OpenAI has unveiled its first in-house silicon, a custom AI inference accelerator called Jalapeño, co-developed with semiconductor partner Broadcom. Announced on June 24, 2026, the chip marks OpenAI's move to build the full stack beneath its models, from products and models down to the processors that run them.
A chip built from scratch for LLM inference
Described as OpenAI's first "Intelligence Processor," Jalapeño is a blank-slate design aimed squarely at large language model inference rather than a general-purpose GPU adapted to AI workloads. OpenAI designed the architecture around its own roadmap of models, kernels and serving systems, while Broadcom contributed silicon implementation and networking technology, including its Tomahawk switching silicon. Hardware partner Celestica handles board, rack and system integration.
Early testing, OpenAI says, shows the first-generation accelerator will deliver performance-per-watt "substantially better" than current state-of-the-art AI GPUs, with realized utilization much closer to theoretical peak. Engineering samples are already running machine-learning workloads in the lab, including GPT-5.3-Codex-Spark.
Nine months from design to tape-out
One of the most striking claims is speed of execution: Jalapeño went from initial design to manufacturing tape-out in just nine months, which the companies believe may be the fastest ASIC development cycle ever achieved in high-performance semiconductors. OpenAI credits the pace partly to using its own models to accelerate parts of the chip design and optimization process.
Jalapeño was handed to OpenAI CEO Sam Altman and President Greg Brockman by Broadcom CEO Hock Tan, underscoring a partnership the two firms frame as a multi-generation compute platform rather than a one-off product.
Scaling to gigawatts
The chip is designed for initial deployment by the end of 2026 and is meant to expand across future generations. Broadcom said the collaboration will help enable gigawatt-scale data centers with Microsoft and other partners beginning this year. The vertical-integration push mirrors OpenAI's broader infrastructure ambitions, following its capital-markets moves and its expansion into robotics. By owning more of the stack, OpenAI argues it can serve intelligence more cheaply and reliably as demand for inference compute surges across the industry.
Reporting based on coverage from OpenAI, Broadcom and CNBC.