Steve Eisman said the main vulnerability in the AI boom is the sector's heavy dependence on OpenAI and Anthropic, but argued it is too early to bet against the trade because investors still lack the kind of operating data that preceded his subprime short. Speaking on his Weekly Wrap podcast, Eisman said there is no equivalent AI data set to the delinquency figures his team used before shorting mortgage markets, leaving investors to infer the economics of the two private companies from fundraising and spending signals. He said the bear case is real because large language models, or LLMs (AI systems trained on vast text data), appear to have limited competitive moats and face cheaper Chinese open-weight rivals, a dynamic that could eventually trigger a price war. Eisman said research estimates suggest OpenAI and Anthropic generate about 70% of AI revenue across Microsoft, Amazon and Alphabet, and 25% to 35% of their cloud revenue, while roughly half of Oracle's approximately $600 billion backlog comes from OpenAI. He said that level of concentration is "huge and quite scary" because weaker economics at the labs could prompt hyperscalers (largest cloud providers) to curb capital spending and send the broader AI chain "into reverse." He added that public listings by OpenAI and Anthropic could provide the first clearer financial evidence, with Polymarket assigning Anthropic a 90% chance of reaching an IPO before OpenAI and only a 20% chance of an OpenAI IPO this year. For now, Eisman said the boom still has momentum as major technology groups continue to spend, Nvidia is working with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize more than $500 billion for AI infrastructure, and Anthropic is projecting $190 billion to $200 billion of 2028 revenue from a $47 billion annualized run rate in May, according to Reuters.