The shift from a training-centric to an inference-centric world has profound implications for the frontier labs that have dominated the AI landscape, such as OpenAI, Google, and Anthropic. These organizations have built their reputations and business models on creating the largest, most powerful models, with training costs running into the hundreds of millions or even billions of dollars. The inference revolution, however, presents both a lucrative opportunity and a significant strategic threat.
