On November 13, 2026, The Wall Street Journal published an investigation alleging that NVIDIA systematically provided funding to artificial intelligence startups on the condition that they use the proceeds to purchase NVIDIA GPUs and build data centers. The logic of this closed loop was straightforward: NVIDIA puts up the money → startups buy cards → GPU demand figures rise in the market → NVIDIA’s valuation is sustained. The report reached the front page of Hacker News the day it was published.
This allegation did not exist in isolation. According to a subsequent Bloomberg review, NVIDIA’s total financing commitments in the AI sector have exceeded $750 billion, spanning equity investments, financing guarantees, and chip purchase commitments, with reach extending across the entire supply chain from small cloud providers to OpenAI.
Where the Panic Began
The starting point disclosed in the report was Jensen Huang’s fear of TSMC overcapacity. Large cloud providers—Amazon, Google, and Microsoft—have accelerated development of in-house AI chips in recent years, gradually reducing their dependence on NVIDIA’s standard products. This trend was already quite pronounced in the first half of 2026: NVIDIA held a large volume of prepaid silicon wafer orders from TSMC, while the originally intended buyers were scaling back purchases. According to the report, Huang described the situation as an oncoming “avalanche.”
A response strategy took shape: rather than cutting prices to clear inventory, NVIDIA would proactively invest in a group of small cloud service providers that could not originally afford GPUs, using its own money to build data centers for them while creating a continuously robust purchasing demand on the books. Orders from this group of customers would fill the demand gap left by the retreat of hyperscale cloud providers, allowing the market narrative of “GPUs in short supply” to be maintained.
The Structure of the $750 Billion Network
Bloomberg’s review sketched the specific contours of this network: in 2026 alone, NVIDIA committed more than $40 billion to AI equity deals, including about $30 billion in OpenAI equity; there were also up to $250 billion in financing guarantees tied to OpenAI’s data center construction project in Ohio; in addition, an additional $350 billion in chip financing agreements were under negotiation.
The most illustrative node in this network is CoreWeave. As of March 31, 2026, this small cloud service provider carried $25.1 billion in principal debt and posted a net loss of $740 million in the first quarter, yet it held $100 billion in revenue backlog—much of which came precisely from compute purchase commitments unlocked by NVIDIA’s investment. When CoreWeave went public in 2026, its public filings for the first time showed the market a new financing instrument, the “high-performance computing collateralized delayed draw term loan,” which in essence involved issuing securities backed by GPU clusters as collateral.
This means that when market investors, data center operators, and regulators try to read the signal of “GPU demand,” what they see is not the genuine purchasing intent of independent buyers, but a circular flow of funds that is seeded by NVIDIA, collateralized by NVIDIA’s own GPUs, and ultimately flows back into NVIDIA’s revenue.
Internal Dissent and the August Showdown
The most dramatic detail in The Wall Street Journal report occurred at an executive meeting in August 2026. NVIDIA CFO Colette Kress, the company’s current acting CEO, explicitly raised two objections at the meeting: first, she called for scaling back circular financing arrangements with AI startups; second, she called for reducing the size of the $250 billion financing commitment to OpenAI.
According to the report, Huang responded to these objections by likening Kress to a “deer in headlights”—someone who freezes rather than acts when danger approaches. The force of this detail lies in how it reveals a fundamental divergence between two strategic judgments: the financial prudence logic represented by Kress holds that providing guarantees to customers who cannot afford GPUs is essentially accumulating bad-debt risk; the growth logic represented by Huang holds that as long as the compute narrative can be maintained, short-term financial exposure is bearable.
Four months after this meeting, The Wall Street Journal made it public. By then, Huang himself had left the CEO position at NVIDIA, and Kress had taken over as acting CEO. Placed side by side, the logic of the timeline is not hard to infer.
Regulators Had Already Noticed
The controversy did not come only from a single media investigation. As early as June 28, 2026, the Bank for International Settlements (BIS) issued a warning, listing the structure in which “equity holdings coexist with multi-year chip purchase commitments” between chipmakers and AI labs as a systemic risk factor. In its July 8 report, the International Monetary Fund (IMF) was even more direct, explicitly naming such “circular financing under opaque private arrangements” as a hidden danger to financial stability.
The warnings from the two major international institutions show that the nature of this issue has gone beyond the scope of corporate compliance. When a chip supplier simultaneously plays the three roles of capital provider, hardware seller, and market demand creator, the supply-and-demand signals of the entire industry lose their independence.
Cracks in the Scarcity Narrative
Over the past three years, “severe GPU shortage” has been one of the most solid consensuses in the AI industry. This narrative has supported NVIDIA’s triple-digit price-to-earnings ratio, supported the premium pricing of cloud service providers, and supported many AI startups in using compute costs as moat logic in fundraising pitches.
The Wall Street Journal investigation, for the first time, opened up the internal construction of this narrative from the supply side: so-called scarcity may partly not be a supply-demand imbalance that formed spontaneously in the market, but a perceived state actively maintained with the supplier’s participation. Between the true demand signal—how much compute independent buyers are willing to buy at market prices—and the financing-amplified demand signal—how much compute startups bought after NVIDIA injected capital—there is a gap that no one can precisely measure.
The existence of this gap means that all analysis inferring actual AI investment based on “GPU purchase volumes” needs to be called into question.
If the money for a GPU purchase order comes from the seller of the GPU, does that order count as “demand”?
The Trust Cost Across the Supply Chain
In the short term, the direct financial impact of this investigation on NVIDIA is still difficult to quantify—the allegations remain a news report not legally confirmed, and NVIDIA’s core product competitiveness does not disappear because of this. But the trust cost is real.
The first to be affected are the institutions that rely on “GPU demand data” to make investment decisions: venture capital firms evaluating AI infrastructure companies, data center operators planning the pace of expansion, and enterprise buyers judging whether they should stockpile compute have all used NVIDIA’s order backlog as an anchor for market signals. If this anchor itself contains the inflationary effect of circular financing, all judgments built on it need to be recalibrated.
Second is the situation of small cloud service providers. The report explicitly states that some small providers funded by NVIDIA “cannot afford to buy GPUs at all”—this means that the commercial sustainability of this group of customers depended from the start on continued funding from NVIDIA, not on independent market competitiveness. Once NVIDIA tightens such arrangements due to its own financial pressure or regulatory pressure, these companies will face the risk of an abrupt break in their funding chain.
Finally, there is the fragility of the entire AI infrastructure investment logic. Over the past two years, hundreds of billions of dollars globally have been invested under the name of “AI compute infrastructure.” The return assumptions of these investments are highly dependent on continued compute scarcity and persistently high rental prices. If the scarcity itself is manufactured, then when this narrative begins to weaken, the adjustment speed of the secondary market is often far faster than the change in the value of the infrastructure itself.
Independent Judgment
Based on the facts that can currently be confirmed, NVIDIA’s circular financing structure is already large and systemic enough to warrant regulatory intervention, regardless of whether the internal conflict details in the WSJ report are ultimately legally confirmed. The backing of the BIS and the IMF represents authoritative judgment beyond a single media report.
The deeper question is this: during the super-cycle of AI infrastructure investment, the credibility of demand signals has never been seriously audited. Every GPU purchase is recorded as an independent market action, but the funding sources, related-party interests, and repayment capacity behind it have never been public information. The value of The Wall Street Journal investigation does not lie in whether it can establish NVIDIA’s legal liability, but in the fact that it forces the entire industry to confront a question no one was previously willing to publicly ask: how solid is the true demand foundation of this compute arms race?
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接