OpenAI has unveiled Jalapeño, its first custom AI inference processor, built with Broadcom to support large language model workloads more efficiently and reduce infrastructure bottlenecks across ChatGPT, Codex, API, and future AI products.
TL;DR
- OpenAI and Broadcom introduced Jalapeño, OpenAI’s first custom Intelligence Processor.
- The chip is designed for LLM inference, not general-purpose AI acceleration.
- OpenAI says engineering samples are already running ML workloads in the lab.
- Jalapeño was developed from design to manufacturing tape-out in about nine months.
- Initial deployment is planned by the end of 2026, with broader multi-generation rollout ahead.
OpenAI And Broadcom Launch Jalapeño
OpenAI and Broadcom have unveiled Jalapeño, OpenAI’s first custom AI accelerator designed specifically for large language model inference. The chip marks OpenAI’s latest move to build more of its own AI infrastructure stack, from models and products to silicon, networking, memory systems, and deployment platforms.
Jalapeño has been positioned as OpenAI’s first Intelligence Processor, built around the company’s understanding of how LLMs run across products such as ChatGPT, Codex, the API, and future agentic tools. Unlike a general-purpose AI accelerator, the chip was designed from scratch for inference workloads, where AI models generate responses after receiving user prompts.
The chip was developed with Broadcom and Celestica. Broadcom contributed silicon implementation, networking, and connectivity technologies, while Celestica is helping with board, rack, and system integration. OpenAI said engineering samples of Jalapeño are already running machine learning workloads in the lab at target frequency and power, including GPT-5.3-Codex-Spark.
Early testing suggests Jalapeño will deliver better performance per watt than current state-of-the-art alternatives, though OpenAI has not yet released final benchmarks. The company said a detailed technical report on performance will be shared in the coming months.
Why Jalapeño Matters
The launch comes at a time when AI companies are racing to secure enough compute to train and run increasingly capable models. Reuters reported that OpenAI is using custom chips as a way to reduce costs and create an alternative to Nvidia GPUs, which remain widely used across the AI industry. Broadcom CEO Hock Tan told Reuters the chip is comparable with Nvidia’s Blackwell and Google’s Tensor Processing Units.
For OpenAI, the bigger story is control over infrastructure. The company already operates major AI products and builds frontier models, but Jalapeño expands that strategy into the hardware layer.
Greg Brockman, President and Co-Founder of OpenAI, said, “The world is moving to a compute-powered economy.” He added that Jalapeño fits into OpenAI’s long-term plan to make compute more abundant and AI more affordable.
Richard Ho, who leads OpenAI’s hardware program, said the chip was optimized around the kernels, memory movement, networking, and serving patterns that matter most for frontier AI models.
Topics For More Insights
A Nine-Month Chip Development Cycle
One of the most notable parts of the announcement is the development timeline. OpenAI and Broadcom said Jalapeño moved from initial design to manufacturing tape-out in just nine months, with OpenAI’s own models helping accelerate parts of the design and optimization process.
This builds on the broader OpenAI-Broadcom partnership announced in October 2025, under which the companies said they would co-develop and deploy 10 gigawatts of OpenAI-designed AI accelerators and network systems. That rollout was targeted to begin in the second half of 2026 and complete by the end of 2029.
Jalapeño is expected to be the first step in a multi-generation compute platform, with initial deployment planned by the end of 2026. Reuters reported that Celestica will build server systems for OpenAI and that the chips will be used only by OpenAI.
Broadcom’s Hock Tan called the collaboration the beginning of a multi-generation roadmap, adding that it would support gigawatt-scale data centers with Microsoft and other partners beginning in 2026.
For users, the impact could eventually show up in faster AI responses, more reliable access during peak demand, and lower operating costs for services built on OpenAI models. However, final performance numbers are still pending, so the real test will come when OpenAI publishes benchmarks and begins deploying Jalapeño at scale.
