OpenAI and Broadcom Unveil "Jalapeño" — A Custom AI Chip Built for LLM Inference
OpenAI and Broadcom have jointly revealed Jalapeño, OpenAI's first custom-designed AI inference chip, purpose-built from the ground up for large language models. The chip marks a major step in OpenAI's strategy to control its full technology stack — from models and products all the way down to silicon.
On June 24, 2026, OpenAI and Broadcom (NASDAQ: AVGO) officially unveiled Jalapeño, described as OpenAI's first "Intelligence Processor" — a custom AI accelerator designed specifically around the requirements of LLM inference.
What is Jalapeño?
Jalapeño is not a general-purpose accelerator adapted from earlier AI workloads. It is a blank-slate design for modern LLM inference, informed by the systems OpenAI runs every day across ChatGPT, Codex, the API, and future agentic products. The chip's architecture is optimized to reduce data movement and to balance compute, memory, and networking resources, achieving realized utilization much closer to theoretical peak performance.
Performance & Technical Highlights
Early testing shows that Jalapeño will deliver performance per watt substantially better than current state-of-the-art accelerators.
Engineering samples are already running ML workloads in the lab at production target frequency and power, including GPT‑5.3‑Codex‑Spark.
The architecture reduces data movement and balances compute, memory, and networking resources to achieve realized utilization much closer to theoretical peak performance.
A full technical performance report is expected in the coming months.
Record-Breaking Development Speed
Jalapeño was co-developed from initial design to manufacturing tape-out in just nine months, representing what OpenAI and Broadcom believe to be the fastest ASIC development cycle ever achieved in high-performance advanced semiconductors. This speed was made possible through deep software-hardware co-development and, notably, the use of OpenAI's own models to accelerate parts of the design and optimization process.
Key Partners and the Full-Stack Vision
The platform was developed with partners Broadcom and Celestica, covering chip implementation, board and rack system integration, high-performance networking, and scalable production systems. OpenAI President Greg Brockman described the chip as part of a "long-term full-stack infrastructure strategy," designed to make compute more abundant, and AI faster, more reliable, and more affordable.
Deployment Plans
Jalapeño is the first step in a multi-generation compute platform designed for initial deployment by the end of 2026, with gigawatt-scale data center deployments planned with Microsoft and other partners in the years ahead.
The Strategic Flywheel
OpenAI frames Jalapeño as a critical piece of a larger growth loop:
Better infrastructure → greater compute efficiency
Greater efficiency → better model training and serving
Better models → better products
Better products → more usage and revenue → reinvestment in next-generation infrastructure
Why This Is Important
Inference is where AI actually reaches people — and owning the inference chip means OpenAI can directly improve the speed, cost, and reliability of every interaction across ChatGPT, Codex, and its API. This move signals a strategic shift: OpenAI is no longer just an AI software company but is becoming a vertically integrated AI infrastructure company, following in the footsteps of Google (TPUs) and Amazon (Trainium/Inferentia). The fact that OpenAI's own AI models helped design the chip also hints at a future where AI accelerates the development of the very hardware it runs on — a potentially self-reinforcing cycle with profound implications for the pace of AI progress.
Source here