AI’s Achilles’ Heel: Why Everything Hinges on Anthropic & OpenAI | The Weekly Wrap

Summary

In this Weekly Wrap, Steve Eisman says the AI trade’s main risk is not the hyperscalers themselves but their dependence on Anthropic and OpenAI, which he sees as loss-making, capital-hungry providers with weak pricing power. He argues the market is largely operating on supposition because there is no AI equivalent of the hard delinquency data that helped him validate the pre-GFC short, and he expects clearer evidence only if OpenAI and Anthropic go public. Eisman points to continued hyperscaler capex and Nvidia’s reported $500 billion financing push as evidence the buildout is still accelerating, while warning that Chinese open-weight models could intensify token price competition. He also reviews earnings and market reactions from CoreWeave, Supermicro, and Cisco as evidence that AI infrastructure spending remains strong, then broadens into a framework about incumbents versus upstarts in streaming, payments, and crypto. Across those examples, he argues Netflix succeeded by changing business models, Circle still faces a hard battle against Visa and MasterCard, and Bitcoin lacks a convincing thesis and has been underperforming despite its high correlation with Nasdaq.

Notes

• Eisman frames the current AI trade as a dependency risk, not a generic capex story.
• He says the AI stack could unwind if OpenAI or Anthropic stumble.
• He argues there is no AI data set comparable to the mortgage delinquency data used in pre-GFC analysis.
• He expects public listings to reveal real operating data for OpenAI and Anthropic.
• He says hyperscalers are still raising capex rather than pulling back.
• Nvidia’s financing move is presented as a new way to fund AI data centers.
• He says the financing structure likely uses securitizations.
• He thinks securitized financing could improve hyperscaler free cash flow.
• He says the LLM providers, not the hyperscalers, are the real weak point.
• He argues there are no meaningful moats around foundation models.
• He says Chinese open-weight models are materially cheaper than Anthropic and OpenAI offerings.
• He thinks enterprise adoption of Chinese models is rising.
• He warns that token pricing power may be limited.
• He cites research estimates that Anthropic and OpenAI drive a large share of hyperscaler AI revenue.
• He says Anthropic and OpenAI may account for 25% to 35% of total cloud revenue.
• He says Oracle’s backlog is unusually exposed to OpenAI.
• He describes Anthropic and OpenAI as loss-making and dependent on capital raises.
• He sees capital dependence at the core of their strategic vulnerability.
• He says CoreWeave posted 112% revenue growth year over year.
• He notes CoreWeave remains unprofitable despite the revenue surge.
• He says CoreWeave management described demand and pricing as very strong.
• He says Supermicro is a server supplier into data centers and competes with Dell.
• He says Supermicro revenue rose 91% and EPS rose 315% year over year.
• He says Supermicro also rallied 19% on Wednesday after the report.
• He characterizes Cisco as another AI data-center beneficiary.
• Cisco EPS was $1.22, up 23% year over year.
• He says Cisco raised both EPS and revenue guidance.
• He says Cisco fell after hours because guidance missed whisper numbers.
• He uses Netflix as an example of an upstart that became the incumbent.
• He says Netflix’s model shift was mailing DVDs and later streaming.
• He says Circle must win in payments, not just as a store of value.
• He argues Visa and MasterCard are extremely hard to disintermediate.
• He says Bitcoin still lacks a satisfying investment thesis.
• He says Bitcoin has behaved more like a Nasdaq proxy than an inflation hedge.
• He suggests prediction markets may be drawing younger traders away from Bitcoin.
• He says Bitcoin is no longer the cool speculative asset for younger investors.
• He argues boards are often ineffective because CEOs handpick directors.
• He says boards are usually window dressing rather than real oversight.