Former OpenAI Researcher Questions Valuation of Frontier Labs
Former OpenAI researcher Andrew Ho expressed cautious views on the trillion-dollar valuation of frontier AI labs. He roughly calculated that even with an 80% gross margin and a 20x price-to-earnings ratio, an annual revenue of $100 billion to $200 billion would be needed to support a $1 trillion valuation, assuming no new models are trained. In reality, labs must continuously invest in next-generation training, or users will turn to cheaper alternatives like Qwen or Kimi, leading to rising income and costs, putting pressure on leaders for ongoing investments. Source: Public Information
ABAB AI Insight
Andrew Ho briefly worked at OpenAI and has previously discussed the limits of model capabilities and data. This time, he directly targets the current high valuations of OpenAI and Anthropic. In terms of capital, private market valuations rely on continuous financing to support training expenses, and once publicly listed, they will face stricter profitability scrutiny and selling pressure during lock-up periods. Similar to the long cycle of early cloud computing companies expanding from losses to profitable scale, current frontier labs are still in a phase of high investment, high competition, and limited product scenarios. Essentially, this represents a valuation correction before a transfer of pricing power, where capability diffusion and data bottlenecks extend the profitability timeline, making current multiples difficult to support with recent cash flows. ABAB News · Cognitive Laws 1. Continuous training makes it hard to stop profits 2. The cost for leaders is an opportunity for followers 3. Capabilities stuck today will take twenty years to diffuse.