It's never a good sign when Bethany McLean takes an interest in your company.
From The Comprehensive Guide to the Bank of Nvidia
The Information recently reported that Nvidia has put $170 billion on the line to support neoclouds. But we found that the company’s overall committed capital extends much further. Tallying the disclosures in Nvidia’s latest 10-Q reveals a whopping $530.5 billion in future commitments and guarantees.
The obligations span several buckets: supply commitments, equity injections into neoclouds, backstop guarantees for data centers, and agreements to purchase unused capacity from partners.
The $530.5 billion figure doesn’t include deals that post-date the 10-Q, like the preliminary agreement Nvidia just reached to inject $100 billion of equity in 10 installments into OpenAI, each one tied to OpenAI placing orders for Nvidia chips, or any of those European deals.
...
Layered on top of that $366 billion is another $108.5 billion, which represents a different way in which Nvidia is supporting data centers. In this case, Nvidia isn’t the tenant. Instead, Nvidia has agreed to cover the shortfall if the tenant defaults.
By far the deal’s largest component — accounting for $105 billion — is tied to SB Energy, a SoftBank portfolio company, and its massive PORTS-Pike Technology Campus in Pike County, Ohio, built for OpenAI. In essence, if OpenAI cannot pay, Nvidia is on the hook for up to $105 billion. Nvidia also has agreed to make a $3 billion investment in SB Energy.
SB Energy was supposed to go public this fall, but postponed its IPO amid pushback about the valuation it wanted and broader questions about data centers. Investors also questioned SB Energy’s reliance on OpenAI. Indeed, what’s striking about this is that OpenAI’s commitment reportedly wasn’t enough. The deal needed Nvidia’s backstop.
That shows the doubt that is increasingly creeping into the ecosystem, and how all roads lead back to Nvidia. If Nvidia is using its credibility to bridge what will turn out to be short-lived shakiness, it’s a smart move. It’ll seem less smart if the temporary wobble turns into a bigger rout.
...
The more you dig into it, the more elaborate the relationships are. Nvidia-backed companies may buy chips from Nvidia, but they may also buy computing from neoclouds that Nvidia backs. For instance, Nvidia has invested in AI software companies Mistral and Cohere, both of which use Nvidia’s chips. Both companies also use CoreWeave for compute capacity. So if Mistral or Cohere were to fail, that would hurt Nvidia’s investment and chip sales. But it would also take customers away from CoreWeave, which might put Nvidia on the hook to buy CoreWeave’s capacity, as well as negatively impact the value of the stake Nvidia has in CoreWeave.
...
Unsurprisingly, Nvidia says its exposure is manageable. Chief Financial Officer Colette Kress said on the August earnings call that Nvidia’s “risk is limited” in backing AI labs because it can redeploy its chips to other customers. Nvidia did not return several requests for comment and clarification.
But that’s true only as long as any problems remain isolated, rather than part of a more systemic breakdown.
Ramp economist Ara Kharazian says that 80% of spend on OpenAI and Anthropic is driven by 1% of firms. In an interview with Derek Thompson, Kharazian pointed out that these customers are heavily concentrated in software and technology. It isn’t clear to what extent these customers are being financed by venture capital, or to what extent apparent demand for AI might be demand created by investors betting on AI — and supposedly disparate demand might all be correlated. If those customers need a continued influx of funds to keep spending, then any downturn could cause a lot of demand to evaporate.
On top of that, research firm Gartner has predicted that by 2030, 80% of neoclouds will cease independent operations. And the need for more and more capital means that the boom is in the hands of an increasingly wary — and expensive — bond market.
Nvidia in the AI bubble era has been a study in wrong way risk.
Any one of these scenarios could trigger losses for Nvidia. If neoclouds default, Nvidia loses the chip sales and its investments and is on the hook for some amount of rent payments. If venture capital backed AI companies that are in the top 1% cut back on their AI spending, that harms Nvidia’s investments and chip sales. If demand for its products declines, Nvidia is on the hook for supply commitments it may no longer need. If OpenAI or Anthropic cannot pull off their IPOs, then Nvidia’s investments in those companies are worth less, and they have to raise additional money from the private market in order to make good on their commitments. (OpenAI reportedly has raised less than half the $30 billion it had initially targeted for its latest funding round.) That in turn would likely increase both the draw from Nvidia as an investor and the likelihood that Nvidia will be liable for the various rent guarantees.
“Nvidia is the central bank of AI” is something that would never be said of a well functioning, stable financial system. www.nytimes.com/2026/11/13/t...
— Kate Jackson (@katejackson.bsky.social) November 13, 2026 at 11:50 AM
[image or embed]
— Ed Zitron (@edzitron.com) November 17, 2026 at 9:42 AM
[I think I'm getting this from some classic adventure story like Swiss Family Robinson or The Mysterious Island. but that doesn't really matter.]
You've got this abandoned ship a few miles offshore, filled with great stuff but slowly sinking. You know that when it finally does go down, anyone who's on it, and quite possibly any small craft next to it, will go down with it.
You want to get as much stuff as possible off the ship, so you're inclined to keep going back as long as you can. But you definitely don't want to be there at the end. I think this is how a lot of investors are approaching the AI bubble. While there are certainly true believers who are convinced "number go up forever," I suspect that there are a substantial number of realists who are long on AI despite knowing how horrific the numbers and long-term projections are because they want to wait until the last possible moment before leaving the ship.
For these traders, Jensen Huang let them know that things would be okay in the short term. No neocloud was going to collapse and set off a chain reaction. Credit wasn't going to freeze up. Construction would continue on the data centers. Anyone who could read a balance sheet or just cracked open The Wall Street Journal or the Financial Times now and then knew that the reckoning was coming.
I believe it's Stein's law that says that something that can't go on forever will eventually stop. The rate of growth in capital expenditures that companies have committed to over the next three years cannot keep going. But as long as the Bank of Nvidia was ready to step in, things were stable enough that you could stay on the ship for a little bit longer.
Nvidia as a company is likely to survive, as is most of the Magnificent Seven, but if the Bank of Nvidia goes away, the future of much of the AI ecosystem is considerably murkier.






