.png)
The World’s Richest Vendor: How NVIDIA Is Financing The AI Bubble
The traditional view of NVIDIA that prevailed was that the firm was the world’s most dominant supplier of AI chips. While that still holds true, over the past eighteen months, the firm has simultaneously pursued the path of a venture investor, a loan guarantor, and a customer-financing arm for the very companies that buy its hardware. The role NVIDIA now plays as a chip vendor and a chip financer is what is driving the already expanding AI bubble. By making credit access easier for buyers who spend that credit on NVIDIA silicon, the company is inflating demand signals that investors then go on to use to justify the valuation, which is now one of the largest in corporate history.
Illustration by The Geostrata
THE SCALE OF INFLATION
NVIDIA closed August at roughly $5.25 trillion market capitalization, the highest of any public company ever. Earlier in the year, the company reached the milestone of becoming the first to reach a $5 trillion valuation, a figure that is way larger than any of its market rivals in the semiconductor business combined. This includes AMD, Intel, TSMC, Micron, Qualcomm, ASML, amongst others. By May of this year, this valuation had climbed high enough to exceed the final economic output of all nations across the globe, with the exceptions being the United States and China. The scale of this inflation is not comparable to a smaller mid-cap firm dabbling in some vendor credit; the scale for NVIDIA is high enough to change the entire demand picture of the AI sector itself.
THE CIRCULAR FINANCING MACHINE
NVIDIA does not simply sell GPUs; it, increasingly, funds those who buy them. In September last year, they committed to invest up to $100 billion in Open AI to help fund a buildout of data centers running on NVIDIA’s own chips. This deal resurfaced the circular concerns that had already trailed the company previously. These concerns were in no way new; NVIDIA had already made fifty separate venture investments in AI companies in 2024, many of which used the funding they got to eventually purchase NVIDIA hardware.
By mid-2026, the pattern has scaled even further, with reports that NVIDIA is weighing a near $250 billion guarantee on Open AI’s data center lease payments, which is layered on top of up to $350 billion in financing for their chip purchases. This is further scaled by an additional $500 billion in an AI infrastructure initiative with South Korea’s SK Group.
According to estimates, every $10 billion NVIDIA commits to OpenAI can eventually translate into $35 billion in downstream GPU purchases and lease payments. This multiplier is already extremely profitable for NVIDIA before a single chip trains any AI model. The worry is that when a company NVIDIA finances or holds equity in typically goes on to buy chips with that same money, the resulting market that the market initially treats as organic growth is actually partially created by the vendor itself, insofar as any shortfall in AI monetization would hit both sides of that loop.
While the company rejects the circular critique by saying that the capital NVIDIA commits is only a fraction of what these companies ultimately raise, historical comparisons do not suggest this defense is strong. In the 1990s, Cisco extended credit in a similar manner to telecom buyers because overall network buildouts were expensive and demand seemed to be limitless. The dynamic contributed to years of capacity overshoot and years of depressed stock when the dot-com bubble eventually burst. NVIDIA’s commitments echo Cisco’s market interactions at the time.
THE DEPRECIATION LEVER
Easy credit only writes half the story; the other half reveals itself in how that credit’s output is booked. Major cloud and AI companies are depreciating NVIDIA’s GPUs over five to six years, whereas their actual life cycle is roughly two to three years. This particular choice lowers reported costs and inflates reported profits across the entire industry. The model puts roughly $176 billion in understated depreciation across major firms between 2026 and 2028. NVIDIA’s response to this was an uncharacteristically detailed internal memo to Wall Street contesting these figures, which also included a dispute of the characterization of its buyback and stock compensation numbers.
Regardless of the response or the argument, when the vendor is also arranging the financing for the buyer, and the buyer controls depreciation assumptions applied to the vendor’s product, the reported demand of the AI boom becomes significantly harder for investors to independently verify.
BEYOND THE BALANCE SHEET
The web of financing extends well beyond NVIDIA into the broader credit system. Warnings surfaced in March 2026 that AI infrastructure is increasingly funded through off-balance-sheet vehicles that ultimately create newer channels of risk running through private credit funds, insurers, and also banks. The structure is effectively similar to how localized mortgage losses in 2007 were turned into a systemic event in 2008.
The role of private credit in this bubble is already sizable and growing even further. Outstanding loans to AI companies went from zero to over $200 billion in the last few years. A specific mismatch compounds the risk of a bubble: the collateral backing most of this debt is the same hardware NVIDIA ultimately sells. That hardware loses 30-40% of its resale value every single year, even if debt raised against it is structured with long, almost real-estate-style maturities.
Ultimately, the biggest critique of the market is that a single company is acting as the supplier, equity investor, lender, and also the guarantor to its own consumer base; this reality makes it incredibly difficult for any investor to separate genuine demand from demand the vendor is purposely creating. Whether NVIDIA’s story ends the same way Cisco’s did is a question the market will eventually have to price in.
BY KRISH
TEAM GEOSTRATA
Comments