OpenAI and Broadcom have unveiled Jalapeño, OpenAI's first custom AI chip. Announced on 24 June 2026, it is an accelerator, what OpenAI calls an "Intelligence Processor", built from scratch for large language model inference. It is the first chip in a multi-generation compute platform the two companies are building together, with initial deployment planned for the end of 2026.
What is Jalapeño?
Jalapeño is a purpose-built inference chip. Inference is the stage where a trained model answers your prompt, as opposed to training, where the model first learns. OpenAI describes Jalapeño as a blank-slate design for modern LLM inference, not a general-purpose accelerator adapted from older AI workloads.
The architecture is built to cut down data movement and to balance compute, memory, and networking. The aim is to push real-world utilisation closer to the chip's theoretical peak, which is where a lot of performance is normally lost.
How fast was Jalapeño built?
OpenAI says Jalapeño moved from early schematics to fabrication readiness in about nine months. For high-performance silicon, that is unusually fast. New processor cycles are normally measured in years.
The companies credit a tight software and hardware co-design process. OpenAI also used its own models to speed up parts of the chip design. That is a clear example of AI helping to build the hardware that will then run AI.
What performance does Jalapeño claim?
OpenAI says early testing shows performance per watt substantially better than current state-of-the-art hardware. Performance per watt matters because power, not the number of chips, is now the main limit on large AI deployments.
