OpenAI and Broadcom unveiled Jalapeño on June 24, the first custom Intelligence Processor OpenAI has ever shipped, and the companies are pitching it on a single number: roughly 50% cost savings versus typical AI GPUs, per Broadcom CEO Hock Tan. For a company whose Nvidia bill is its largest single cost line, that figure is the entire story.

Jalapeño is an inference part, not a training one. TSMC is manufacturing it. As an ASIC, it trades the general flexibility of a GPU for hard-wired efficiency on the workloads OpenAI actually runs in production, which is exactly the bet AWS made with Trainium and the bet hyperscalers keep making whenever a single tenant’s traffic gets large enough to justify silicon of its own.

The development timeline is the other tell. OpenAI and Broadcom say they went from initial design to manufacturing tape-out in nine months, which the companies are billing as possibly the fastest ASIC cycle ever in high-performance semiconductors. Greg Brockman told CNBC’s David Faber that OpenAI used its own models to compress the design work. “The degree to which our models have been able to accelerate it was very surprising to us,” he said. Engineering samples are already running ML workloads in the lab at production frequency and power, including GPT-5.3-Codex-Spark, with performance-per-watt the companies describe as substantially better than state-of-the-art.

The deployment arc, in Tan’s words, is “small prototype development” late this year, ramping in 2027, going “full tilt in first half ‘28.” A successor is slated for 2028, with new chips annually after that. This isn’t a one-off; it’s a roadmap.

Read it against the rest of OpenAI’s 2026 procurement: earlier deals with AWS Trainium, AMD, and Cerebras, plus the standing Nvidia relationship. Brockman says the company “cannot get compute fast enough.” Tan describes demand across his six customers as “simply insatiable.” Both statements function less as marketing than as confessions about the shape of the market.

Tan reportedly handed the first chip sample to Sam Altman and Brockman in person. The image is the point. The buyer-supplier relationship that defined this cycle of AI is becoming something stranger, in which the largest customers are also, increasingly, the designers.

Sources