If Mistral is making an attempt to be a French model of OpenAI, its lack of hyperscale compute is a deadly weak spot. It gained’t outspend OpenAI, Oracle, Microsoft, Google, Amazon, SpaceX, or Anthropic. It in all probability gained’t out-recruit them throughout each frontier analysis space, both. The AI market is already suffering from corporations that underestimated how rapidly “good mannequin” grew to become “not adequate.”
But when Mistral is making an attempt to turn out to be the enterprise-controlled AI layer for organizations that don’t need all intelligence to stay behind another person’s API, compute turns into a extra nuanced concern. It nonetheless wants infrastructure, and Mistral appears to understand it. In spite of everything, Mistral raised $830 million in debt to purchase 13,800 Nvidia chips for a knowledge middle close to Paris. That’s a rounding error in comparison with OpenAI and Anthropic, in fact, however the true query is whether or not Mistral can flip relative compute shortage right into a advantage, like Amazon’s Management Precept “Frugality” on steroids. If decrease compute capability leads Mistral to ship smaller, extra environment friendly, and extra specialised fashions, which in flip helps enterprises keep extra management of their information at decrease value, then much less actually does turn out to be extra.
Mistral’s compute problem, then, is to not try to have as a lot compute as OpenAI. It’s to make prospects care much less about uncooked compute scale and extra about deployment flexibility, specialization, and management.
