largest builders. You miss the smaller builders, the roads, the substations and the apartment blocks built on spec.

How much of it is debt?

This is where the story becomes a financing story. According to semianalysis estimates, by 2029 roughly $7 trillion of debt will be outstanding to fund the build-out, and that net new funding needs are currently running at about $1 trillion per year. Even the richest technology companies cannot fund trillions purely from operating cash flow, and many participants have far less cash. So a large share must be borrowed, through bonds, bank loans, project finance, private credit and asset-backed securities.

Example. A 100 MW AI campus at the episode's figure of about $40 million per MW for GB300-class deployments costs roughly $4 billion. A young neocloud might have $400 million of equity. The other $3.6 billion must come from lenders. Multiply that across hundreds of planned campuses and the trillions add up quickly.

Why this matters outside finance: The cost and availability of capital at the bottom of the stack shows up at the top, that is in GPU-hour rates, token prices, the capacity limits vendors impose, and which regions get new AI capacity first.

Financing is a supply chain, and like any supply chain, its bottlenecks ripple outward.

How the Binding Constraint Moved - Chips, Then Power, Then Money

The "Trinity" of AI project constraints - For every AI infrastructure project needs the following at the same time

Every AI project needs three things at once

If any one is missing, the project stalls. The question at any moment is which of the three is scarcest.

The binding constraint has moved from compute, to power and sites, to financing.

Phase 1: Compute and memory - In 2023 and 2024 the scarce resource was silicon: GPUs on allocation, long lead times, constrained high-bandwidth memory and packaging. According to the SemiAnlayis, this has largely been relieved through more supply and "de-speccing", configuring systems with less of a scarce component. The reduced SOCAMM memory on NVL72 racks. SOCAMM is a compact, low-power memory module; fitting less per rack means each rack consumes less of a constrained part.

What is de-speccing

Example: If a supplier can deliver modules for 1,000 racks at full specification, trimming memory per rack by a quarter covers about 1,333 racks. Some workloads lose a little performance, but many more customers get capacity.

Phase 2: Power and sites - Next came finding somewhere to plug it all in: hundreds of megawatts per campus, slow grid connections, limited suitable land. According to the SemiAnlayis, this has also largely been addressed through capital mobilisation, money paying for faster construction and early equipment orders, plus refurbished sites with existing grid connections, and behind-the-meter power built on site.

Example: A developer buys a retired industrial site with an existing 300 MW connection, refurbishes the electrical infrastructure and adds on-site gas generation. Within 18 months it has a powered shell ready for GPUs, possible only because investors funded the purchase up front.

Phase 3: Financing - With chips and sites increasingly available, the scarcest part of the Trinity is now capital, specifically debt that is willing to lend on acceptable terms to the companies that need it. When the constraint was chips, power came from allocation. When it was sites, power shifted to developers and utilities. Now power accrues to whoever has the strongest balance sheet, and to whoever is willing to lend that balance sheet to others.

When the constraint was chips, power came from allocation. When it was sites, power shifted to developers and utilities. Now power accrues to whoever has the strongest balance sheet, and to whoever is willing to lend that balance sheet to others.

Project Finance 101 - Offtake, Investment Grade and Spreads

To understand backstops, we need four project finance ideas. None is complicated once we go through an example.

Key themes for backstops

Concept 1: Offtake - An offtake agreement is a contract in which a customer commits in advance to buy a project's output. In energy, a utility might buy all the electricity from a wind farm for 15 years. In AI, the output is compute capacity, GPU-hours or reserved megawatts. Lenders love offtake because it converts an uncertain future into contracted revenue. The question changes from "will anyone want these GPUs?" to "will this customer honour this contract?" The episode notes that most AI project financing requires a long-term offtake, typically five years, from an investment-grade counterparty.
Concept 2: Investment grade - Rating agencies (S&P, Moody's, Fitch) grade borrowers on their likelihood of repaying. Ratings of BBB-/Baa3 and above are "investment grade" (IG); below that is "high yield." The distinction matters because many of the largest pools of capital, insurers, pension funds, many bank programs, are restricted to, or strongly favour, IG exposure. Most hyperscalers are IG, several very highly. Most AI labs and neoclouds are not.
Concept 3: Non-recourse project finance: In project finance, lenders lend to a special-purpose vehicle that owns a specific asset, and are repaid only from that asset's cash flows. If the project fails, they can take the asset but usually cannot chase the parent company. That's why the quality of the offtake contract is so crucial: it is effectively the only thing between the lender and a loss.
Concept 4: Credit spreads: A credit spread is the extra interest a borrower pays above a risk-free benchmark (such as US Treasuries) to compensate for default risk. It is measured in basis points (bp); 100 bp equals 1 percentage point. A Z-spread is a technical measure of that premium across the whole yield curve; think of it simply as "the risk premium."

The investment-grade gap

In one sentence: the companies most urgently needing AI compute (labs and Neoclouds) generally aren't IG, but the lenders with the most money generally want IG offtake. The solution is to insert an IG company in the middle

Credit intermediation: a hyperscaler stands between a non-IG AI lab and the lender.

Example: An AI lab wants 200 MW of capacity for five years. Lenders won't finance a neocloud against the lab's contract alone. So the lab signs with a hyperscaler, which signs a five-year offtake with a Neocloud that builds and runs the cluster. The Neocloud takes the hyperscaler contract to lenders. Because the payer is now IG, lenders lend more at a lower spread. The hyperscaler earns a margin for standing in the middle, and takes the risk that the lab doesn't pay.

Backstop credit is not zero risk Even with a strong counterparty, lenders charge for the risk that the project itself goes wrong: the cluster is late, underperforms or is poorly run. According to SemiAnalysis this is known as execution risk, linking it to what they call it as "ClusterMAX risk." Example: Meta's debt at roughly a 97 bp Z-spread versus CoreWeave project financing at around 200 bp.

Source : SemiAnalysis - Even with a strong counterparty, lenders price in execution risk.

Example (illustrative). On a $2 billion loan, an extra 100 bp costs $20 million a year, about $100 million over five years. That's why Neoclouds care so much about operational track records: better execution can translate directly into cheaper debt.

Backstops Explained - What They Are and How They Work

In baseball, the backstop is the fence behind home plate that stops balls going any further. In finance, a backstop is a commitment by a stronger party to absorb a loss, fill a funding gap or buy an asset if things don't go to plan. It doesn't usually provide money on day one; it provides certainty that money will be there if needed. That certainty is what lenders are really buying.

Everyday example how Backtops work

A helpful everyday analogy is a parent guaranteeing a young adult's apartment lease. The landlord would hesitate to rent to someone with no credit history. With a parent's guarantee, the landlord is happy, not because the parent will pay the rent, but because the parent would pay if the tenant didn't. The tenant gets the apartment; the parent takes on a contingent obligation; the landlord gets comfort. Replace "tenant" with neocloud, "parent" with Nvidia or a hyperscaler and "landlord" with lenders, and you have the backstop universe.

The main types of backstop

Why backstops work: Transferring credit - The essence of a backstop is credit transfer. A lender's risk depends on who ultimately pays. If the ultimate payer becomes a company with a strong balance sheet, the lender's risk falls, the loan gets bigger and the interest rate falls. The backstop provider is effectively selling insurance and getting paid in fees, in upside sharing, or in extra product sales.

Backstop - Credit Transfer

The cost of a backstop: contingent liabilities - Backstops aren't free. They are contingent liabilities, obligations that only become real if something goes wrong. Many sit off the provider's balance sheet, appearing in notes to the accounts rather than as debt. That creates two concerns:

Visibility. Investors may underestimate total exposure if commitments are scattered across many types of agreements.
Correlation. Backstops tend to be called at exactly the time the provider's core business is weakest. If AI demand fell, Nvidia would sell fewer chips and be asked to honour more floors and guarantees at the same time.

A useful way to judge any backstop is to ask two questions: how likely is it to be called, and how painful would it be if it were?

Backstop risk: probability of being called vs severity if called.

Comfort (unlikely, small). Priority supply allocations, memoranda of understanding, soft support. Useful signals; low cost.
Routine support (likely, small). Cloud services agreements where the provider genuinely uses the capacity. These are "called" every month, but the provider gets real value in return.
Tail risk (unlikely, large). Residual value guarantees, lease and PPA guarantees. Normally never paid, but potentially very large in a downturn or a technology leap.
Danger zone (likely and large). Broad rental floors across many customers in a weak market. A sensible provider limits how much sits here.

Example: Nvidia's commitment to rent capacity for its own research sits in "routine support": it uses the GPUs. Its residual value guarantee on a large complex sits in "tail risk": unlikely to be paid if GPU prices hold, very costly if they collapse. A floor of $3.25 per hour across many Neoclouds sits closer to the danger zone if market rates fell well below the floor.

Backstops are not new: Aircraft manufacturers support airline financing through export credit agencies and guarantees. Car makers run captive finance companies. Equipment makers routinely offer vendor finance. Energy projects are routinely backed by government or utility offtake. What's new in AI is the scale, the speed and the short life of the underlying asset.

Inside Nvidia's Backstop Universe - Roughly $497 Billion of Commitments

What the SemiAnalysis team found: Their claim is that Nvidia has committed roughly $497 billion across multiple categories of guarantees and support. Many sit off its balance sheet, not recorded as debt in the traditional sense, but representing real or contingent obligations. The categories are:

AI cloud agreements with neoclouds.
Datacenter leases.
Power purchase agreement (PPA) guarantees.
Cloud services agreements.
Supply and capacity commitments, especially memory.
Residual value guarantees.

Source : SemiAnalysis - Off Balance Sheet Commitments

Why would a chip company guarantee other people's obligations?

Nvidia's growth depends on customers being able to afford its systems. If neoclouds and labs can't raise money, they can't place orders. In a world where financing is the bottleneck, the fastest way to sell more systems is to make customers more financeable. Nvidia has an exceptionally strong balance sheet; by putting some of it behind customers, it unlocks financing that would otherwise be unavailable or prohibitively expensive. It converts creditworthiness into product demand aka "Balance Sheet as a Service."

There's a strategic angle too. Nvidia's largest customers, the hyperscalers, are scaling their own chips. A thriving neocloud ecosystem that buys Nvidia systems diversifies Nvidia's customer base. Backstops help keep that ecosystem growing.

Who backstops whom: Nvidia and hyperscalers sit between capital markets and less-creditworthy builders and users

Structure 1: AI cloud agreements and the AICP rental floor - The most distinctive structure is what the SemiAnalysis team calls is the NVIDIA AI Compute Partnership (AICP) program. Nvidia guarantees a minimum rental rate on GPU capacity operated by a Neocloud, on average about $3.25 per GPU-hour for six years. If demand falls short, Nvidia effectively stands ready to take the capacity at that floor. In exchange, it commonly receives a share of the upside, revenue-share arrangements of often 50% of rental income above the floor.

An AICP-style rental floor turns an uncertain rental stream into financeable revenue.

Example (illustrative). A Neocloud deploys 10,000 GPUs. At $3.25/hr, around the clock, each GPU earns about $28,470 a year, so the fleet earns roughly $285 million a year — about $1.7 billion over six years — even in a downside case. Lenders can size a loan against that. If the market rate is $4.25/hr, the extra $1/hr is shared: $0.50 to the neocloud, $0.50 to Nvidia. Nvidia has turned its guarantee into an option on the upside; the neocloud has turned an uncertain business into a financeable one.

Economically, Nvidia is selling insurance on GPU rental prices and being paid in upside participation. If demand stays strong, it earns extra income at little cost. If demand weakens, it pays for capacity no one else wants, just as its own chip sales would likely be falling.

Structure 2: Datacenter leases - Developers typically need a long-term lease from a creditworthy tenant before they can raise construction finance. When Nvidia signs or guarantees a lease, the developer borrows against Nvidia's credit.

Example: A developer is building a 150 MW powered shell for a Neocloud whose credit isn't strong enough for the developer's lenders. Nvidia guarantees the lease for the initial term. The construction loan is approved at a much tighter spread.

Structure 3: PPA guarantees - Power generators and their lenders need to know the buyer of the power will pay for many years. Nvidia may guarantee a Neocloud's or operator's PPA so the power project itself can be financed.

Example: An energy company will build on-site generation for an AI campus only if the 10-year offtake is creditworthy. Nvidia guarantees the operator's PPA payments. The generator is financed, the campus gets power sooner, and Nvidia's GPUs get a home.

Structure 4: Cloud services agreements - Under cloud services agreements, Nvidia itself commits to buy cloud capacity, for research, its own cloud services or resale. That gives the provider contracted revenue from a highly creditworthy counterparty. The CoreWeave case in Chapter 7 shows Nvidia renting capacity since 2023.

Structure 5: Supply and capacity commitments - Nvidia commits to buy components, particularly memory, to secure supply. These commitments give suppliers the confidence to expand manufacturing, but must be paid for whether or not Nvidia needs the volume.

Example: A memory maker will add a production line only with firm multi-year orders. Nvidia signs a take-or-pay commitment. The line gets built and supply is secured, but if demand fell, Nvidia would still owe for the volume.

Structure 6: Residual value guarantees - A residual value guarantee (RVG) promises an asset will be worth at least a certain amount at a future date; the guarantor pays any shortfall. For GPUs, RVGs address lenders' biggest worry: what the hardware will be worth when the loan ends. The episode cites a $105 billion residual value guarantee associated with the Ports Pike complex as an indication of scale.

Example: A vehicle buys $1 billion of GPUs on a six-year loan structured so $300 million remains outstanding at maturity, to be repaid by selling the hardware. Nvidia guarantees the hardware will be worth at least $300 million. If it sells for $200 million, Nvidia pays $100 million. Because the "balloon" is protected, lenders accept smaller repayments in early years, improving the project's cash flow.

Hyperscalers play the same game

Hyperscalers also act as balance-sheet providers. When Microsoft, Oracle, Google or Amazon sign long-term capacity agreements with Neoclouds on behalf of AI labs, they substitute their credit for that of a less-creditworthy end user. The motive differs, hyperscalers secure capacity they can resell or use; Nvidia sustains demand for its product but the mechanism is the same.

Part IV - The Economics

A Simple Financial Model of a GPU Cluster

Talking about backstops and spreads in the abstract only gets you so far. This section builds a deliberately simple financial model of a single GPU cluster, the kind a Neocloud might finance, so you can see how the numbers fit together and why lenders, equity investors and guarantors each care about different things.

Everything discussed here is illustrative. The inputs are round, plausible numbers chosen to make the mechanics clear, not estimates of any real company's economics. The model is pre-tax and ignores working capital, refinancing and many other real-world details.

Step 1: The inputs

Source: SemiAnalysis - Inputs for A GPU Cluster

Step 2: Revenue

Revenue = GPUs × price per hour × hours per year × utilisation.
25,000 × $3.50 × 8,760 × 95% ≈ $728 million per year.

Step 3: Operating costs

Power: 25,000 × 1.5 kW × 8,760 hours × $0.08 ≈ $26 million.
Datacenter lease: 25,000 × $0.25 × 8,760 ≈ $55 million.
Operations and support: 10% × $728 million ≈ $73 million.
Total operating costs ≈ $154 million.

Step 4: EBITDA (earnings before interest, tax, depreciation and amortisation) is a proxy for operating cash flow: $728 million - $154 million ≈ $574 million per year, an EBITDA margin of about 79%. That seems very high, and it is, because the biggest cost of a GPU cluster isn't operating cost; it's the capex. The real question is whether $574 million a year for five years, plus whatever the hardware is worth at the end, pays back $1,750 million with an adequate return.

Step 5: Debt service - The Neocloud borrows 70% of capex $1,225 million at 9.5% over five years, repaid in equal annual instalments. The annual payment (interest plus principal) is about $319 million.

Step 6: Cash to equity - Cash left for equity holders = EBITDA debt service ≈ $574 million - $319 million ≈ $255 million per year, plus the residual value of about $262 million at the end of year five.

Source: SemiAnalysis - GPU Cluster Economics

Step 7: The key outputs of a GPU Cluster Economics: The project IRR of about 22% sits within the 20-30% range the where projects are typically underwritten to, without relying on residual value. Remove the residual entirely and the project IRR in this model falls by about 2.6 percentage points, still close to 20%.

Source: SemiAnalysis - Key Outputs of a GPU Cluster Economics

Understanding the outputs - Three numbers and three audiences

Source: SemiAnalysis - Three numbers and Three audiences

Project IRR is the annualised return on the whole $1,750 million, as if it were all equity. It measures the quality of the asset.
Equity IRR is the return to the $525 million of equity after debt is serviced. Because debt is cheaper than the project's return, borrowing magnifies equity returns, here from about 22% to about 44%.
DSCR (debt service coverage ratio) is what lenders watch. A DSCR of 1.80x means EBITDA covers debt payments 1.8 times. Lenders typically require a minimum, say 1.25-1.40x, with a cushion for bad years.

The downside case: Now suppose the market weakens and there is no contract or backstop: the price falls to $2.00 per hour and utilisation to 70%.

Source: SemiAnalysis - Utilisation case (Base vs Downside)

A DSCR of 0.61x means the cluster cannot pay its debt. Equity is largely wiped out and lenders face losses unless the hardware can be sold for enough. The same cluster that looks wonderful at $3.50 and 95% is a disaster at $2.00 and 70%. That gap is precisely what backstops and long-term contracts are designed to close.

Sensitivity: price × utilisation - The heatmap below shows project IRR across combinations of price and utilisation. Think of it as a 5×5 version of a 2×2: high price and high utilisation is the "boom" corner; low and low is the "bust" corner.

Project IRR sensitivity to price per GPU-hour and billable utilisation (illustrative model).

Three observations stand out:

First, the model needs both a decent price and high utilisation; neither alone is enough.
Second, below about $2.50 per hour the project struggles at almost any utilisation.
Third, the slope is steep: each $0.50 change in price moves project IRR by roughly 5-7 percentage points.

Why backstops change the maths: debt capacity: Lenders size loans on what they believe is a realistic downside, not the base case. A common method is to ask: what is the largest loan that still has a DSCR of at least 1.30x in the case we're underwriting? The answer depends heavily on who stands behind the revenue.

Source: SemiAnalysis - Debt capacity of the same cluster under three credit scenarios (illustrative model, 1.30x minimum DSCR, 5-year amortising).

No backstop: Lenders underwrite a cautious case ($2.00 at 70%) and charge 11%. Maximum debt: about $554 million, roughly 32% of capex. The Neocloud would need to find more than $1.2 billion of equity.
Nvidia-style floor: Lenders can underwrite the guaranteed floor ($3.25 at 95%) and, with less risk, charge 8.5%. Maximum debt: about $1.6 billion, over 90% of capex. In practice lenders would cap loan-to-value around 70–80%, but the floor lets them go to that cap comfortably.
IG hyperscaler offtake: Lenders underwrite the contract ($3.50 at 95%) with an investment-grade payer at about 7%. Debt capacity exceeds the cost of the cluster.

This is the single clearest demonstration of "Balance Sheet as a Service." The cluster, the GPUs and the datacenter are identical in all three bars. The only thing that changes is whose balance sheet stands behind the revenue, and that alone roughly triples how much can be borrowed.

Adding the backstop to the model: what a rental floor is worth

Now add an Nvidia-style floor of $3.25 per GPU-hour, with a 50% share of any rental income above the floor going to the guarantor. Compare four cases on the same cluster:

Source: SemiAnalysis

Read the above table as an insurance contract.

In the downside, the floor turns a cluster that cannot pay its debt service( DSCR 0.61x) into one that comfortably can (1.65x).
In the upside, the Neocloud gives up about 7 points of project IRR and roughly 19 points of equity IRR to the guarantor.
The Neocloud is trading away part of its best case to eliminate its worst case, exactly what anyone buying insurance does.

From the guarantor's side, the arithmetic is mirrored. In good times it collects $0.50 per GPU-hour on the excess, about $104 million a year on this cluster at a $4.25 market rate. In bad times it pays $1.25 per hour on capacity it must absorb, roughly $260 million a year here if it took all the capacity at $3.25 against a $2.00 market. Whether that is a good trade depends on the probability of each state of the world, and as discussed earlier under Backstops, the bad state for the backstop is also the bad state for the guarantor's core business.

Example: A Neocloud chief financial officer presents both versions to the board. Without the floor, the lender offers about a third of the capex as debt at 11%; the company must raise more than $1.2 billion of equity, diluting existing shareholders heavily. With the floor, the lender offers 70% at 8.5%. The board accepts the floor and the upside share because the cost of not having it, dilution, or not building at all, is far higher than the upside given away.

Financial Levers - What Really Moves Returns

A "lever" is any input that management, lenders or guarantors can influence to change the outcome. Understanding which levers matter most tells you where to focus negotiation, engineering effort and risk management.

The tornado chart: The tornado chart below changes one input at a time, holding everything else at base case, and shows how far project IRR moves.

Tornado chart: change in project IRR when each lever is flexed (illustrative model, base project IRR 21.8%).

The levers, from most to least powerful

Price per GPU-hour: A ±20% change moves project IRR by about ±9–10 points. Price is set by contracts (fixed for the term) or by the market (volatile). Levers: contract length, rate floors, revenue sharing, and moving up the value chain into managed services that command higher prices.
Utilisation: Dropping from 95% to 76% costs about 10 points of IRR; going to 100% adds only about 2.5. That asymmetry matters: there is little upside but lots of downside. Levers: take-or-pay contracts, good scheduling software, a diverse customer base, and fast deployment so GPUs start earning sooner.
Capex per GPU: Paying 20% more for the same capacity costs about 7 points; paying 20% less adds about 10. Levers: purchasing scale, vendor relationships, "de-speccing" configurations, buying at the right point in the product cycle, and avoiding construction delays that tie up capital.
Asset life / contract term: A four-year life instead of five costs almost 6 points; six years adds about 3.5. Levers: longer contracts, re-leasing to inference workloads once frontier training moves on, and maintaining hardware so it stays rentable.
Residual value: Moving from 15% to 0% or 30% shifts IRR by about ±2.5 points. Modest in this model, because projects are underwritten not to depend on it, but crucial to lenders' balloon repayments and to RVG providers.
Datacenter lease and power: A ±20% change shifts IRR by less than 1 point each. In GPU economics, power and rent matter far less than price, utilisation and capex. (They matter a lot more to the datacenter developer whose whole business they are.)

The leverage lever: a double-edged sword - Debt is the lever that most dramatically changes returns to equity, in both directions.

Source: SemiAnalysis - Equity IRR at different levels of debt, base case vs a soft case (illustrative model).

In the base case, raising debt from zero to 80% of capex lifts equity IRR from about 22% to about 58%.
In a soft case ($2.75 at 80% utilisation), the same increase drags equity IRR from about +5% to about –10%, and at 80% debt the DSCR falls below 1.0x, the cluster can no longer cover its debt.

This is why lenders insist on equity cushions and why backstops that support the downside are so valuable: they let borrowers use more leverage without facing ruin if the market softens.

The cost-of-capital lever: As discussed earlier, the gap between Neocloud debt (about 8.5–11%) and SPV/IG-style funding (about 5.5–7.2%). In this model, cutting the interest rate from 9.5% to 6.5% on $1,225 million of debt reduces annual debt service from about $319 million to about $294 million and raises cash to equity by roughly 10%. Credit quality is a financial lever in its own right, and backstops are the most direct way to pull it.

Source: SemiAnalysis - Cost of Capital

Levers: impact versus controllability

The top-left cell is where backstops live. Nobody can control the market price of GPU-hours five years out, but a floor, a take-or-pay contract or an RVG can transfer that risk to someone better able to bear it.

Levers for each player

Neocloud: contract length and coverage, capex discipline, deployment speed, operational excellence (to narrow the execution-risk spread), and adding software to escape commodity pricing.
Lender: loan-to-value, DSCR covenants, amortisation profile, cash sweeps, security over GPUs and contracts, and requiring backstops.
Nvidia (as guarantor): floor level, term, upside share, and how much total exposure to accept in each quadrant of the backstop risk matrix.
Hyperscaler (as intermediary): the margin it earns between lab and neocloud, contract terms on both sides, and its own rating headroom.
AI lab (as end customer): reserved vs on-demand mix, model efficiency (more tokens per GPU-hour), and pricing.

Profitability Is Extremely High at Most Layers

Revenue per megawatt: According to SemiAnalysis, Frontier token generation can generate up to $100 million of revenue per megawatt per year on GB300 systems that cost about $40 million per megawatt to deploy. Neoclouds renting capacity on five-year contracts were earning roughly $12 million per MW per year a few months ago.

Source: SemiAnalysis Per-megawatt economics

In other words, at the upper end, a megawatt of the latest systems serving frontier tokens at high utilisation can earn its entire deployment cost in about five months. The Neocloud renting out that same megawatt earns far less, but against $40 million of cost still pays back in about three and a half years before operating costs, and its revenue is contracted.

Example: A neocloud deploys 50 MW at $40 million per MW, $2 billion. At $12 million per MW per year, it earns $600 million a year in rent. The lab renting that capacity might generate several billion dollars a year of token revenue on it if it keeps utilisation high. The lab captures most of the value; the neocloud captures a smaller but more secure slice.

Gross margins on tokens: Gross margins on frontier API tokens can approach 90% in some models, accounting for cache hits and high utilisation.

Cache hits - when part of a prompt has been seen before (a long system prompt or a document reused across requests), the model can reuse earlier computation. Providers charge less for cached tokens, but their cost to serve is lower still.
Utilisation - the share of time GPUs do paid work. Idle GPUs still cost money. High utilisation spreads fixed costs across more revenue.

Example: A lab sells 1 million output tokens for $15. If its cost, after cache reuse and high batch utilisation is $1.50, its gross margin is 90%. If utilisation halves, cost per token roughly doubles to $3, and margin falls to 80%.

Profitability across the stack

Profitability at each layer

The catch: Margin is not cash flow, and gross margin is not net profit. Labs spend heavily on training, research staff and free usage; Neoclouds like CoreWeave report net losses while growing quickly because of interest and depreciation. The financing question isn't "is AI profitable at the margin?", the question often is, but "is that profit reliable enough, for long enough, to repay trillions of dollars of debt?"

Residual Value and the Strange Economics of GPU Depreciation

What residual value means: Residual value is what an asset is worth at the end of a period of use for a car lease, what the car is worth when you hand it back; for a GPU cluster, what the hardware can be sold or re-rented for after the financing term. It determines how much of a loan must be repaid from operating cash flows versus from selling or re-leasing the asset.

The conservative approach: don't count on it: Most AI projects, per the episode, are underwritten to achieve 20–30% IRR over 5–6 years without relying on residual value. Anything the hardware is worth afterwards is upside. A 20–30% hurdle is high compared with traditional infrastructure, reflecting the perceived risk.

Example: A $1 billion cluster generating about $350 million a year of net cash for five years, with zero residual, produces an IRR of roughly 22%. If the hardware is worth $200 million at the end, the IRR rises to roughly 25%. The underwriting doesn't need the residual, but every dollar of it helps.

The theoretical depreciation curve: In theory, a GPU's value should fall as better GPUs arrive. If a new GPU delivers three times the performance at a similar hourly price, the old one should rent for roughly a third as much per unit of work. This "relative performance" model implies steep depreciation.

What actually happened: the curve broke: This curve broke in late 2025 because demand outstripped supply. H100 rental prices, for a GPU that shipped in volume in 2023, rebounded sharply toward early-2023 levels of about $3 per hour on one-year terms. Newer GPUs like GB300 are sold out, so buyers "cascade down the stack" to older hardware.

Example: A start-up needs 512 GPUs for inference next month. No GB300 or H200 capacity is available. It signs a one-year deal for H100s at about $3/hr. Multiply that across thousands of buyers and older GPUs, which were supposed to be getting cheaper, are in demand again.

The cascade effect: Think of the GPU market as a staircase. New demand arrives at the top step (the latest generation). When it's full, demand spills to the next step, then the next. As long as total demand exceeds supply, every step stays full and prices hold. If supply at the top expands suddenly, demand drains back up, and the lowest steps empty first.

Source: SemiAnalysis - The cascade effect

Why this matters for financing: The rebound is good news for anyone who financed H100 clusters, and it makes residual value guarantees look less risky in the short term. But is today's strength permanent, or a symptom of shortage? If supply catches up, or if a shift to custom chips like ASIC changes what inference runs on, the theoretical curve could reassert itself quickly. This uncertainty is exactly what makes lenders cautious and backstops valuable.

The Shift From GPUs to ASICs and What It Means for Financing

Major tech companies including Microsoft, Meta, Amazon, and Google are actively scaling their own custom ASICs (application-specific integrated circuits). As AI workloads continue to shift from model training to inference, the adoption of specialised ASICs is accelerating, positioning custom silicon to capture a rapidly growing share of total chip volumes by 2030

What is an ASIC: An ASIC (application-specific integrated circuit) is a chip designed for a particular job. A GPU is a flexible, programmable engine that can run almost any AI workload well. An ASIC gives up some flexibility to run a narrower set of workloads more cheaply or more efficiently. In AI, the main examples are Google's TPUs, Amazon's Trainium and Inferentia, Meta's MTIA and Microsoft's Maia. Many are designed with partners such as Broadcom or Marvell.

Why inference favours ASICs

Stability: Training frontier models changes constantly, new architectures, new techniques. Flexibility is valuable, and GPUs, with Nvidia's mature software ecosystem, dominate. Inference for a popular model is more stable: the same model runs billions of times, so it's worth optimising hardware for it.
Cost per token: In inference, the winning metric is cost per token at acceptable speed. A chip that does one job very efficiently can beat a more general chip.
Control: Hyperscalers that design their own chips capture the margin they would otherwise pay to a chip vendor and control their own supply.

The market-share projection

Projected compute market share, training vs inference, 2024 vs 2030 (redrawn from the supplied chart).

Training: Nvidia GPUs hold about 78% in both 2024 and 2030, with others (ASIC, CPU, smaller GPUs) at about 22%. Training remains GPU territory.
Inference: GPUs (Nvidia, AMD and others) fall from about 95% in 2024 to about 45% in 2030, while ASICs/TPUs rise from about 5% to about 55%.

Combine that with the three eras discussed in the beginning, demand is shifting from training to inference and agents, and the implication is clear: a growing share of the total compute market could run on non-GPU silicon.

GPUs vs ASICs

Where GPUs vs ASICs win: workload novelty against buyer scale and software control.

GPU heartland (novel workloads, labs and Neoclouds). Frontier research and varied training. Flexibility and software ecosystem win.
Contested (novel workloads, hyperscalers). Hyperscalers train frontier models on both GPUs and their own chips.
Flexible inference (stable workloads, many customers). Neoclouds and enterprises serving many models for many customers still value GPUs' flexibility, which also helps GPUs keep resale value.
ASIC heartland (stable workloads, hyperscalers). High-volume inference on a hyperscaler's own models, where custom chips can win on cost per token.

Why this matters for AI financing

Residual values: If more inference moves to ASICs, the "cascade" that has kept older GPUs in demand could weaken. That increases the risk on residual value guarantees and on loans with large balloon payments.
Who borrows: ASICs are mostly deployed by hyperscalers, financed from IG balance sheets. GPUs are the main asset of Neoclouds, financed with GPU-backed debt and backstops. A shift to ASICs shifts the financing mix toward hyperscaler balance sheets.
Nvidia's incentive: Backstopping Neoclouds is partly a strategic defence: a strong, independent Neocloud ecosystem is a market for Nvidia systems that isn't building its own chips. The more hyperscalers shift to ASICs, the more valuable that ecosystem becomes to Nvidia, and the more it may be willing to stand behind it

Correlated risk: Should customer adoption shift heavily toward ASICs and reduce GPU demand, Nvidia would face a dual vulnerability: falling product sales alongside heightened financial exposure from vendor backstops, mirroring past telecom sector financing patterns.

The counter-arguments: To be balanced, Nvidia continues to release new generations quickly, its software ecosystem is deeply entrenched, its systems are also used for inference at very large scale, and agentic workloads may be less "stable" than classic inference, keeping flexibility valuable. Custom chips are costly to design and take years to mature. The projection is a forecast, not a certainty.

Example: A lender is considering a six-year loan against an H100-class fleet with a 25% balloon. In a world where inference stays largely on GPUs, the fleet might still rent for inference in year six. In a world where 55% of inference has moved to ASICs, the same fleet may struggle to find tenants. The lender's answer: a shorter tenor, a smaller balloon, or a residual value guarantee from someone else.

Part V - Markets and Risk

Credit Markets Are Already Reacting

Hyperscaler spreads are widening: One clear indicator being monitored is that credit spreads on hyperscaler debt have widened, Amazon and Oracle are among those. Wider spreads mean investors want more compensation to lend to these companies than before. Why would that happen to some of the most creditworthy companies in the world?

Supply: They are issuing much more debt, and markets need a higher yield to absorb it.
Concentration: Bond investors have limits on exposure to any single issuer or sector.
Contingent obligations: Capacity agreements and guarantees may not appear as debt, but analysts increasingly factor them in.
Uncertainty: The long-term return on AI investment at this scale isn't yet proven.

Example : A hyperscaler that used to borrow at 70 bp over Treasuries now pays 100 bp. On $50 billion of new bonds, the extra 30 bp costs $150 million a year, and signals that bond investors are pricing in AI-related risk.

AI debt is becoming a major asset class

According to SemiAnalysis, AI-related debt is rapidly scaling to become the second-largest US asset-backed financing market behind mortgages, already overtaking auto loans, student loans, and credit cards. Furthermore, 2026 is projected to be the point at which AI debt accounts for more than 10% of all US debt issuance. Mortgages, auto loans and credit cards are mature markets built over decades with deep data on defaults and recoveries. AI infrastructure debt has emerged in a few years, backed by assets whose long-run behaviour is still largely unknown.

Why ABS matters: Asset-backed securities pool assets that generate cash (mortgages, car loans, equipment leases) and issue bonds backed by that cash, often in tranches. AI debt is moving into ABS-style structures because they let lenders slice risk and reach a much wider pool of investors. But they also mean investors who may not deeply understand GPU economics are buying exposure to them.

Example: A Neocloud pools 30 customer contracts backed by GPU clusters and issues a $3 billion ABS deal in three tranches. An insurer buys $500 million of the AAA senior tranche because it pays more than comparable corporate bonds. The insurer is now indirectly exposed to GPU rental markets, customer credit and residual values — without ever having analysed a datacenter.

What lenders need to learn: Lenders need better tools to evaluate GPU economics, cluster execution risk, token profitability and residual values. Without them, credit is priced conservatively, or, more dangerously, inconsistently.

How a Lender Looks at an AI Deal, A Practical Checklist

Everything so far can be condensed into the questions a careful lender, credit committee or rating analyst asks about an AI infrastructure financing. It is also a useful checklist for anyone on the other side of the table like a Neocloud, a developer, or a vendor whose customers need financing.

Source SemiAnalysis: Lender Checklist

Example: A credit committee reviewing a $1.5 billion GPU loan scores each of the eight areas from 1 to 5. The deal scores well on payer quality (IG hyperscaler offtake), revenue firmness (five-year take-or-pay) and structure (70% LTV with cash sweep), but poorly on concentration (one end customer) and residual value (large balloon on hardware likely to face ASIC competition). The committee approves the loan on condition that the balloon is reduced or covered by a residual value guarantee, a typical outcome, and a direct illustration of why backstops exist.

Bond Yields Versus Equities - Why Markets Are Worried

The AI build-out isn't happening in a vacuum. It is being financed in global bond markets that are also absorbing government deficits, inflation risk and central bank policy. When bond yields rise, it affects AI in two ways at once: it makes AI debt more expensive, and it changes how investors value AI equities. This chapter explains why.

Bond yields in plain English: A bond's yield is the annual return an investor earns for lending money by buying it. The US 10-year Treasury yield is the benchmark "risk-free" long-term interest rate for the world. Almost every other long-term rate, mortgages, corporate bonds, project finance, is priced as the Treasury yield plus a spread.

Bond prices and yields move in opposite directions. If yields rise, existing bonds with lower coupons are worth less, so their prices fall.

Why rising yields hurt equities, especially growth equities

The value of a share is, in theory, the present value of all the cash a company will generate in the future. To work out present value, investors discount future cash flows using an interest rate that starts with the risk-free yield. When yields rise:

Future cash is worth less today: A dollar in ten years is worth less when the discount rate is higher.
Long-duration companies are hit hardest: Companies whose value depends on profits far in the future, like many AI and technology stocks, fall more than companies with steady near-term profits.
Bonds become competition: If a safe Treasury pays a high yield, investors need a bigger expected return to justify the risk of owning shares.

Example. An AI company is expected to generate $10 billion of cash in year ten. At a 7% discount rate, that's worth about $5.1 billion today. At 9%, it's worth about $4.2 billion. A two-point rise in the discount rate removes nearly a fifth of the value of that future cash — without anything changing about the business.

The stock–bond correlation and the 5.25% threshold

Why rising bond yields worry equity markets during the AI build-out.

The chart above (Bloomberg/MacroBond) plots the US 10-year yield against the rolling correlation between stocks and bonds. Its message is that stocks and bonds are almost never strongly negatively correlated when the 10-year yield is above about 5.25%.

Why does that matter: For most of the past two decades, when yields were low, bonds and stocks tended to move in opposite directions in a sell-off: when stocks fell, investors rushed into government bonds, bond prices rose, and portfolios were cushioned. That negative correlation is what makes the classic "60/40" portfolio (60% shares, 40% bonds) work.

When yields are high, typically because inflation and the cost of money are the main concern, that cushion tends to disappear. Stocks and bonds can fall together, because the same thing (rising rates) is hurting both. Historically, as the chart shows, above about 5.25% on the 10-year the strongly negative correlation almost never appears.

Why markets are worried in the context of AI: Put the pieces together and you can see why investors watch bond yields so closely during the AI build-out:

The AI build-out adds to bond supply: With roughly $1 trillion a year of new funding needs and AI debt expected to exceed 10% of US issuance in 2026, AI is a significant new borrower competing with governments and other companies for the same pool of savings. More supply of bonds can mean a higher yield is needed to clear the market.
Higher yields raise the cost of AI projects: Every AI loan is priced off Treasuries plus a spread. If base yields rise and spreads widen at the same time, project IRRs fall as discussed earlier how cost-of-capital lever works in reverse.
Higher yields lower AI equity valuations: AI-exposed companies are some of the longest-duration assets in the market, so they are especially sensitive to discount rates.
The hedge breaks: If yields rise into the zone where stocks and bonds stop offsetting each other, investors heavily exposed to AI equities lose their usual shock absorber. That can make sell-offs sharper.

A feedback loop: Lower equity valuations make it harder for AI companies to raise equity; higher yields make debt more expensive; that makes projects harder to finance; which in turn can weigh on AI equities again.

Yields versus AI demand

Yields vs AI Demand

Example:. A Neocloud plans to refinance a $3 billion construction loan into a five-year bond. When it planned the project, Treasuries plus spread implied a 7.5% coupon. By completion, yields have risen and spreads widened, so the coupon is 9.5%. On $3 billion, that's $60 million more interest a year, and the Neocloud's share price has also fallen, making new equity expensive. Nothing about its GPUs or customers has changed; the cost of money has.

Keeping it in perspective: Bond-market concerns don't mean AI is a bubble or that a crisis is coming. They mean the financing of AI is now big enough to interact with, and be affected by, the broader economy. That is a defining feature of an infrastructure build-out, and it's exactly why credit has become the bottleneck.

Where Durable Demand Comes From - Agentic AI and the Decision Matrix

Lenders ultimately need to believe that demand will last for the life of their loans. The most important source of new, durable demand is the move to agentic AI. But how far will organisations actually let AI act on their behalf? A useful framework comes from MIT Sloan.

The AI Decision Matrix: The following Matrix (adapted from I. Sebastian, P. Weill, T. Haskamp and J. vom Brocke, "Designing Decision Rights for AI," 2026, MIT Sloan) is a framework for deciding how humans and autonomous AI agents should share decision-making, based on two dimensions: ambiguity and risk.

AI Decision Matrix

Each decision has three components: Frame (define the decision), Act (take or recommend actions) and Learn (own and improve outcomes). Each component can be done by a human, a human with AI, or AI alone.

Routine (low ambiguity, low risk) - well-defined decisions with limited downside. Approach: automate and monitor. Frame, act and learn can all be done by AI.
Consequential (low ambiguity, high risk) - well-defined decisions with broad impact on many people or significant resources. Approach: manage to ensure reliability. AI frames and acts; humans own learning.
Exploratory (high ambiguity, low risk) - new or unclear decisions where learning matters more than precision. Approach: enable experimentation to learn fast. Human with AI frames; AI acts; humans own learning.
Strategic (high ambiguity, high risk) - new or unclear decisions with organisation - or ecosystem-wide consequences. Approach: lead with continuous oversight. Human with AI frames and acts; humans own learning.

Why this matters for compute demand: The matrix is about governance, but it has a direct link to infrastructure economics. Every decision that moves into the "AI acts" section becomes a source of continuous, machine-driven compute demand, the "all time" era. Routine decisions are the first and largest pool: invoice matching, ticket triage, scheduling, data validation, claims checks. As governance frameworks like this one give organisations confidence to delegate, the volume of agentic work grows, and so does demand for inference.

Example:. A large insurer maps its decisions onto the matrix. It finds 70% of claims decisions are routine (automate), 20% consequential (AI with managed oversight), and 10% need human judgement. Moving the routine 70% to agents adds millions of model calls a month, demand that is continuous, contract-like and tied to business processes rather than to staff typing prompts.

What lenders should take from this

•Routine-quadrant demand is the most financeable: It is high-volume, repetitive and embedded in operations, closer to utility demand than to discretionary spend.
Governance accelerates adoption: Clear decision rights reduce the risk that organisations pull back after an AI mistake, making demand more durable.
Demand quality varies by quadrant: Exploratory usage is valuable but volatile; routine and consequential usage is stickier. Lenders assessing an AI lab's revenue could ask how much of it comes from embedded, process-level use.

The same matrix applied to AI financing decisions can also guide how financial institutions use AI in their own credit decisions:

Routine: automated monitoring of covenants, utilisation reports and payment schedules.
Consequential: AI-assisted underwriting of standard GPU loans with human sign-off on outcomes.
Exploratory: AI-assisted modelling of residual value curves for new hardware generations.
Strategic: whether to commit billions to a new asset class, humans lead, with AI analysis.

The Core Debate - Willingness Versus Ability

The emerging debate is about the future of AI as a contest between two forces, that is Willingness vs Ability.

Willingness: demand and profitability - Willingness is whether customers want AI enough to pay for it, and whether sellers can do so profitably.

Token prices for frontier capabilities are holding steady or rising.
Model efficiency keeps improving, so each dollar of compute delivers more useful work.
User utility keeps increasing - models are doing more valuable tasks, and agents are taking on whole workflows.
"You can't go back" - once people experience high-quality AI tools, they don't want to work without them.

Example: A legal team adopts an AI drafting assistant. A contract review that took two days now takes two hours. At renewal, the question isn't whether to keep it but whether to upgrade. That is willingness, revenue at the top of the stack that ultimately repays lenders at the bottom.

Ability: financing capacity - whether the capital system can fund the build-out fast enough. Here the episode sees the binding constraint. Nvidia's balance sheet can support a large but finite amount of backstops, an estimated 46 GW. Total incremental capacity required is about 240 GW.

Source: SemiAnalysis - Nvidia's estimated backstop capacity covers only a fraction of incremental capacity required.

Example: If the remaining ~194 GW were deployed at about $40 million per MW, it would imply several trillion dollars of investment that must be financed by someone other than Nvidia like hyperscalers, sovereign funds, infrastructure investors, private credit and public bond and ABS markets.

The Willingness vs Ability

Four possible futures for the AI build-out, depending on willingness and ability.

Credit-constrained boom (high willingness, low ability): Demand is strong but money is scarce. Capacity is rationed, prices stay high, backstops are in heavy demand. The episode suggests this is roughly where we are today.
Sustained build-out (high willingness, high ability): Capital markets learn to underwrite AI on its own merits and the 240 GW gets built.
Overbuild / bust (low willingness, high ability): Easy credit funds capacity that demand doesn't fill, the telecom 2001 pattern.
Stall (low willingness, low ability): Projects are cancelled, and the industry consolidates.

Why IG backstops alone can't close the gap: Only a limited number of investment-grade (IG) companies have the scale and appetite to stand behind tens of billions of dollars of AI commitments, and each is limited by rating agencies, shareholders and its own risk appetite. If they all approach their limits at once, the model stops scaling. Therefore the lending market must evolve beyond pure IG backstops, through:

Lender education: understanding GPU economics, cluster execution risk and token profitability, so lenders can lend against the asset and business, not just the guarantor.
Better residual value models: so lenders can confidently value hardware at the end of a financing.
Risk-tranching structures: so conservative investors buy safe slices and risk-tolerant investors take the rest.
Better tools: independent operator ratings, utilisation data and pricing benchmarks that make AI infrastructure behave like a mature asset class.

Example: In five years, a pension fund analyst might evaluate a GPU-backed bond the way she evaluates an aircraft lease today: checking the operator's execution score, the hardware generation's historical residual curve, the contracted-versus-spot revenue mix and the token-revenue health of end customers. If those tools exist and are trusted, capital can flow without Nvidia signing every guarantee.

What Could Go Wrong - Stress-Testing the Backstop Model

Any honest explanation of a financing system should ask how it behaves under stress. The structures in this section outlines various failure modes.

What Could Go Wrong

Scenario 1: Demand pauses -Enterprise adoption slows for a year, an economic downturn, regulation, or models that don't visibly improve. Spot rental prices fall. Neoclouds with uncontracted capacity feel it first; those with floors lean on guarantors.

Example: Market rates fall from $4/hr to $2.50/hr. A Neocloud with a $3.25 floor calls on the guarantee for $0.75/hr across 10,000 GPUs, that is about $66 million a year that the guarantor must cover. Multiply across many Neoclouds and the cost shows up in the guarantor's results, just as its chip sales are also under pressure.

Scenario 2: A technology leap - A new architecture, a breakthrough in model efficiency, or a rapid move of inference to ASICs makes existing clusters much less valuable. Residual values fall; RVGs are called; junior tranches absorb losses.

Example: A new architecture delivers five times the inference throughput per watt. The expected resale value of a previous-generation fleet drops from 30% of cost to 10%. On a $5 billion fleet, that's a $1 billion swing, absorbed first by equity, then the B-piece, with any RVG stepping in according to its terms.

The irony: efficiency gains strengthen the willingness case but can erode the value of assets already financed.

Scenario 3: An operator stumbles - A Neocloud delivers a cluster late or suffers serious reliability problems. Its customer may claim service credits or terminate. This is the execution risk that explains the ~100 bp gap between Meta's bonds and CoreWeave's project financing.

Example: A 100 MW cluster is four months late because of a cooling defect. The hyperscaler customer is owed several months of service credits. First-year cash falls well short of plan, reserves are drawn, and the Neocloud's next financing is priced wider.

Scenario 4: Credit markets tighten - Rates rise or a broader credit event makes lenders cautious. Sound projects struggle to refinance; new ones are delayed. With AI debt projected to exceed 10% of US issuance in 2026, a pullback would be felt quickly, and rising yields hit AI equities at the same time.

Scenario 5: Concentration: Many of the largest commitments depend on a small number of AI labs and hyperscalers. CoreWeave's 67% revenue concentration with Microsoft in 2025 is one visible example. If a key customer cut its compute plan, multiple lenders, Neoclouds and guarantors would feel it at once.

Example: Three Neoclouds each sign offtake agreements with the same hyperscaler, whose end customer for all three is a single AI lab. On paper, lenders see three projects with an IG offtaker. In reality, all depend on one lab's growth.

Scenario 6: Circularity unwinds: Where suppliers invest in customers who buy their products, a slowdown can travel around the loop: lower demand → lower supplier sales → less appetite for further investment and backstops → harder financing for customers → lower demand. This is the vendor-financing dynamic that hurt telecom equipment makers after 2001.

Why stress testing doesn't contradict the bull case: None of these scenarios means the build-out will fail. Railways, electricity, telecom and the internet all went through financing booms, busts and restructurings and still delivered transformative infrastructure. Financing structures determine who bears the pain when things go wrong and how fast the system recovers. Understanding backstops, tranches and residual value guarantees is how you understand who those people will be

Who bears the pain

Part VI - Bringing It Together

The End-to-End Flow - Financing a 200 MW AI Campus

Here is one end-to-end, illustrative walk-through that ties every concept together.

1Demand appears. An AI lab forecasts it needs 200 MW more inference capacity because enterprise API and agentic revenue is growing quickly. (Eras; willingness; decision matrix.)
The lab can't borrow cheaply. It isn't investment grade. (The IG gap.)
A hyperscaler steps in. The lab signs a multi-year compute deal with an IG hyperscaler, which signs a five-year offtake with a full-stack neocloud. (Credit intermediation.)
The Neocloud needs a site. An infrastructure-led developer offers a refurbished industrial site with an existing grid connection plus behind-the-meter gas. (Neocloud map; power constraint solved by capital.)
The developer needs a creditworthy lease. Nvidia guarantees part of the Neocloud's lease and the PPA for on-site power. (Lease and PPA guarantees.)
The Neocloud buys compute. It orders GB300 racks; memory supply is secured by Nvidia's multi-year commitments. (Supply commitments; de-speccing.)
The financing route is chosen. Part of the hardware is financed with GPU-backed debt on the Neocloud's balance sheet; part goes into a bankruptcy-remote SPV leased to the hyperscaler at a lower cost of capital. (Two funding strategies.)
Lenders size the debt. They underwrite the hyperscaler offtake and Nvidia's support, at a DSCR cushion, charging a spread that still includes execution risk. (Financial model; ClusterMAX risk.)
The capital stack is tranched. Senior A notes, a B-piece and equity, with partial Nvidia residual value support. (Tranching; RVGs.)
Some of it reaches ABS markets. Senior notes are sold to insurers and pension funds, priced off Treasury yields plus a spread. (AI debt as an asset class; bond yields.)
Capacity goes live. The lab serves tokens at high margins; revenue flows down through the hyperscaler to the Neocloud, which pays the developer, the power provider and lenders. (Revenue down the stack.)
Upside is shared. If rental prices exceed any floor, Nvidia shares the excess. (AICP revenue share.)
In year five or six, the hardware is refinanced, re-leased for inference, or sold. If it's worth less than guaranteed, Nvidia makes up the gap. If inference has largely moved to ASICs, that's when it will show. (Residual value; ASIC shift.)

Who carries which risk?

Who Carries the Risk in AI Build

The big picture in one paragraph: AI demand creates revenue at the top of the stack. That revenue is attractive but held by companies lenders don't fully trust. Nvidia and the hyperscalers lend their trust through their balance sheets, so money can flow from capital markets to the builders. It's working today, and profitability across most layers is strong. But the amount of trust available is finite, bond markets are starting to push back, custom chips are changing the long-run value of GPUs, and the build-out ambition is far larger than any single balance sheet. The industry's next challenge is to make AI infrastructure financeable on its own merits

The Big Picture

What It Means for You

For technology sellers and account executives

Capacity and pricing at the top reflect financing at the bottom. If credit tightens, expect slower capacity growth and firmer pricing for premium models.
Vendor stability is a selling point. Customers increasingly ask where AI workloads run and who stands behind them. Being able to explain the Hyperscaler vs Neocloud distinction builds credibility in technical due diligence and third-party risk reviews.
Platform choice is a risk conversation. Solutions on a major hyperscaler inherit that provider's resilience and balance sheet, a legitimate point in regulated sectors like health, government and financial services.
Sell to the decision matrix. Frame AI use cases by quadrant: routine processes to automate, consequential ones to manage with oversight. That helps customers adopt faster and makes the value easier to measure.

Example: An enterprise risk team asks whether a vendor's AI features depend on a single small GPU provider. The vendor explains that inference runs through a major cloud provider's managed AI service with regional failover. That answer, grounded in the dynamics in this article, can close a third-party risk question that might otherwise delay a deal.

For enterprise buyers

Expect AI pricing to reflect real infrastructure economics. Frontier token prices, per the episode, are holding or rising.
Lock in capacity where it matters. If AI underpins critical workflows, consider committed capacity or reserved throughput, just as labs do with offtake.
Ask about concentration risk and exit strategy. Know which clouds and providers your AI vendors depend on, and how portable your workloads are (the control-versus-portability 2×2 in Chapter 2).

For delivery and architecture teams

Efficiency is economic. Caching, batching and right-sizing models drive the high gross margins discussed in Chapter 15, and reduce your own costs.
Design for portability where practical, so capacity or pricing changes don't strand your solution.
Expect hardware diversity. As inference moves partly to ASICs, design so models can run on different accelerators through managed services.

For investors and finance teams

Read the footnotes. Off-balance-sheet commitments, capacity agreements and guarantees can be substantial contingent exposure.
Watch spreads and yields. Hyperscaler and AI-related spreads, and the 10-year Treasury yield relative to the threshold where stocks and bonds stop hedging each other, are early signals.
Watch GPU rental prices and the ASIC share of inference. They are the best real-time indicators of residual value risk.
Use the levers. Price, utilisation, capex and leverage dominate returns; power and rent matter far less for the GPU owner.

Conclusion

The Most Important Chip May Be the Credit Rating

For three years, the AI race looked like a race for silicon, then a race for power. The SemiAnalysis episode makes a compelling case that it has become a race for credit. Nvidia's roughly $497 billion of commitments, the hyperscalers' role as credit intermediaries, AICP rental floors, residual value guarantees, SPV leasebacks and tranched capital structures are all versions of the same move: using strong balance sheets to make weaker ones financeable.

That move has worked remarkably well. Demand looks durable, especially as agentic AI turns compute into an "all time" utility. Token economics look strong. Even older GPUs are earning more than anyone expected. But the picture has real tensions. Neoclouds like CoreWeave are growing fast but carry heavy debt and customer concentration. Nvidia's customers are building their own chips. Bond markets are absorbing an unprecedented wave of AI debt, and rising yields could raise the cost of capital and remove the hedge investors rely on. And the industry's long-term structure is not yet settled (software-like, telecom-like or split).

Nvidia's estimated 46 GW of backstop capacity is a fraction of the 240 GW the industry wants to build. Closing that gap will require lenders to understand AI infrastructure deeply enough to underwrite it on its own merits: GPU economics, cluster execution, token profitability and residual values. Whether that happens fast enough, and whether today's demand proves durable through a full credit cycle, will decide how big and how fast the AI build-out becomes.

For everyone working in technology, it is worth watching not just the next chip launch, but the next bond deal.

References and Further Reading: core figures in this article come from the SemiAnalysis episode summary and the charts supplied with the brief. The further reading below is a starting point for anyone who wants to go deeper. Titles and authors are given so they can be searched directly; please check the original sources for the latest versions and figures before citing them.

Primary sources used in this article

SemiAnalysis Weekly, "Nvidia's Backstop Universe" (episode and summary) — the source of the $11.3T capex, ~$7T debt, ~$497B Nvidia commitments, AICP floor, 46 GW vs 240 GW, and related figures.
SemiAnalysis, ClusterMAX ratings of GPU cloud providers - background on execution risk.
Boring Biz, "How Money and GPUs Move Between NVIDIA and CoreWeave" (graphic, as of 13 October 2026), citing CoreWeave Q2 2026 results, 10-K and 8-Ks, Nvidia 13F and SEC filings, Reuters and Bloomberg.
CoreWeave investor relations - quarterly results, 10-K and 8-K filings.
Nvidia investor relations - annual reports and SEC filings, including notes on purchase obligations and guarantees.
Bloomberg / MacroBond chart, "Stock–bond correlation vs US 10-year yield" (as supplied).
"Market Share Splits 2024 to 2030: Training vs Inference" (chart, as supplied).
Social media post dated 13 October 2026 on hyperscalers scaling custom chips and ASIC volumes surpassing GPUs by 2030 (as supplied; an opinion, not advice).
"AI Compute Funding Strategies: Neocloud/PaaS vs SPV Leaseback" (comparison graphic, as supplied).
"The Neocloud Map"
"Cloud Evolution - Adoption 2008 to 2022" and "Distributed Cloud - Battle AWS vs Google" (slides, as supplied, citing Gartner, HBR, Forrester, McKinsey and BCG).
I. Sebastian, P. Weill, T. Haskamp and J. vom Brocke, "Designing Decision Rights for AI" (MIT Sloan, 2026), source of the AI Decision Matrix.
Nvidia keynote framing of AI eras (training, inference, agentic AI), as shown in the supplied "The AI Revolution" slide

Further reading on AI investment and financing

David Cahn (Sequoia Capital), "AI's $600B Question" (2024) - on the gap between AI infrastructure spend and end-user revenue.
Goldman Sachs Global Investment Research, "Gen AI: Too Much Spend, Too Little Benefit?" (2024) — a debate-style report on AI capex returns.
McKinsey & Company, "The cost of compute: a $7 trillion race to scale data centers" (2025).
Bain & Company, Technology Report 2025 — on revenue needed to fund AI compute demand.
International Energy Agency, "Energy and AI" (2025) — on datacenter power demand.
Rating agency commentary (S&P, Moody's, Fitch) on datacenter ABS, GPU-backed financing and hyperscaler credit

Further reading on infrastructure cycles and telecom

Carlota Perez, Technological Revolutions and Financial Capital (2002) - the classic framework for how financial bubbles accompany technology build-outs.
Andrew Odlyzko, writings on internet traffic growth and the telecom bubble - on how real demand and mistimed financing coexisted.
Analyses of the 2000–2002 telecom bust and vendor financing (e.g. histories of Lucent, Nortel and fibre carriers) on supplier-financed demand.

The $200 Billion AI Capex Shift: How Big Tech Offloads GPU Costs

2026-10-03T16:04:33.140+10:00

The artificial intelligence arms race has forced tech giants and cloud providers into an unprecedented spending spree. Deploying state-of-the-art AI models requires tens of thousands of top-tier GPUs, such as Nvidia’s Grace Blackwell architecture, housed in specialised data centers equipped with immense power and cooling infrastructure.

However, buying this hardware upfront requires billions of dollars in Capital Expenditures (Capex). On Wall Street, massive Capex spikes depress Free Cash Flow (FCF) yields, lower Return on Invested Capital (ROIC), and weigh down balance sheets. Furthermore, because AI hardware evolves at a breakneck speed, today’s $35,000 chip faces severe obsolescence and depreciation risks within 3 to 4 years.

To scale AI infrastructure without alienating investors, two major off-balance-sheet models have emerged: Renting via Neoclouds and Securitising Silicon through SPV Sale-Leasebacks.

Visualising the Framework

The diagram below outlines the two structural pathways tech companies use to offload Capex while securing the massive compute power required for modern AI workloads:

Strategy 1: Renting Compute via Neoclouds (PaaS / On-Demand)

Instead of building massive data centers and buying hardware directly, companies rent GPU capacity on-demand or through multi-year Platform-as-a-Service (PaaS) agreements from specialized Neocloud providers like CoreWeave, Lambda Labs, and Nebius.

How It Works

The Neocloud provider assumes the balance sheet burden: they raise debt, procure physical chips, construct facilities, and handle day-to-day operations. The enterprise simply pays a recurring usage fee for compute capacity.

Key Financial & Operational Benefits

Complete Capex to Opex Conversion: 100% of the hardware and data center procurement costs disappear from the enterprise’s capital expenditure line. Payments are treated strictly as operating expenses (Opex).
Elimination of Hardware Obsolescence Risk: As next-generation hardware architectures arrive (e.g., Rubin following Blackwell), the tenant is not stuck with depreciated, obsolete hardware. They simply finish their contract term and negotiate access to newer clusters.
Accelerated Time-to-Compute: Constructing data centers and securing grid power can take years. Neoclouds allow companies to bypass procurement lead times and access pre-built, active GPU clusters immediately.

Strategy 2: The SPV Sale-Leaseback Model (Securitizing Silicon)

For hyperscalers (like Amazon, Microsoft, or Google) that require custom-designed data centers, proprietary network topologies, and full physical control, renting from third parties is not always ideal. Instead, they use a Special Purpose Vehicle (SPV) structure, treating silicon as an income-generating financial asset much like commercial aircraft, real estate, or shipping containers.

How It Works

Purchase & Transfer: The hyperscaler purchases GPUs, installs them across its data centres, and then transfers legal title of those assets into an independent, bankruptcy-remote SPV.
External Financing: Third-party institutional investors (pension funds, private credit firms) inject equity and debt into the SPV to fund the multi-billion-dollar asset purchase.
The Leaseback: The hyperscaler leases the GPUs back from the SPV for an agreed monthly fee.

Key Financial & Operational Benefits

Preservation of Free Cash Flow (FCF): The upfront cash proceeds received from selling the GPUs to the SPV offset the initial Capex outlay. This prevents multi-billion-dollar FCF dips in single fiscal quarters.
Off-Balance-Sheet Leverage & Credit Protection: The debt issued to finance the chips sits inside the SPV as non-recourse debt. This allows hyperscalers to utilise financial leverage to scale compute without swelling their consolidated corporate debt-to-EBITDA ratios or risking corporate credit rating downgrades.
Retained Operational Control: Unlike third-party cloud renting, the physical hardware remains inside the hyperscaler's own data center, preserving proprietary interconnects, security protocols, and operational workflows.
Shifting Residual Value Risk: Equity investors inside the SPV absorb the financial risk of hardware devaluation at the end of the lease term rather than the parent tech giant.

Executive Strategy Comparison

Metric / Aspect	Strategy 1: Neocloud Renting	Strategy 2: SPV Sale-Leaseback
Physical Location	Hosted in Neocloud facilities	Maintained inside hyperscaler data centers
Balance Sheet Impact	Pure Opex (PaaS usage model)	Off-balance-sheet asset refinancing
Hardware Ownership	Never owned by customer	Owned briefly, sold, and leased back
Primary Financial Goal	Zero capital commitment; maximum flexibility	Smooth FCF trajectories while maintaining physical scale
Ideal User Profile	AI startups, mid-market enterprises, tier-2 tech	Global hyperscalers running $5B+ chip deployments

Conclusion: Financial Engineering Meets Silicon

As annual AI infrastructure budgets cross the $200 billion mark, financial engineering has become as vital as software engineering in maintaining market leadership. By leveraging Neocloud renting for speed and flexibility or SPV sale-leasebacks for balance-sheet-light control, tech companies are proving that you don't need to carry every asset on your balance sheet to dominate the AI landscape.

Atlassian's Evolution: Navigating Market Pressure and the AI Revolution

2025-10-22T08:45:00.002+11:00

The Collaboration Giant at a Crossroads

For over two decades, Atlassian has been the backbone of how software teams work. If you've ever created a Jira ticket, documented something in Confluence, or moved a Trello card, you've touched the Australian software giant's ecosystem. With over 300,000 organisations relying on their tools and $5.2 billion in trailing revenue, Atlassian has become an indispensable infrastructure for modern knowledge work.

But here's the uncomfortable truth: despite this dominance, the stock has been hammered. From its October 2021 peak of $458.13, shares have plummeted to around $150—a gut-wrenching 67% decline. Year-to-date in 2025 alone, it's down 37.3%. Investors are concerned about the sustainability of growth, profitability timelines, and whether Atlassian can withstand an increasingly competitive landscape in an AI-transformed world.

So what's really going on? Let's dig into Atlassian's product evolution, revenue composition, strategic bets, and how artificial intelligence is reshaping everything.

1. Understanding the Product Ecosystem

Atlassian's real power isn't in any single product—it's in how all its products work together. Think of it as a connected web of tools spanning software development, project management, IT operations, and team collaboration.

1.1 The Revenue Powerhouses

Jira: The Crown Jewel

Jira isn't just a product; it's practically a verb at this point. "Let me Jira that" has become workplace shorthand for tracking work. The Jira family (Software, Service Management, Work Management, Align) dominates agile project management and IT service management, driving the core of Atlassian's $4.93 billion in subscription revenue, which represents 95% of total revenue in FY2025.
It's deeply integrated with everything: Bitbucket, GitHub, Confluence, and countless third-party dev tools. For many software teams, Jira is simply the industry standard. You might not love it, but good luck convincing your engineering department to switch.

Confluence: The Documentation Hub

Confluence is where institutional knowledge lives. Wiki-style documentation, real-time collaboration, meeting notes, project plans—it's all there. While Atlassian doesn't break out exact numbers, Confluence represents a substantial chunk of revenue and is often bundled with Jira. That bundling is strategic: once you're documenting in Confluence and tracking in Jira, you're locked into the ecosystem.

1.2 The Solid Contributors

Trello: The People's Project Manager

Remember when Trello boards were everywhere? The visual, drag-and-drop interface made project management accessible to non-technical teams. It still generates steady revenue with its high user base, though the average revenue per user is lower than Jira's. The criticism? While competitors innovated, Trello felt stuck in time. Atlassian is now racing to inject AI features to revitalise it.

Bitbucket: Fighting an Uphill Battle

Bitbucket is Atlassian's Git repository hosting service for developers. It makes steady money, but let's be honest—it's fighting a losing battle against GitHub (owned by Microsoft) and GitLab, both of which have stronger developer communities and more modern features.

1.3 The Strategic Investments

Opsgenie & Statuspage

These are niche but valuable: Opsgenie handles incident management and alerting, while Statuspage communicates status updates. They're not huge revenue drivers, but they strengthen Atlassian's IT service management story.

Atlas & Compass

These newer products focus on team alignment and developer experience. They're essentially Atlassian's bet that the future of work requires better visibility into who's doing what and how software components are performing. Right now, they're strategic investments rather than revenue engines.

Advanced Roadmaps

This premium add-on (formerly Portfolio for Jira) helps enterprises do strategic planning and portfolio management. It's part of why big companies pay more for Jira—they need to see the forest, not just the trees.

1.5 How It All Connects

The magic—and the trap—of Atlassian is integration. Here's how it flows:

Code gets written in Bitbucket → tracked in Jira → documented in Confluence → incidents monitored via Opsgenie → status communicated through Statuspage.
Atlassian calls this the "Teamwork Graph"—a proprietary data model connecting billions of relationships across all their products. It creates genuine network effects: the more products you use, the stickier the platform becomes. This is both Atlassian's moat and its potential weakness. Customers appreciate the integration but resent feeling locked in, especially when products don't evolve fast enough.

2. The Profitability Paradox

Here's where things get interesting. Atlassian's financials look both impressive and concerning, depending on what you focus on:

The Good:

Gross margins of 84-85% (software economics at their finest)
Subscription revenue of $4.93B, growing 30% year-over-year
Free cash flow of $1.42B

The Confusing:

Still GAAP unprofitable

How can a company with 84% margins be unprofitable? Two big reasons:

First, Atlassian spends 51% of revenue on R&D, heavily focused on AI (specifically Rovo) and platform expansion. Second, stock-based compensation eats up 26% of revenue. Add in cloud migration incentives and new product development, and you get a company that's investing for the future rather than optimising for today's profits.

2.1 Revenue Breakdown (Estimated)

While Atlassian doesn't publish product-level numbers, industry analysts estimate:

Jira family: 60-65% of total revenue
Confluence: 20-25%
Trello: 5-8%
Everything else: 5-10%

This concentration is both a strength (proven core products) and a vulnerability (heavy dependency on Jira/Confluence).

3. Why the Stock Has Cratered

Let's address the elephant in the room: that 67% decline from peak to trough.

3.1 The Numbers Are Brutal:

Peak (Oct 2021): $458.13
Current (Oct 2025): ~$150
52-week range: $144.32 - $326.00
YTD 2025: Down 37.3%

3.2 What's Driving This?

Growth Questions: 30% subscription growth is objectively strong, but investors worry about sustaining it at $5B+ scale. Where's the next wave of growth coming from?
The Profitability Timeline: Post-2021, investors stopped rewarding "growth at all costs." They want to see a path to sustainable GAAP profits.
Cloud Migration Backlash: Atlassian's forced migration from Server to Cloud created genuine customer frustration. Some felt rushed, some encountered issues, and nobody likes being forced into anything.
Rising Competition: New entrants like Linear and Notion are peeling off customer segments with superior user experiences.
Macro Reality: Higher interest rates compressed valuations for all unprofitable growth software companies. Atlassian wasn't uniquely punished, but it wasn't spared either.

3.3 The Contrarian View:

Interestingly, some analysts argue the market has overreacted.

Simply Wall St's discounted cash flow analysis suggests Atlassian is undervalued by 45.6%, with a P/S ratio of 7.67x versus a "fair" 13.52x. If they're right, this could be a significant buying opportunity.

4. Strategic Bets: Loom and Guard

4.1 Loom: The $975 Million Async Video Bet

In 2023, Atlassian acquired Loom for $975 million—a significant premium for an async video messaging platform. At the time, Loom had 25+ million users, 200,000 customers, and was recording nearly 5 million videos monthly.

The Strategic Logic: Remote and hybrid work created structural demand for asynchronous communication. Text and screenshots can only convey so much; video adds the human context—tone, emotion, nuance—that makes remote work feel less remote. Atlassian saw a product gap in its collaboration suite and a chance to make Jira and Confluence stickier by embedding video everywhere.
The Reality Check: Loom's peak valuation was $1.53B in 2021, so Atlassian got a relative bargain. But has it paid off? Loom now contributes about 1.5 percentage points to cloud revenue growth—it's growing, but it's still relatively small. Integration continues, with AI features like automated transcripts and summaries rolling out.

The concerning part? Two Loom co-founders (Vinay Hiremath and Shahed Khan) left in early 2025 to work on Elon Musk's DOGE project. Co-founder departures post-acquisition always raise questions about cultural fit and retention.

4.2 Guard: The Security Play

Guard isn't an acquisition—it's a product launch from 2024 that deserves attention. It's a comprehensive cloud security solution unifying existing capabilities (Atlassian Access and Beacon) with new data loss prevention features.

Why It Matters: Large enterprises have sophisticated security requirements. If Atlassian wants to convince them to migrate from on-premises to the cloud, it needs robust security controls. Guard Premium addresses this with data classification, threat detection, investigation tools, and extended audit logs.
It's also a revenue opportunity. Security products command premium pricing, and every major cloud vendor (Microsoft, Google) offers advanced security tooling. Atlassian had to play in this space to remain competitive.

5. The Competitive Gauntlet

Atlassian fights battles on multiple fronts, and the threats vary by product.

5.1 Jira's Competition: The Old Guard Gets Challenged

Who's Competing:

Monday.com and Asana own the non-technical, business-team market with friendlier interfaces
Linear is the rising star among developers, praised for speed and simplicity
Azure DevOps is Microsoft's integrated answer, bundled with Azure
ClickUp positions itself as the all-in-one replacement

The Threat:

Developers increasingly complain that Jira is powerful but bloated with a steep learning curve. Linear's clean UX is winning hearts and minds. Meanwhile, Microsoft's bundling strategy with Azure creates ecosystem lock-in that's hard to break.

5.2 Confluence's Challenge: Notion Changed Everything

Who's Competing:

Notion completely changed expectations for what documentation tools should look like
Nuclino explicitly markets itself as "Confluence without the bloat"
Microsoft SharePoint is the boring enterprise standard, often "free" with Office 365
GitBook, Slite, Coda, and others target specific niches

The Threat:

Confluence feels dated. Search is slow. Pricing gets expensive at scale. And Notion showed that documentation tools can be beautiful, fast, and intuitive. For startups and smaller teams, the choice is increasingly clear: use the modern tool.

5.3 The Real Strategic Risk

Individual competitors aren't the biggest problem—it's ecosystem fragmentation. Companies like Notion, Linear, and various open-source projects are unbundling Atlassian's suite, offering "best-of-breed" experiences. Why settle for Atlassian's version when you can pick the best tool for each job?
On the flip side, Microsoft offers a counter-narrative: a fully integrated stack (Azure DevOps, Teams, SharePoint, GitHub) all backed by existing Office 365 relationships and bundled pricing.

Atlassian is caught in the middle: attacked from below by nimble startups and from above by Microsoft's bundling power.

6. The AI Revolution: All-In on Rovo

If there's one thing Atlassian is betting the farm on, it's AI. And the centrepiece is Rovo.

6.1 What Is Rovo?

Introduced in May 2024 and made generally available in October 2024, Rovo is Atlassian's AI platform built on three pillars:

Rovo Search: Unified search across all Atlassian products plus 80+ third-party SaaS tools (Slack, Google Drive, GitHub, Figma, etc.). It's powered by the Teamwork Graph and respects permissions—you only see what you're authorised to access. The promise: eliminate information silos.
Rovo Chat: An AI assistant that finds answers and triggers actions. Need to create a Jira issue, summarise a document, or post to Slack? Rovo Chat can do it. It understands your company's jargon and acronyms because it's trained on your organisation's data.
Rovo Studio: A low-code/no-code platform for building custom AI agents. Anyone can create agents using natural language—no coding required. There are ~20 pre-built agents (release notes writer, bug report assistant, OKR generator), and teams can build custom ones for their specific workflows.

6.2 The Bold Move: Making It Free

Here's where things get wild. In April 2025, Atlassian made Rovo free for all Standard, Premium, and Enterprise cloud customers. They abandoned the initial $20/user/month pricing model.

Why would they give it away?

Remove adoption barriers: Teams can experiment without budget approval
Drive consumption: Build dependency now, monetise usage-based pricing later
Create lock-in: Once AI is embedded in workflows, switching becomes exponentially harder
Match Microsoft: Copilot is bundled with Microsoft 365; Atlassian had to respond

This is a massive bet. At $20/user, Rovo could have generated hundreds of millions in revenue. Instead, Atlassian is sacrificing short-term revenue for long-term strategic positioning.

6.3 Is It Working?

Early signals are promising:

Atlassian's own developers report saving 1-2 hours per week just from having internal docs accessible via Rovo—more than they save from AI code generation tools
Average teams report saving 45 minutes per week using Atlassian Intelligence features
Over 30,000 organisations are now using Atlassian Intelligence

But the real question remains: is Rovo differentiated enough to matter? Every competitor ships AI features now. Notion has Notion AI, Linear has automation, and GitHub has Copilot.

Atlassian argues that the Teamwork Graph gives Rovo unique context—it "knows" your organisation in ways generic LLMs don't. Whether that advantage is defensible or just a temporary head start will determine Rovo's success.

7. AI's Broader Impact on the Business

7.1 Product Enhancement Everywhere

AI is being woven throughout the portfolio:

Jira: AI breaks epics into issues, generates code plans, creates pull requests, and automates backlog cleanup
Confluence: AI summarises documents, generates content, and suggests related pages
Jira Service Management: AI helps support agents respond faster and spot incident patterns
Loom: AI creates transcripts, summaries, chapters, and action items from videos

7.2 The Cost of the AI Arms Race

Heavy AI investment explains some profitability challenges:

That 51% R&D spend is heavily focused on AI capabilities
LLM costs from OpenAI partnerships (and potentially Atlassian-hosted models)
Competing for AI talent in a brutally competitive market

7.3 Business Model Evolution

The shift to free Rovo with potential future usage-based pricing hints at a larger transformation. Atlassian may eventually charge based on AI usage, data connections, or automation executions—following companies like Snowflake and Databricks toward consumption-based models.

8. What's Next: Challenges and Opportunities

8.1 The Headwinds

Market Saturation: With 300,000+ customers, where's the next wave of growth? It has to come from expansion and upsell.
Cloud Migration Hangover: Customer resentment from forced migrations lingers. Trust takes time to rebuild.
Show Me the Profits: Investors want GAAP profitability, not just adjusted EBITDA promises.
Product Bloat: Jira and Confluence are criticised as slow and cluttered. Speed matters.
Talent and Culture: Loom co-founder departures highlight integration challenges that could affect other acquisitions.

8.2 The Tailwinds

AI Differentiation Potential: If Rovo delivers unique value from the Teamwork Graph, it could create defensible competitive moats.
Enterprise Momentum: Deals over $1M in annual contract value are growing rapidly—the enterprise market still believes.
Network Effects: As more products integrate, switching costs increase exponentially.
Cash Flow Strength: $1.42B in free cash flow funds innovation without dilution.
Potential Undervaluation: If analysts are right about 45% undervaluation, there's significant upside.

9. The Verdict: Evolution or Disruption?

Atlassian stands at a genuine inflection point. The company that disrupted enterprise software with product-led growth now faces its own disruption—from nimbler competitors, from Microsoft's bundling power, and from AI's rapid commoditisation of knowledge work.

The twin bets on Loom and Rovo represent a clear thesis: the future of work is more visual, more asynchronous, more automated, and more AI-assisted. Whether these bets pay off depends on execution, timing, and whether customers believe Atlassian's integrated platform is genuinely superior to best-of-breed alternatives.

My take? The 67% stock decline feels overdone relative to fundamentals. Atlassian still has dominant market positions, incredible margins, solid growth, and strong cash flow. The real question isn't survival—it's whether the company can recapture growth momentum and convert it into sustainable profitability.

The company is making the right strategic moves: going all-in on AI, strengthening security, and adapting to changing work patterns. But in a market where customers have more choices than ever, Atlassian must prove that integrated, AI-powered teamwork is worth the premium—and the lock-in.

The next 18-24 months are decisive. If Rovo adoption accelerates and delivers measurable productivity gains, Atlassian emerges stronger with a defensible AI advantage. If differentiation fails to materialise, the company risks becoming just another collaboration tool in an increasingly crowded market.

For investors, this is a classic "show me" moment. For customers, it's a chance to see whether Atlassian can innovate as fast as the market is moving. And for employees, it's a reminder that past success doesn't guarantee future dominance.

One thing is certain: the Atlassian of 2025 will look dramatically different from the Atlassian of 2020. Whether that evolution leads to renewed dominance or gradual disruption is still being written—and the next few chapters will be fascinating to watch.