Wells Fargo's Ohsung Kwon: OpenAI, Anthropic growth slowing

Wells Fargo's Ohsung Kwon and colleagues point to a specific pattern: combined reported or rumored annual run-rates for OpenAI and Anthropic show that while growth continues at a fast pace, the rate of growth is slowing. The bank attributes this partly to the law of large numbers, and notes that both companies reportedly plan to launch IPOs — a context in which a growth slowdown is a problem.

The note offers a second thread: token usage is rising, but the cost of tokens has plateaued. Kwon's team suggests one LLM behaves much like another for 90% of tasks, and that heavy-user customers are budgeting toward cheaper models accordingly, calling plateauing spend concerning for frontier labs. That dynamic sits alongside the capex side, where the market has grown used to large upward revisions to future hyperscaler spending estimates; Kwon warns that if those revisions get smaller — specifically if capex estimates don't get revised up by more than 10% — AI buildout stocks could sell off.

The broader contributing views are consistent. Adrian Cox at Deutsche Bank and Lisa Shalett at Morgan Stanley show estimated capex growth slowing significantly, with Shalett noting that consensus contemplates significant deceleration in growth rates in 2027, which could prove a headwind to equities next year. Spending would still be a substantial $1 trillion per year, though the direction of revisions is what Kwon flags as the risk.