The detail that should be humiliating but mostly reads as clarifying: Grok's usage is so low that xAI is selling spare compute to Anthropic. Elon Musk built a competitor to OpenAI partly out of spite, spent billions on GPU infrastructure, and is now effectively renting that infrastructure to the company founded by people who left OpenAI over disagreements with Musk. Meanwhile, Sam Altman and Musk are in court over the soul of OpenAI. The competitive map of AI is a relationship diagram drawn by people who really dislike each other.

The Proxy War Already Happening in Politics

The New Yorker's Gideon Lewis-Kraus reported this week on a congressional primary that became a proxy battle between OpenAI's influence and Anthropic's, with candidate Alex Bores caught in the middle of a regulation debate. The AI companies are not just competing for users and compute. They are competing for the regulatory environment they will operate in. The court case and the primary are the same story told in different registers. around ChatGPT product inflection and model capability expansion.

What Capital Actually Follows

The fast company piece inadvertently reveals something about how moats are built in AI. Anthropic needed compute. Grok had spare compute. The ideological enemies traded infrastructure because economics doesn't care about grudges. A 2026 paper in arXiv CS.CY by Gringras and Salahshoor on frontier lag in AI capability evaluations found that academic papers consistently misrepresent what AI systems can currently do, creating a gap between public perception and actual capability. Which means the real power in AI is not who has the best benchmarks. It is who controls the narrative while the benchmarks catch up.