There have been rumors that OpenAI would partner with Broadcom to produce AI chips, aiming to reduce reliance on NVIDIA GPUs. Recently, OpenAI has finally revealed more details. Its first self-developed AI inference chip has been officially named “Jalapeño” and is positioned as OpenAI’s first “Intelligence Processor” — an accelerator specifically designed for LLM (large language model) inference.
In other words, Jalapeño is not a chip designed for installation in consumer computers; rather, it will be deployed in data centers in the future, becoming part of the infrastructure behind OpenAI’s own AI services.
Image source:OpenAI
OpenAI unveils its first self-developed AI chip, Jalapeño: partnering with Broadcom, it will be deployed in data centers by year-end.
OpenAI said Jalapeño is designed from the ground up for modern large language model inference, rather than being adapted from a general-purpose accelerator originally built for other AI workloads. Its design draws on the systems OpenAI actually runs every day in ChatGPT, Codex, API, and future agent products.
這個平台的目標，是讓先進 AI 變得更快、更可靠，也讓更多人更容易使用。
This chip went from initial design to tape-out in just nine months. The term tape-out refers to the important milestone at which the chip design is formally finalized and handed over to a foundry for manufacturing. For high-performance advanced semiconductors, completing this stage within nine months is already a fairly rapid development pace.
OpenAI added that achieving such a development timeline relied not only on co-designing hardware and software, as well as Broadcom’s experience in chip implementation, but also on using OpenAI’s own models to assist with certain design and optimization processes, further shortening development time.
Engineering samples are already running machine learning workloads in the lab at target production frequency and power, including GPT-5.3-Codex-Spark. Early tests indicate that Jalapeño will significantly outperform current top-tier solutions in performance per watt, with a detailed technical report to be published in the coming months.
Overall, Jalapeño can be seen as the first officially named product to emerge from OpenAI and Broadcom’s 10GW customized AI accelerator partnership. When the two sides announced the collaboration in October 2025, they already mentioned that they would jointly develop systems including AI accelerators and Ethernet solutions, with Broadcom deploying AI accelerator and network system racks, targeting a start in the second half of 2026 and completion of the related deployments by the end of 2029.
Following Jalapeño’s official unveiling, OpenAI again emphasized that it is not a single chip product, but the first step in a multi-generation computing platform from both companies. OpenAI expects to begin initial deployment by the end of 2026, and will continue to scale up over the next few years with data center partners such as Broadcom, Celestica, and Microsoft.
Source: KOCPC Chinese