26 Oct 2025 7 min read

NVIDIA's $110B AI Bubble: The Circular Financing Time Bomb

NVIDIA'S $110 BILLION CIRCULAR FINANCING MACHINE: THE AI BUBBLE'S TICKING TIME BOMB

NVIDIA isn't just selling chips anymore. It's financing the entire AI infrastructure boom by investing billions into the very companies that buy its GPUs—then watching that money flow right back as revenue. With $110 billion in direct investments and another $15+ billion in GPU-backed debt, the semiconductor giant has built a self-reinforcing financial loop that looks increasingly like the vendor financing schemes that collapsed during the dot-com crash.

Here's what's happening beneath the surface of AI's explosive growth—and why it should terrify anyone paying attention to market fundamentals.

THE CIRCULAR MACHINE: HOW IT WORKS

The mechanism is elegant in its simplicity and alarming in its scale.

Step 1: NVIDIA invests in AI companies

In 2024, NVIDIA poured approximately $1 billion into AI startups globally, either directly or through its venture capital arm NVentures. But the real firepower came in 2025:

September 2025: NVIDIA announces up to $100 billion in investments into OpenAI, structured as 10 tranches of $10 billion each, tied to infrastructure deployment milestones
October 2024: NVIDIA had already participated in OpenAI's $6.6 billion funding round
April 2023: $100 million investment in CoreWeave when the company was valued at around $2 billion

Step 2: Those companies buy NVIDIA GPUs

Here's where it gets interesting. NewStreet Research estimates that for every $10 billion NVIDIA invests in OpenAI, it will see $35 billion worth of GPU purchases or lease payments—an amount equal to about 27% of its annual revenues.

OpenAI CFO Sarah Friar confirmed the obvious: "Most of the money will go back to NVIDIA."

The numbers bear this out:

CoreWeave has purchased at least 250,000 NVIDIA GPUs—mostly H100 Hopper models at roughly $30,000 each—spending approximately $7.5 billion buying these chips from NVIDIA
Lambda signed a $1.5 billion deal with NVIDIA for an 18,000 GPU server lease-back over four years

Translation: Essentially all the money NVIDIA invested in CoreWeave has returned to it as revenue.

Step 3: NVIDIA backstops the risk

In October 2025, NVIDIA signed a $6.3 billion agreement to purchase any cloud capacity that CoreWeave can't sell to others. This isn't venture capital—it's vendor financing with a safety net.

NVIDIA's total exposure: $110 billion in direct investments plus $15+ billion in GPU-backed debt guarantees.

That's 67% of NVIDIA's $165 billion in last-twelve-months revenue. For context: Lucent Technologies, during the telecom bubble peak of 1999-2000, had vendor financing commitments of $8.1 billion—24% of its $33.6 billion revenue. NVIDIA's exposure is 2.8 times larger relative to revenue.

THE GPU-AS-COLLATERAL GAME

If the circular investments weren't concerning enough, a parallel financial innovation has emerged: AI startups are now using NVIDIA GPUs as collateral for massive loans.

CoreWeave pioneered this model:

August 2023: Secured $2.3 billion in debt financing led by Magnetar Capital and Blackstone, using NVIDIA H100 GPUs as collateral
May 2024: Raised another $7.5 billion in debt led by Blackstone
Total debt load: $10.6 billion
Annual interest payments on the Blackstone/Magnetar $2.3 billion loan alone: $892 million
Interest rates: 11-15%

Those aren't venture capital rates. Those are distressed debt rates.

The model is spreading:

London-based Fluidstack received approval from Macquarie and other lending institutions to borrow over $10 billion using NVIDIA GPUs as collateral. The total GPU-backed debt market now exceeds $20 billion.

The pitch to lenders is straightforward: NVIDIA GPUs are scarce, in-demand assets that hold their value. The problem? CoreWeave itself estimates that the effective lifespan of a chip is six years, and NVIDIA's accelerated product development cycle could lead to rapid depreciation.

Due to this depreciation risk, current investors are demanding double-digit interest rates—rates that exceed most high-yield bond markets.

Here's the circular logic at its finest:

Startups buy NVIDIA GPUs with investor money (including NVIDIA's own investments)
Those GPUs serve as collateral for loans at 11-15% interest
The loan proceeds buy more NVIDIA GPUs
NVIDIA recognizes revenue from both the original purchase and the debt-funded purchase
If the GPUs depreciate faster than expected, lenders take losses—but NVIDIA has already booked the revenue

It's a financial perpetual motion machine. Until it isn't.

THE CONCENTRATION RISK NO ONE'S TALKING ABOUT

NVIDIA doesn't just have massive vendor financing exposure. It has extreme customer concentration.

39% of NVIDIA's revenue comes from just two customers. 46% comes from four customers.

During the telecom bubble, Lucent's top two customers—AT&T at 10% and Verizon at 13%—accounted for 23% of revenue. NVIDIA's concentration is nearly double that.

And here's the kicker: 88% of NVIDIA's revenue comes from data centers.

This isn't a diversified business model. It's a single-point-of-failure waiting to happen.

The new $10+ billion GPU-backed debt market is built on the assumption that GPUs will hold their value over 4-6 years. But what happens when:

Hyperscalers build their own custom chips (Amazon's Trainium/Inferentia, Google's TPUs, Meta's MTIA)?
Model efficiency improvements reduce GPU requirements by 10-25x?
Inference workloads (which require far fewer GPUs than training) become the dominant use case?

All three are already happening.

COREWEAVE: THE CANARY IN THE COAL MINE

CoreWeave's financials reveal the precariousness of this entire ecosystem.

The company claims a revenue backlog of $25.9 billion as of March 31, 2025. Impressive—until you realize it's only contractually obligated to receive $14.7 billion of that.

According to JP Morgan analyst Mark Murphy, just two companies drove 77% of CoreWeave's revenues in 2024, with Microsoft accounting for 62%.

CoreWeave is essentially a Microsoft subsidiary masquerading as an independent cloud provider—financed by Blackstone debt, built on NVIDIA chips, and propped up by NVIDIA's $6.3 billion backstop guarantee.

In May 2025, DA Davidson analyst Gil Luria downgraded CoreWeave to Underperform, citing its capital structure as "very risky." The company's capital expenditure forecast for 2025: $23 billion—against projected revenue of around $5 billion.

CoreWeave reported $9 billion in current and non-current debt as of its IPO filing. Interest expenses jumped roughly 550% to $264 million in Q1 2025 from the prior year.

Short-seller Kerrisdale Capital called CoreWeave a "debt-fueled GPU rental business" engaged in a "circle jerk with NVIDIA," highlighting the "round-trip arrangement" where NVIDIA was CoreWeave's second-largest customer in 2023, accounting for 15% of total revenue.

Translation: NVIDIA invests in CoreWeave, which buys NVIDIA GPUs, which it rents back to NVIDIA. The money goes in a circle.

THE BANK OF ENGLAND WARNING

This isn't just speculation from bearish analysts. Central banks are noticing.

The Bank of England has warned of a dot-com-era risk of a "sharp market correction" in AI-linked equities, according to reports from the Financial Times.

When central banks start using language like "sharp market correction," it's time to pay attention. And it's not an isolated warning—yesterday I detailed how both France and the United States received unprecedented sovereign credit downgrades https://nardaggio.com/two-sovereign-downgrades-over-the-weekend-will-monday-be-different/ , signaling that institutional risk monitors are seeing cracks across the entire financial system.

The AI circular financing machine is just one more fault line in a system showing increasing stress.

WHY THIS MATTERS NOW

The AI infrastructure buildout is real. The demand for compute is real. But the financing mechanisms underlying this boom are starting to look disturbingly familiar to anyone who lived through the dot-com collapse.

The parallels are striking:

Dot-com era:

Telecom equipment makers (Nortel, Lucent, Cisco) lent money to startups and telecom companies to purchase their equipment
Just before the bubble burst in 2001, the amount of financing Cisco and Nortel had extended to their customers exceeded 10% of annual revenues
When the music stopped, those loans became worthless, and the vendors took massive write-downs

Today:

NVIDIA finances AI startups to purchase its GPUs (vendor financing: 67% of revenue)
Those startups use the GPUs as collateral for additional debt at 11-15% interest rates (desperation rates)
The loans are predicated on GPUs maintaining value over 4-6 years (questionable assumption)
Customer concentration is nearly double what Lucent experienced

The key difference: In the telecom bubble, the fiber and infrastructure had real, lasting value. The networks still carried data for decades after.

GPUs depreciate. Fast. NVIDIA CEO Jensen Huang himself said in March 2025: "When Blackwell starts shipping in volume, you couldn't give Hoppers away."

That's the CEO of NVIDIA telling you that last-generation chips become worthless overnight. Now imagine you're a lender who just used those Hoppers as collateral for a $10 billion loan at 11% interest.

THE SYSTEMIC RISK

The problem isn't that NVIDIA is engaging in circular financing. Vendor financing is common and sometimes necessary to seed new markets.

The problem is the scale, concentration, and opacity.

NVIDIA has struck numerous investment and financing deals, many of which are too small individually for the company to consider "material" and report in its financial filings—yet collectively they may be significant.

There are so many interlocking rings of circularity—where NVIDIA has invested in a company like OpenAI, which in turn purchases services from a cloud service provider that NVIDIA has also invested in, which then also buys or leases GPUs from NVIDIA—that disentangling what money is flowing where is far from easy.

This is systemic risk masquerading as innovation.

If AI demand drops sharply, the consequences could be catastrophic:

CoreWeave and similar startups can't pay their debt service
GPU values collapse as Blackwell and future generations ship
Lenders holding GPU-backed debt take massive losses
NVIDIA's customers—which are also its debtors and investees—cut orders
NVIDIA's revenue collapses despite having "booked" billions in circular deals

And because 88% of NVIDIA's revenue comes from data centers, there's no diversification cushion.

THE BOTTOM LINE

NVIDIA has constructed a $110 billion self-reinforcing financial ecosystem where it invests in companies that buy its products, using those products as collateral for debt, which funds more purchases of NVIDIA products.

It's elegant. It's profitable. And it's exactly the kind of vendor financing that historically precedes spectacular collapses.

The AI boom is real. The infrastructure buildout is necessary. But when the supplier becomes the investor, the lender, the backstop guarantor, and the revenue beneficiary all at once—and when central banks start warning about "sharp market corrections"—it's worth asking whether we're funding the future or inflating a bubble.

NVIDIA's market cap briefly touched $3.5 trillion in 2024. That valuation assumes the circular financing machine keeps spinning forever.

History suggests it 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.