OpenAI pulls ahead in custom chip development race, per Semafor reporting.

Since the ChatGPT moment nearly four years ago, the AI race has been viewed largely as a contest of model performance. Yet Sam Altman, OpenAI's co-founder and CEO, recognized early that compute infrastructure would become a critical battleground—and his company is pulling ahead in a key area: enabling faster, more efficient, and less costly model inference. His long-term strategy crystallized this week with OpenAI's announcement of Jalapeño, its custom inference chip.

Today, AI tools are constrained by a fundamental supply-demand imbalance that caps their performance. By designing custom chips optimized for their specific models, companies like OpenAI can reduce serving costs and improve performance per watt. This optimization phase will eventually prove critical to building a sustainable AI business. While OpenAI may lag behind Anthropic in coding performance, it maintains an infrastructure advantage, commanding significantly more compute than Anthropic, which has partnered with SpaceX for compute access.

With this week's announcement, OpenAI's lead in custom silicon is now unmistakable.