Steve Eisman: What If OpenAI Actually Fails?

Steve Eisman has spent most of the years since the financial crisis being asked, in his words, to predict the end of the world. In his latest weekly wrap – recorded Thursday night as the 10-year brushed 4.8%, he says he’s still not there on AI, but if he were – he lays out exactly how it would happen.

Eisman is not predicting that OpenAI fails – but it is the weak link in a chain that runs from two money-losing labs, through hyperscaler capex, to roughly half of projected US GDP growth – and arguing that it’s “not too early to think about” what happens if the link breaks.

“I predicted the end of the world once, and believe me, it was no fun. I am in no rush to predict the end of the world again, unless I am really convinced that it’s going to happen. But I’m not going to make such a prediction just because it will get a lot of press. There is no question in my mind that the entire US economy hinges on the success of AI. The amount being spent is just so large that were it to stop, the economy would go into a recession almost immediately.”

The chain: two companies, $700 billion of capex, half of GDP growth

Eisman waves off the two “bubble” arguments echoing through the halls – and that both hyperscalers’ vanished free cash flow, and Nvidia’s circular financing – are survivable if AI pays off. The real vulnerability, he argues, sits one layer down:

“So where is the Achilles heel? I think that it resides with Anthropic and OpenAI, because they are so central to the entire AI food chain. According to reports from various Wall Street firms, something like 70% of hyperscaler AI revenue comes from Anthropic and OpenAI… I can’t confirm those statistics, but they sound right given what we actually know about Oracle.”

From there it’s arithmetic:

“Hyperscalers are spending about $700 billion in capex this year, and even more next year, and that spend accounts for around half of the 2% GDP growth projected for 2026. So one must conclude that the health of the US economy is extremely dependent on hyperscaler capex, and hyperscaler capex is highly dependent on the health of Anthropic and OpenAI. That’s the chain.”

OpenAI is… the weakest link

Between the two labs, Eisman says, “OpenAI is the weaker entity” – pointing to a WSJ report on the 2nd quarter. 

“OpenAI’s June quarter revenue reached $6.7 billion, up only 18% versus the March quarter. Compare that to Anthropic’s revenue of $11 billion-plus in the June quarter, which was up over 100%… OpenAI’s costs reached $12.3 billion, up $3 billion versus the March quarter. So, in three months, revenue increased $1 billion, but costs surged $3 billion. Things are not moving in the right direction.”

(ZH Note; the $12.3 billion Eisman calls “costs” is OpenAI’s operating loss, including stock-based compensation, up from $9.3 billion in the first quarter, per WSJ. On $6.7 billion of revenue, that implies an expense line closer to $19 billion. Revenue rose $1 billion; the loss rose $3 billion.)

Then the departures. Chief revenue officer Denise Dresser left in August after roughly eight months, two days after Brad Lightcap ended an eight-year run. Eisman reads both through the lens of an IPO that keeps sliding:

“Supposedly, OpenAI is getting closer to an IPO. That’s the big payday for employees, because it means that eventually they can sell some of their shares. That two such senior employees would leave now is an important data point.”

Two fairness notes: Lightcap had already been moved out of the COO role in April, so his exit was telegraphed. And Eisman doesn’t mention Fidji Simo, who stepped down in July and was arguably the bigger loss.

The heart of the argument is what unprofitability does to a company’s relationship with its funders:

“When you lose billions upon billions, appearances matter a lot. OpenAI is completely dependent on the kindness of strangers funding its cash flow needs. When a company is growing and very profitable, appearances don’t matter nearly as much… But when a company is not profitable and has an insatiable need for capital, appearances matter more than anything, because if the narrative turns negative, raising capital becomes much more difficult.”

That’s why he flags last week’s “good news” – OpenAI’s ad business hitting a $1 billion annualized run rate – as bad news: earlier this year the company projected $2.4 billion of ad revenue for all of 2026, and $1 billion annualized in September doesn’t get there. 

pic.twitter.com/WGH5sNDwYz

— cocktail peanut (@cocktailpeanut) September 8, 2026

Oracle is the first domino – and the market has already run the drill once

“If OpenAI fails, Oracle is in immediate trouble because of the large increase in Oracle’s debt levels. Oracle’s debt rating is barely above junk. Oracle’s S&P credit rating is triple-B-minus, which is quite weak. Like I said before, it has a $600 billion backlog, and half of that backlog is from OpenAI.”

That isn’t Eisman’s inference; it’s S&P’s. When the agency cut Oracle to BBB- on July 9, it named OpenAI a “key credit risk,” put the lab at roughly half of a $638 billion RPO, and spelled out the failure path: if OpenAI can’t pay, Oracle is left holding data center leases it can’t exit or must re-lease on worse terms.

Eisman’s point is that investors have already seen the preview:

“Prior to the earnings report, the stock was $230 a share. In just a few days, it jumped to $330 a share. Then analysts started publishing reports pointing out that 50% of the RPO was from OpenAI, and the stock gave back all of its gains, plus, in a few months. Today the stock is around $145… From the peak, the stock is down over 50%. That decline is because the market perceives an over-reliance on OpenAI. Imagine what the market would do to Oracle stock if OpenAI fails.“

Why it doesn’t stop at tech – and what he’s doing about it

“The ramifications of an OpenAI failure extend far beyond just Oracle. Remember I said that AI capex accounts for 50% of US GDP growth. While the other hyperscalers are not quite as dependent on Anthropic and OpenAI as Oracle, they are dependent enough. If OpenAI failed, the hyperscalers, I am sure, would cut back on their capex. So I’m starting to think that the demise of OpenAI could push the US into an almost immediate recession.“

Affected sectors are all over the place… It isn’t just Amazon, Google, Microsoft, Oracle and Nvidia. It’s the investment banks, sitting at peak valuations on a financing cycle that AI is feeding. It’s GE Vernova and Quanta on power, Eaton and Rockwell on electrification and automation. The uncomfortable implication: a portfolio that “diversifies” across tech, financials and industrials may own three versions of the same trade.

His answer is reallocation, not stock-picking – healthcare, consumer staples, and within financials the property-and-casualty names – and he names three ETFs by ticker: LVHD, SPLV and KBWP. Then the caveat that should anchor this whole piece:

“It’s still early, and I want to emphasize that I am not making a major call. Not yet. I’m just preparing.”

That Said…

OpenAI has its own numbers. CFO Sarah Friar told employees that July’s annualized revenue already exceeded the entire second quarter, and the company says its run rate has topped $40 billion. Worth knowing: that is a latest-month annualization, while recognized Q2 revenue annualizes closer to $27 billion. Second, strangers have been extremely kind. A March round at a valuation above $852 billion reportedly raised more than $122 billion. Runway isn’t the near-term issue, it’s the next raise – which is Eisman’s point.

One more: Nvidia, where “both things can be true”

Eisman’s read of Nvidia’s $96.2 billion quarter – revenue up 106% year over year – is that the AI story “continues but is displaying potential weakness,” and that “both apparently contradictory ideas can be true.” His evidence for the weakness is Note 7 of the 10-Q: five direct customers at 22%, 14%, 13%, 11% and 10% of accounts receivable, which he sums to 70% and assumes “must be the hyperscalers.”

Careful there. That disclosure is receivables, not revenue, and Nvidia’s direct customers include distributors, ODMs and system integrators, not just clouds. The revenue disclosure in the same filing shows one direct customer at 16% of the quarter. The better version of Eisman’s point is one sentence lower in the 10-Q: Nvidia estimates that one “AI research and deployment company” – OpenAI’s own description of itself – contributed a meaningful amount of revenue by buying cloud services from Nvidia’s customers. Same dependency, no arithmetic error.

And the circularity he mentions in passing is in Nvidia’s own release: roughly $7.8 billion of gains on equity securities ran through other income this quarter, which is why GAAP net income ($59.7 billion) tops non-GAAP ($54.0 billion). Nvidia invests in the companies that buy its chips, then books the markups.

Eisman’s closing line on all of it:

“Once again, it looks like the entire AI ecosystem is dependent on the future health and success of two companies that currently lose billions. Again, if Anthropic or OpenAI ever get into trouble, the whole AI ecosystem will slow to a crawl.”

Watch the entire episode below: