Eisman's Concentration Warning: "70% of AI Revenue Is Just OpenAI and Anthropic" — the NVDA→MSFT→OpenAI Chain
A new bear vector crystallized on Sept 21-22, 2026: not financing capacity, but customer concentration at the top of the stack. Steve Eisman (the Big Short's subprime short) warned on his September 21 podcast that "70% of AI revenue of the hyperscalers is just from OpenAI and anthropic" — which he pegged at about 25-35% of the hyperscalers' entire cloud revenue. His conclusion: "the entire ecosystem is dependent upon two companies" and "one of them I think is in trouble, which is OpenAI." (Caveat: Eisman offered no source for the 70% figure, per 24/7 Wall St.)
The chain, as traced by 24/7 Wall St. (Sept 22):
- Nvidia (market cap ~$5.52T): fiscal Q2 revenue $96.22B (+105.8% YoY), data center revenue $89.02B, Q3 guide $108.0B; supply commitments swollen to $279B, with $108.5B in guarantee obligations to AI cloud and data center partners; nearly $50B invested in the Frontier AI Labs (OpenAI/Anthropic et al.).
- Microsoft (middle of the chain): Azure crossed $100B annual revenue; FY2026 capex $115.95B; commercial RPO $678B; but Q4 free cash flow fell to $19.64B, down 23% YoY, as capex outran operating cash. The restructured OpenAI deal gave Microsoft a
27% stake ($135B) in exchange for OpenAI contracting $250B in incremental Azure services. - OpenAI (the fulcrum): reportedly burning ~$12.3B/quarter; six days before Eisman's warning, CFO Sarah Friar told CNBC (Sept 15) the business is "a diversified set of revenue streams" with strong margins — and the next day CNBC reported investors had approached OpenAI about a round at as much as $1.5T valuation, with no active raise underway.
The crux, in 24/7 Wall St.'s words: "That $250 billion is the Eisman question in one line. If OpenAI cannot generate the cash to consume what it promised to buy, the RPO becomes uncollectable and Microsoft's capex intensity looks less like leverage to a secular boom and more like credit exposure to two loss-making labs.12"
What it means: this reframes the financing debate (The Circular-Financing Web Adds Its Largest Layers Yet: Broadcom's $50B+ OpenAI Chip Package, Oracle's Independent Chip-Lessor — and SpaceX's $40B Now Priced at 194bps CDS) around end-demand quality — the same "who ultimately pays" question, but pointed at the labs rather than the bond market. It also sharpens why the Anthropic IPO (Anthropic's ~$2T IPO: Mid-November Target, Oct 14 Investor Meetings — and the S-1's 2025 Revenue Baseline Is Now Public via Reuters) matters so much: it is the first public, audited look at one of the two companies the ecosystem reportedly depends on. Note the asymmetry with Nvidia's own disclosures: Nvidia's concentration risk is usually framed as supplier commitments ($279B); Eisman's point is that the revenue side of the buyers is even more concentrated than any 10-K discloses.
-
An instance of Circular loops of chipmaker equity and hyperscaler capex cannot scale specialized AI clouds — The circular financing loop is only as strong as its weakest demand node — if the labs' cash generation fails, the chipmaker equity and hyperscaler capex propping it up become uncollectable credit exposure. ↩︎
-
An instance of The entire AI ecosystem is now credit exposure to two private labs. — Eisman's 70%-concentration math restates the ecosystem's dependence on OpenAI and Anthropic as a systemic credit position. ↩︎