OpenAI and Broadcom Unveil 'Jalapeño': The Custom Inference Chip That Could Reshape AI Economics
On June 24, 2026, OpenAI and Broadcom unveiled Jalapeño — OpenAI's first custom-designed AI chip — in a ceremony at OpenAI's San Francisco headquarters. According to Broadcom's official press release, engineering samples of the chip were physically delivered to OpenAI CEO Sam Altman and President Greg Brockman by Broadcom President and CEO Hock Tan and Semiconductor Solutions President Charlie Kawwas. The chip, formally described as an "Intelligence Processor," is specifically designed for inference — the process of running pre-trained AI models to generate responses — rather than for the computationally intensive task of training new models. TechCrunch reported that the chip is an application-specific integrated circuit, or ASIC, which is less flexible than Nvidia's general-purpose GPUs but can be purpose-built for targeted AI tasks at lower cost.
The strategic significance of Jalapeño cannot be overstated. As VentureBeat reported, OpenAI spent approximately $14 billion serving ChatGPT in 2025 on third-party Nvidia GPUs, operating at a structural cost disadvantage compared to rivals such as Google, which has long used its own Tensor Processing Units, and Amazon, which deploys its in-house Trainium chips. OpenAI's President Greg Brockman told CNBC directly that the company "cannot get compute fast enough," and Broadcom's Hock Tan echoed that sentiment, saying demand from his company's hyperscale customers is "simply insatiable." By designing and owning its inference silicon, OpenAI is attempting to rewrite the unit economics of serving AI at scale, potentially cutting inference costs by up to 50% according to reporting by AI Tools Recap.
The pace of development is itself remarkable. According to Broadcom's official announcement, Jalapeño went from initial design to manufacturing tape-out in just nine months — what the companies describe as potentially the fastest ASIC development cycle ever achieved in high-performance semiconductors. OpenAI's own models were used to accelerate parts of the design and optimisation process, a recursive loop that the company highlighted as proof of its technology's real-world utility. Richard Ho, who leads OpenAI's hardware programme, said in the official announcement that the chip was "optimised around the kernels, memory movement, networking, and serving patterns that matter most for frontier AI models," and that early testing shows Jalapeño will execute key workloads close to the hardware's theoretical limits.
The deployment roadmap is ambitious. As reported by both Broadcom's investor relations team and CNBC, prototype data centre deployment is targeted for the end of 2026, with production ramp planned through 2027 and full scale in the first half of 2028. Microsoft is named as a primary data centre partner, with the partnership ultimately targeting gigawatt-scale deployments. The Semiconductor Industry Association, in a recent report produced with Deloitte, noted that government and industry are expected to invest over $4 trillion in new data centre infrastructure through 2028, of which up to $2.8 trillion will be spent on semiconductors — a macroeconomic backdrop that gives further weight to OpenAI's bet on vertical integration.
For Nvidia, Jalapeño represents the clearest signal yet that its largest customers are engineering a path around it. Nvidia currently commands an estimated 80 to 90 percent of the AI data centre accelerator market for training workloads, according to industry analysis, but inference — where models run continuously to serve billions of users — is an equally vast and growing frontier. As OpenAI prepares for what reports describe as a highly anticipated IPO targeting a valuation near $1 trillion, the Jalapeño chip offers private investors and public markets a concrete answer to a persistent question: how does a company that spent $14 billion serving its own products ever become profitable? The answer OpenAI is now betting on is simple, and it runs on custom silicon.