Former OpenAI researcher questions trillion-dollar valuations of frontier AI labs

According to Beating Monitor, former OpenAI researcher Andrew Ho posted a critique questioning the trillion-dollar valuations of leading AI labs. According to Ho’s rough calculations, even with an 80% gross margin on inference services and a 20x price-to-earnings ratio, a company would need annual revenues of $100 billion to $200 billion to justify a $1 trillion valuation—assuming it stops spending on training new models. In reality, companies like OpenAI and Anthropic must continuously train next-generation models; if they pause, users may shift to cheaper alternatives like Qwen or Kimi. While revenues are rising, so too are the training costs for next-generation models, forcing market leaders to keep increasing their investments. Ho also doubts that AI capabilities will rapidly reach full maturity. He believes large models remain uneven in performance, progress may be slower, and is heavily constrained by the availability of high-quality data. Even if model capabilities stagnated at today’s level, it could still take over 20 years for AI to meaningfully penetrate every industry. He views programming agents as perhaps the only high-value product already identified; even deploying teams of engineers to help enterprises integrate AI may not quickly uncover the next profitable use case. Ho remains optimistic about the long-term growth of the AI industry but is unwilling to buy into OpenAI or Anthropic at their current valuations.