OpenAI and chip-design partner Broadcom publicly unveiled Jalapeño on June 24, 2026 — OpenAI’s first custom silicon, built to handle the inference workloads that power ChatGPT and the company’s other AI products.
What Jalapeño does
Unlike training chips, which require enormous compute to teach a model from scratch, inference chips serve users in real time. Jalapeño is purpose-built for this task, optimized for running large language models efficiently and at low cost.
Early benchmarks show the chip delivers “significantly better performance-per-watt than current state-of-the-art alternatives,” according to the announcement. It is specifically tuned to accelerate real-time coding models — one of the most resource-intensive inference categories.
An unusually fast build
OpenAI president Greg Brockman said Jalapeño went from design to manufacturing tape-out in nine months — a timeline the companies describe as the fastest ever for an advanced semiconductor. OpenAI’s own AI models contributed to the chip’s design, a notable instance of AI being used to build AI infrastructure.
The Broadcom partnership was first announced in October 2025. Initial deployment of Jalapeño chips is targeted for the end of 2026.
Reducing reliance on Nvidia
OpenAI currently depends heavily on Nvidia GPUs for both training and inference. Jalapeño is a strategic move to reduce that dependence, following a path already taken by Google (with TPUs) and Amazon (with Trainium and Inferentia). As OpenAI stated: “By designing more of the stack ourselves, we can serve more intelligence with greater efficiency and keep pushing advanced AI toward broader access.”
Why it matters for Georgia
Lower inference costs benefit the entire global AI ecosystem, including Georgia’s growing technology sector. As the country allocates $18.4 million toward an AI research and competence centre (2026–2029), more efficient inference infrastructure means Georgian startups and researchers can access capable AI services at reduced cost — accelerating local development without requiring large capital outlays on hardware.