Sign in Subscribe
8 min read

The $800B AI Infrastructure Crisis: How OpenAI, Oracle, and CoreWeave Are Shifting CapEx Risk

The $800B AI Infrastructure Crisis: How OpenAI, Oracle, and CoreWeave Are Shifting CapEx Risk

THE GREAT CAPEX SHIFT: HOW AI COMPANIES ARE OFFLOADING BILLIONS ONTO ORACLE, COREWEAVE, AND PRIVATE EQUITY

OpenAI announced in July 2025 it would pay Oracle $300 billion over five years for data center capacity—$60 billion per year. There's just one problem: OpenAI is projected to generate $13 billion in revenue this year and lose $9 billion doing it.

So how will a company bleeding nearly $10 billion annually commit to spending five times its entire revenue on infrastructure?

The answer: It won't. Oracle will.

What we're witnessing isn't AI companies building infrastructure—it's AI companies systematically offloading capital expenditure obligations onto cloud providers, equipment leasing firms, and private equity, creating a complex web of interdependent financing that obscures who's actually footing the bill for the AI boom.

And the trend is accelerating dramatically in 2025.

THE $300 BILLION ORACLE DEAL: WHO'S REALLY PAYING?

In July 2025, OpenAI agreed to a five-year deal with Oracle valued at more than $300 billion. Oracle previously sold $24.5 billion worth of cloud services to all customers combined in its fiscal 2025 TechCrunch—meaning OpenAI's single contract represents more than the entire company's prior annual cloud revenue.

OpenAI had a $5 billion net loss in 2024. The company is on pace to generate $13 billion in revenue this year CNBC. Yet it's committing to $60 billion annually for Oracle infrastructure.

The math doesn't work. Unless someone else is funding it.

KeyBanc Capital Markets estimates that Oracle may need to borrow roughly $100 billion over the next four years—about $25 billion a year—to build the datacenters required for the OpenAI contract. Oracle has around $10 billion in cash on hand and $9 billion of debt due within a year. Oracle's free cash flow has declined 152% year-over-year due to massive capex increases, with the company spending $8.5 billion in Q1 2026, up from $2.3 billion a year earlier.

Oracle already had $82.2 billion in long-term debt as of August 31, 2025, and issued $18 billion worth of bonds in September to fund its AI expansion plans The Register.

Translation: OpenAI commits to infrastructure spending it cannot afford. Oracle borrows $100 billion to build that infrastructure. OpenAI pays Oracle over time using funding from... investors who are betting on future AI monetization.

The capital expenditure has been shifted from the AI company's balance sheet to the cloud provider's balance sheet. But the risk remains: if OpenAI cannot generate sufficient revenue, Oracle is left holding $100 billion in debt and data centers full of depreciating GPUs.

MICROSOFT'S QUIET RETREAT—AND OPENAI'S QUICK PICKUP

While Oracle takes on debt for OpenAI, Microsoft is quietly redistri

buting its AI infrastructure commitments.

In March 2025, Microsoft chose not to exercise a nearly $12 billion option to buy more data center capacity from CoreWeave. OpenAI snapped up the contract within days Semafor.

TD Cowen analysts reported in February 2025 that Microsoft had withdrawn from around 200MW of data center leasing agreements. Weeks later, this expanded to 2GW of data center projects across the US and Europe. By April, reports suggested the pullback had expanded to the APAC region Data Center Dynamics.

Microsoft maintains it's "still committed to spending $80 billion" on AI infrastructure in fiscal 2025. But the company acknowledged it "may strategically pace or adjust our infrastructure in some areas" CNBC.

What's actually happening: Microsoft is shifting OpenAI's infrastructure needs from direct Microsoft builds to third-party providers like CoreWeave and Oracle—providers that OpenAI is now contracting with directly.

Microsoft had $94.8 billion in finance leases that had not been completed as of March 31, 2025, down from nearly $109 billion two quarters earlier.

That's $14+ billion in lease commitments Microsoft walked away from in six months—and much of that capacity is being picked up by OpenAI through other providers.

The CapEx hasn't disappeared. It's been transferred.

COREWEAVE: THE MIDDLEMAN ABSORBING BILLIONS

CoreWeave has become the poster child for this capital expenditure shell game.

In March 2025, CoreWeave announced an $11.9 billion infrastructure deal with OpenAI. By May, that deal had increased to $22.4 billion nearly doubling in two months.

CoreWeave is also providing $14.2 billion in AI infrastructure to Meta through separate agreements.

The company's financial profile reveals the precariousness of this model:

  • CoreWeave reported approximately $2 billion in revenue in 2024 but lost $863 million
  • Q1 2025 revenue: $982 million. Net loss: $314.6 million
  • Total debt as of Q1 2025: $9 billion. Cash on hand: ~$1.3 billion
  • Capital expenditure forecast for 2025: $23 billion—against projected revenue of around $5 billion

CoreWeave is spending nearly 5x its annual revenue on capital expenditures, financed almost entirely through debt and equity raises.

Just two companies drove 77% of CoreWeave's revenues in 2024, with Microsoft accounting for 62% MLQ.

CoreWeave's model: borrow billions to buy NVIDIA GPUs, use those GPUs as collateral for more debt, lease the capacity to hyperscalers, use the lease payments to service debt and buy more GPUs.

It works as long as demand remains insatiable and GPU values hold. If either assumption breaks, the entire structure collapses.

META'S $27 BILLION PRIVATE EQUITY PLAY

Even Meta—one of the most profitable tech companies globally, with $164.5 billion in revenue in 2024 and free cash flow of $52.10 billion The American Prospect—is offloading infrastructure financing.

In October 2025, Meta formed a joint venture with Blue Owl Capital for the $27 billion Hyperion data center in Louisiana. Blue Owl owns 80% of the joint venture, Meta retains 20%. Blue Owl contributed approximately $7 billion in cash, and Meta received a one-time distribution of approximately $3 billion PR Newswire.

A portion of capital raised by Blue Owl will be funded by debt issued to PIMCO and select other bond investors through a private securities offering The American Bazaar.

Meta—a company that could easily self-finance—chose to bring in private equity to fund 80% of a massive data center build.

Why? To shift capital expenditure off its balance sheet and onto private investors, while maintaining operational control through lease agreements.

Meta entered into operating lease agreements with the joint venture for use of all facilities once construction is complete. These lease agreements will have a four-year initial term with options to extend.

The pattern is clear: even cash-rich companies are using financial engineering to distribute infrastructure costs across external capital sources.

THE $800 BILLION SHORTFALL NO ONE'S TALKING ABOUT

Here's the systemic problem underlying all of this capital shuffling:

According to Bain & Company's 2025 Global Technology Report, by 2030, AI companies will need $2 trillion in combined annual revenue to fund the computing power required to meet projected demand. Building the data centers needed would require about $500 billion of capital investment each year.

Even if companies shifted all of their on-premise IT budgets to cloud and reinvested AI-related savings into capital spending on new data centers, the amount would still fall $800 billion short of the revenue needed to fund the full investment.

David Crawford, chairman of Bain's Global Technology Practice, stated: "By 2030, technology executives will be faced with the challenge of deploying about $500 billion in capital expenditures and finding about $2 trillion in new revenue to profitably meet demand".

Let that sink in: Even with aggressive assumptions about AI adoption and cost savings, the industry faces an $800 billion annual funding gap by 2030.

Bain's analysis estimates global incremental AI compute requirements could reach 200 gigawatts by 2030 equivalent to the power consumption of several major cities.

THE ACCELERATION IN 2025

What's most alarming isn't the absolute numbers—it's how rapidly this capital shifting has accelerated in the past 9-12 months.

Q3-Q4 2024:

  • NVIDIA invests $100M in CoreWeave (April 2023 baseline)
  • Early CoreWeave-Microsoft partnerships establish the leasing model

Q1 2025:

  • Microsoft begins pulling back on direct data center commitments (February)
  • CoreWeave announces $11.9B OpenAI deal (March)
  • Microsoft walks away from $12B CoreWeave option; OpenAI takes it (March)

Q2 2025:

  • CoreWeave increases OpenAI commitment to $22.4B—doubling in 2 months (May)
  • Oracle reveals $30B/year deal with OpenAI (July)

Q3-Q4 2025:

  • NVIDIA announces $100B investment in OpenAI (September)
  • OpenAI + Oracle formalize $300B five-year agreement (September)
  • Meta + Blue Owl announce $27B Hyperion data center (October)

The CapEx shifting isn't a gradual trend—it's exponentially accelerating as AI companies realize they cannot self-finance the infrastructure their ambitious plans require.

WHY THIS MATTERS

This isn't normal infrastructure financing. This is a systemic redistribution of capital expenditure risk across an increasingly interconnected web of companies, investors, and lenders—all betting on AI monetization that hasn't materialized yet.

Consider the chain of dependencies:

  1. OpenAI commits $300B to Oracle over 5 years despite generating $13B in annual revenue
  2. Oracle borrows $100B to build infrastructure for OpenAI
  3. CoreWeave borrows $10.6B using NVIDIA GPUs as collateral to lease capacity to OpenAI and Microsoft
  4. NVIDIA invests $100B in OpenAI, which uses that money to buy NVIDIA GPUs through Oracle and CoreWeave
  5. Meta offloads $27B in infrastructure costs to Blue Owl/PIMCO debt investors

Each link depends on the next. Each assumes AI demand justifies the spending. And each is leveraged far beyond what traditional infrastructure financing would allow.

According to American Prospect's analysis, OpenAI is on pace to bring in $12.7 billion this year, expects to lose $9 billion, predicts losses will swell to $47 billion by 2028, and doesn't expect to break even until 2029 The American Prospect.

Yet it's committing to infrastructure spending that would bankrupt it five times over.

The only way this works is if someone else pays—and right now, that "someone else" is Oracle (borrowing $100B), CoreWeave (borrowing $10.6B), and private equity investors (funding Meta's $27B data center).

THE WARNING SIGNS

Several indicators suggest this capital shifting strategy is reaching its limits:

1. Moody's Concern: Ratings firm Moody's expressed concern over the Oracle/OpenAI deal and its financial feasibility, noting Oracle's debt is just as concerning as the "counterparty risk" that OpenAI might not be able to pay its bills The Register.

2. Microsoft's Retreat: Microsoft—OpenAI's biggest backer—is pulling back from direct infrastructure commitments and forcing OpenAI to source capacity elsewhere. That's not confidence.

3. The Bain Gap: An $800 billion annual shortfall by 2030 means current financing models are structurally insufficient even under optimistic scenarios.

4. CoreWeave's Fragility: A company with $2B in revenue, $863M in losses, and $23B in annual CapEx is one demand shock away from insolvency.

THE BOTTOM LINE

The AI infrastructure boom isn't being funded by AI companies. It's being funded by:

AI companies have successfully offloaded capital expenditure obligations onto a daisy chain of intermediaries, each more leveraged than the last.

It's financial engineering masquerading as infrastructure investment.

And the acceleration in 2025 suggests these companies know something the market doesn't: they cannot self-finance the future they're promising investors.

When Bain projects an $800 billion funding gap, when Moody's questions Oracle's ability to get paid, when Microsoft quietly backs away from direct commitments—these aren't isolated concerns.

They're warning signs that the CapEx shifting game is approaching endgame.

The question isn't whether AI infrastructure is necessary. It's whether the financing mechanisms propping it up can survive when revenue growth fails to keep pace with debt service requirements.

History suggests they won't.

Welcome to Nardaggio. Let's separate signal from noise.

Disclaimer

This content is for informational and educational purposes only. Nothing in this article constitutes financial, investment, legal, or tax advice.

I am not a registered financial advisor. I do not know your financial situation, risk tolerance, investment objectives, or time horizon. The strategies and positions discussed here reflect my personal approach and should not be interpreted as recommendations for you to follow.

Markets are inherently risky. Trading futures, options, and leveraged instruments can result in significant losses, including losses exceeding your initial investment. Past performance—mine or anyone else's—does not guarantee future results.

Do your own research. Consult with a qualified financial advisor before making any investment decisions. Any action you take based on information in this article is strictly at your own risk.

CODERA LAB LTD and the author assume no liability for any losses or damages resulting from the use of this information.

Published by: