The circular financing debate cuts to a question investors have been circling for months (many, many months): how much of the current AI infrastructure buildout reflects genuine end-user demand versus vendor financing chasing its own tail. Nvidia’s customer concentration disclosure, with three direct customers accounting for 16%, 15% and 13% of first-half fiscal 2027 revenue, gives the concern some statistical weight regardless of how the investment structuring is characterised. Regulatory commentary from the IMF and BIS on AI financing risk more broadly adds a macro dimension, since both flagged growing use of debt and private credit across the AI value chain rather than singling out any one company. For now, the debate looks more reputational than financial for Nvidia specifically, but any sign that hyperscaler capital expenditure plans are slowing would likely reignite scrutiny of exactly this financing structure.
—
Earlier:
—
Huang says Nvidia isn’t buying its own demand, but the scale of its investment web means the question isn’t going away.
Summary:
- Nvidia CEO Jensen Huang dismissed allegations that Nvidia is subsidising its own AI demand through its investments, calling the concern immaterial relative to the business it generates
- Critics’ core argument: Nvidia invests in AI firms or cloud companies, those firms then buy Nvidia hardware, and the revenue flows back, blurring the line between an investment and a sale
- Confirmed deals cited include a $2 billion Nvidia stake in CoreWeave at $87.20 a share, a $110 billion OpenAI funding round including $30 billion from Nvidia, and an August 10 platform with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR aiming to mobilise more than $500 billion in third-party AI infrastructure capital
- Nvidia’s SEC filings show three direct customers made up 16%, 15% and 13% of its first-half fiscal 2027 revenue
- The IMF (in April) and the BIS (on September 10) both flagged rising financing risk across the AI investment ecosystem, including growing reliance on debt and private credit, without singling out Nvidia specifically
- S&P Global estimates the five largest hyperscalers could spend a combined $5.3 trillion in capital expenditure through 2030, while Stanford’s 2026 AI Index puts Nvidia’s share of global AI compute capacity above 60%
Nvidia CEO Jensen Huang has pushed back on claims that the chipmaker is effectively financing its own demand, telling the Goldman Sachs Communacopia and Technology Conference that the company’s investments are too small relative to the business they generate to support the theory. The allegation, sometimes described as circular financing, centres on a pattern critics have identified across the AI industry: Nvidia invests in an AI company or cloud provider, that company then purchases Nvidia hardware, and the resulting revenue flows back to Nvidia. The concern is not that the underlying demand is fake, but that the accounting distinction between an investment and a genuine sale becomes difficult to separate.
The companies most often named in this discussion are OpenAI and CoreWeave. Nvidia is both an investor in and a commercial partner of OpenAI, while CoreWeave relies on Nvidia-supplied infrastructure and financial backing. Huang has previously said Nvidia does not use its investment capital to artificially sustain customers, evaluating each investment on its own commercial merits instead. The scale of the numbers involved is real. In January, Nvidia put $2 billion into CoreWeave Class A shares at $87.20 each, alongside a plan to build more than 5 gigawatts of AI data centre capacity with CoreWeave by 2030. OpenAI subsequently announced a $110 billion funding round at a $730 billion pre-money valuation, with $30 billion coming from Nvidia, $30 billion from SoftBank and $50 billion from Amazon. On August 10, Nvidia went further, partnering with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR on platforms intended to mobilise more than $500 billion in third-party capital for AI infrastructure, a fundraising target rather than a direct Nvidia commitment.
Huang’s central defence rests on reframing what a GPU represents commercially. “In AI, compute is revenue,” he said, arguing that Nvidia’s chips function as productive, revenue-generating assets rather than one-time hardware sales, which he says changes how the investment relationships should be read. Nvidia’s own SEC disclosures add some texture to the concentration question, showing three direct customers accounted for 16%, 15% and 13% of revenue in the first half of fiscal 2027. Huang has been managing expectations on this front for some time. In February, addressing earlier talk of a possible $100 billion OpenAI investment, he said “it was never a commitment,” adding that Nvidia would invest “one step at a time.”
The scrutiny extends beyond Nvidia to the wider AI financing ecosystem. The IMF said in April that AI-related investments could face strain in a downturn, noting increased circular financing across the value chain, though it judged the financial stability impact minor at this stage. The Bank for International Settlements went further on September 10, warning that rising use of debt and private credit to fund AI capital spending could contribute to a larger financial shock if returns fail to materialise as expected. The stakes are considerable given the size of the buildout: S&P Global estimates the five largest hyperscalers may spend a combined $5.3 trillion in capital expenditure through 2030, while Stanford’s 2026 AI Index puts global AI compute capacity at the equivalent of 17.1 million H100 GPUs, with Nvidia accounting for more than 60% of that total. The underlying question, whether genuine end-user demand is keeping pace with the financing, orders and valuations being layered on top of it, remains unresolved and is likely to keep resurfacing as the buildout continues.
This article was written by Eamonn Sheridan at investinglive.com.