--- title: "OpenAI Reveals Its First Custom Chip, 'Jalapeño,' in Push to 'Build the Full Stack'" description: "OpenAI and Broadcom unveiled Jalapeño, the ChatGPT maker's first in-house AI chip — an inference accelerator the companies say marks OpenAI's deepest move yet to design its own hardware and lessen its reliance on Nvidia. Broadcom shares rose about 3% on the news." category: "Tech" category_url: https://boursel.com/category/tech author: "Sofia Marchetti" published: 2026-06-24T13:28:00.000Z updated: 2026-06-24T13:28:00.000Z canonical: https://boursel.com/article/openai-broadcom-jalapeno-first-chip tags: ["openai", "broadcom", "ai-chips", "nvidia", "semiconductors"] --- # OpenAI Reveals Its First Custom Chip, 'Jalapeño,' in Push to 'Build the Full Stack' OpenAI and Broadcom unveiled Jalapeño, the ChatGPT maker's first in-house AI chip — an inference accelerator the companies say marks OpenAI's deepest move yet to design its own hardware and lessen its reliance on Nvidia. Broadcom shares rose about 3% on the news. OpenAI and Broadcom unveiled Jalapeño, described as OpenAI's first "Intelligence Processor" and the first piece of custom silicon to emerge from a partnership the two companies [first announced in October 2025](https://openai.com/index/openai-and-broadcom-announce-strategic-collaboration/). Broadcom shares [rose about 3%](https://www.investing.com/news/stock-market-news/broadcom-rises-3-on-openai-custom-jalapeno-chip-reveal-4758239) on the reveal. The announcement is the hardware reality behind a previously disclosed plan. In October 2025, OpenAI and Broadcom said they would jointly develop and deploy [10 gigawatts of OpenAI-designed AI accelerators](https://investors.broadcom.com/news-releases/news-release-details/openai-and-broadcom-announce-strategic-collaboration-deploy-10), with rack systems slated to begin shipping in the second half of 2026. Today's news is the first actual chip from that effort. ## What Jalapeño is Jalapeño is a custom AI **accelerator** — a chip purpose-built to run the math behind large language models rather than a general-purpose processor. According to the companies, it is designed specifically for **inference**, the stage at which a trained model answers a user's query, as opposed to the more compute-heavy training phase. OpenAI describes it as a blank-slate design for modern LLM inference rather than a chip adapted from older workloads. The companies said the chip went from design to "tape-out" — the point at which a finished design is sent for fabrication — in roughly nine months, and that early lab samples have run a version of OpenAI's models at production frequency and power, [per the Broadcom announcement](https://www.stocktitan.net/news/AVGO/open-ai-and-broadcom-unveil-llm-optimized-intelligence-jqpk7vkxf7jd.html). OpenAI claims Jalapeño will deliver performance per watt "substantially better than current state-of-the-art," with detailed benchmarks promised later. Those are vendor claims and are not independently verified. Initial deployment is targeted for the end of 2026 at gigawatt scale, across OpenAI facilities and partner data centers including Microsoft. ## Why build its own chip "Full stack" refers to controlling every layer — models, software, chips, networking and data centers — rather than buying hardware off the shelf. Hyperscalers including Google, Amazon and Meta have for years designed custom chips to cut costs, secure supply and tune silicon to their own software — economics that grow more compelling at OpenAI's scale. By co-designing its own accelerator, OpenAI can reduce its dependence on Nvidia's GPUs, which dominate the AI market and command premium prices. ## The financial scale The partnership sits within an enormous compute build-out. OpenAI has lined up tens of gigawatts of capacity across deals with Nvidia, AMD and Broadcom; the AMD agreement alone covers [up to 6 gigawatts of accelerators](https://www.cnn.com/2025/10/06/tech/amd-openai-nvidia) plus a stake option. Terms of the Broadcom deal were not disclosed. For Broadcom, custom AI silicon is now a core business. The company reported [$8.4 billion in AI revenue in fiscal Q1 2026, up 106% year over year](https://www.cnbc.com/2026/06/03/broadcom-avgo-earnings-report-q2-2026.html), and Chief Executive Hock Tan has said it has a "line of sight" to AI chip revenue above $100 billion in 2027; he called Jalapeño "just the beginning of a multi-generation roadmap." ## What it means for Nvidia Jalapeño signals that OpenAI's largest supplier relationships are diversifying. Custom inference chips threaten the slice of demand Nvidia serves at high margins, though OpenAI continues to buy Nvidia and AMD GPUs in volume, and the new chip is narrowly aimed at inference. Whether Jalapeño meaningfully shifts the competitive balance will depend on real-world performance and how much of OpenAI's fleet it can ultimately power — neither of which is yet established.