The Spiraling Cost of Making AI

The Wall Street Journal reports that OpenAI and Anthropic project combined training and inference costs of nearly $65 billion this year, rising to $127 billion next year and approaching $250 billion by 2029. Those spending forecasts exceed current revenues for both firms and, in OpenAI’s case, are expected to outstrip revenue through 2029. The story highlights that runaway AI compute costs could keep standalone AI firms unprofitable for years, while large tech incumbents such as Alphabet (and Meta) have lucrative core businesses to fund AI investment. Market implications include greater demand for AI-related infrastructure and chips (supportive for suppliers like NVIDIA), continued pressure on smaller AI pure-plays’ margins and valuations, and potential investor concern about the sustainability of returns given accelerating capital intensity.