OpenAI's Jalapeño chip rivals Nvidia Blackwell in AI inference

OpenAI has officially revealed Jalapeño, its first custom-designed AI processor made in partnership with Broadcom. The chip is an Application-Specific Integrated Circuit (ASIC) optimized exclusively for AI inference—the stage where models like ChatGPT process user requests in real time. According to Broadcom CEO Hock Tan, early benchmarks show Jalapeño matches the performance of Nvidia's Blackwell chips and Google's Tensor Processing Units while offering substantially better performance per watt. This efficiency gain is critical for OpenAI, which runs massive server clusters to power ChatGPT and Codex agents. The company calls Jalapeño the 'first step in a multi-generation compute platform' and expects to deploy it by the end of 2026.

Jalapeño arrives just nine months after OpenAI announced its chip-making partnership with Broadcom, reflecting the intense industry race to reduce dependence on Nvidia's GPU supply constraints. Microsoft, Meta, and Amazon have all launched their own custom AI chips recently, though most still trail Nvidia on overall performance. By controlling its own silicon, OpenAI hopes to lower operational costs and scale its large language models more efficiently. The chip is designed specifically for inference workloads, leaving training to GPUs, but future generations may expand capabilities. For now, Jalapeño solidifies OpenAI's hardware strategy and puts pressure on Nvidia's dominance in the AI chip market.