AI
OpenAI unveils Jalapeño, its first custom inference chip, signaling deeper AI stack control
OpenAI's new custom chip, co-developed with Broadcom, marks the company's latest push to own the entire AI infrastructure stack alongside other Big Tech players building their own silicon.
The New Stack
OpenAI has announced Jalapeño, a custom inference accelerator co-developed with Broadcom and manufactured by Celestica, positioning it as the company's opening move into custom semiconductor design for AI workloads.
The announcement follows a pattern: Google built TPUs for TensorFlow in 2016; Amazon launched Inferentia in 2018; Microsoft released Azure Maia in 2023. Now OpenAI joins the ranks of Big Tech firms designing silicon to reduce dependence on third-party suppliers and lower compute costs amid exponential growth in AI training demands.
The pitch: ownership as efficiency#
OpenAI frames Jalapeño as a vehicle for making AI "faster, better, and cheaper." The company claims it optimized the chip around the specific patterns and workloads of large language models, with early tests showing performance "substantially better than current state-of-the-art." Engineering samples are already running GPT-5.3-Codex-Spark in the lab, and the company plans gigawatt-scale deployment at Microsoft data centers by year-end.
However, technical specifics remain sparse. OpenAI has declined to release benchmarks, promising a detailed technical report "in the coming months."
The unanswered question#
The core tension: OpenAI's vertical integration strategy may deliver better performance and lower prices, but it also tightens the company's control over the AI ecosystem. While OpenAI insists Jalapeño was designed to work with "all" large language models, the lack of transparency raises questions about whether developers will ultimately become locked into OpenAI's infrastructure.
Richard Ho, OpenAI's head of hardware, noted the chip was engineered around "kernels, memory movement, networking, and serving patterns that matter most for frontier AI models." What remains unclear is how open that architecture will be to competitors.
Speed and ambition#
OpenAI credits its own AI models with accelerating the chip's design cycle, claiming Jalapeño went from design to manufacturing tape-out in nine months—a timeline the company describes as "the fastest ASIC development cycle ever achieved" in advanced semiconductors. The multi-generation roadmap hints at further vertical integration ahead.
