OpenAI has unveiled its first custom artificial intelligence chip, codenamed Jalapeño, marking a major strategic shift as the company seeks to reduce its reliance on Nvidia and gain greater control over the infrastructure powering its AI models.
Developed in partnership with Broadcom and announced on June 24, Jalapeño is a custom-built inference processor designed specifically to run large language models more efficiently. Unlike Nvidia's general-purpose graphics processing units (GPUs), the new chip is an Application-Specific Integrated Circuit (ASIC), built for a single task: processing AI queries after models have already been trained.
What Is Jalapeño?
According to OpenAI, Jalapeño has been engineered around the specific requirements of its AI systems, with a focus on improving data movement, networking efficiency, and the balance between computing power and memory.
One of the most notable aspects of the project is its development timeline. OpenAI and Broadcom said the chip went from initial design to manufacturing tape-out in just nine months, a pace they believe could be among the fastest ever achieved for an advanced semiconductor.
OpenAI President Greg Brockman told CNBC that the company's own AI models helped accelerate the design process. "The degree to which our models have been able to accelerate it was very surprising to us," he said.
Early testing indicates that Jalapeño delivers significantly better performance per watt than current industry standards, though OpenAI said final benchmarking is still underway and a detailed technical report will be released in the coming months. Engineering samples are already running production AI workloads, including GPT-5.3-Codex-Spark.
The project brings together several industry players. OpenAI designed the chip's architecture, while Broadcom is providing implementation, networking and connectivity technologies. Hardware manufacturer Celestica is supporting board and system integration.
Microsoft has been confirmed as a key deployment partner. Reports indicate that Broadcom secured a commitment from Microsoft to purchase 40% of the first production run, underscoring the scale of investment behind the project. Deployment is expected to begin before the end of 2026 at gigawatt-scale data centres.
The move also reflects a broader trend in the AI industry, where major technology companies are increasingly designing their own chips to reduce dependence on Nvidia's hardware. Google and Amazon have already developed custom AI processors for their cloud businesses.
For OpenAI, the economics are significant. More efficient inference chips could lower the cost of running AI models, giving the company greater flexibility to improve margins or reduce prices in an increasingly competitive AI market.
OpenAI has described Jalapeño as the first generation of a multi-generation computing platform, signalling its intention to control not only its AI models and products but also the silicon that powers them.