EquitiesMarket noteTuesday 13 October 2026, 06:18
Nvidia is too big to fail
Nvidia guarantees $108.5bn of data-centre leases, $105bn of them for OpenAI, and holds $51.2bn of private stakes. Its filings show why it is too big to fail.
By The Notebook Desk
Nvidia’s shares fell 1.3% on Monday to $226.32, according to Nasdaq, after Microsoft said it would rely mainly on its own Maia chips to run its AI services, CNBC reported. The fall was small. What it points to is not. Much of the AI industry now depends on Nvidia not only for chips but for money, and the question of whether Nvidia is too big to fail is no longer a figure of speech. “Nvidia is officially too big to fail,” Liz Hoffman, Semafor’s business editor, wrote on Monday. Nvidia’s own filings show why.
What Nvidia has guaranteed
In August Nvidia agreed to guarantee up to $105bn of lease and power payments at a data-centre campus in Pike County, Ohio, which SB Energy is building for OpenAI, its latest quarterly report shows. The campus will have nine data centres and about 4.25 gigawatts of computing load, leased to OpenAI for 20 years. If OpenAI defaults, Nvidia pays part of the bill. The guarantees end early if, among other things, OpenAI achieves a “satisfactory credit rating”. Add $3.5bn of similar guarantees for other AI cloud companies, and Nvidia’s maximum exposure is $108.5bn.
That is 47% of Nvidia’s shareholders’ equity of $229bn, by our calculation. It is also more than twice OpenAI’s annualised revenue of about $50bn, as we reported on 11 October. Nvidia books the smaller guarantees as credit derivatives: contracts that pay out when someone else cannot.
The bank of the AI industry
The guarantees are one part of a wider role. At 26 July Nvidia held $51.2bn of stakes in unlisted companies, up from $22.3bn six months earlier, and $42.8bn of listed shares, up from $12.9bn. In all it reported equity investments of $99bn and commitments to invest $25bn more. Many of the companies it owns stakes in are also its customers, Hoffman noted, and she calls Nvidia “AI’s bank and security blanket”.
It also stands behind the cloud companies that buy its chips. Under agreements worth $36bn, Nvidia has agreed to buy computing capacity from AI cloud firms if they cannot sell it to other customers. And it is owed money by very few buyers: five customers accounted for 70% of its receivables at the end of July.
The IMF sees the same loop
The International Monetary Fund made the same point in general terms on Tuesday. In its Global Financial Stability Report, it warns that weaker AI investment or profits could cause a correction in AI-related asset prices, “amplified by rising leverage and concentrated exposures”. It reviewed 14 circular deals in the AI sector since 2023, in which equity and debt flow between AI firms, and found about $300bn of equity and $650bn of debt, in exchange for $1.4tn of promised future revenue, the IMF’s blog says. Stress at a key company that has received such investment could lower revenue forecasts and valuations, and with them the value of the debt in those deals, the IMF says.
Too big to fail: is Nvidia the new AIG?
The comparison is with AIG, the insurer to which the Fed agreed to lend up to $85bn in September 2008, with the Treasury’s support. AIG’s central place in the financial system only became obvious when it was close to collapse, Hoffman wrote. Nvidia’s can be seen on “a blue-sky day”.
There are big differences. Nvidia is hugely profitable: it earned $59.7bn on revenue of $96.2bn in its latest quarter. Its largest guarantee only starts when the first Ohio data centre opens, which Nvidia expects in its 2029 fiscal year, and it shrinks as OpenAI pays its rent. Hoffman’s own test is political rather than financial: a company is too big to fail when the answer to “would the government step in to save it?” is yes. By that test, size matters too: Nvidia’s 24.1bn shares were worth about $5.5tn at Monday’s close, by our calculation.
Microsoft’s move shows the other side of the risk. Its largest customers are building their own chips. Maia 300 goes into production in 2027, and Microsoft claims it handles 40% more tokens per megawatt than GPUs. Analysts at Evercore ISI estimate that Maia will make up about 10% of Microsoft’s spending on AI chips by early 2027, CNBC reported. The strongest buyers are slowly reducing their dependence on Nvidia, while the weakest lean on its balance sheet.
What to watch
Nvidia’s next quarterly report will show whether the guarantees and stakes have grown. Talks to buy or back Reflection AI, which we cover separately, would add to them. OpenAI’s next funding round, and any credit rating it obtains, will decide how long Nvidia stays on the hook in Ohio.
Disclosure
This is analysis and opinion, not investment advice.