Startups

OpenAI's first custom chip, 'Jalapeño,' arrives via Broadcom with a nine-month design cycle

The LLM-optimized inference ASIC is showing roughly 50% cost savings versus standard AI GPUs, with initial deployment targeted by the end of 2026.

By Giulia Donatello · Startups & funding · 2026-06-28 · 3 min read

Photo: Unsplash / Alexandre Debiève — Close-up of a printed circuit board with processors and chips

OpenAI and Broadcom on Tuesday unveiled Jalapeño, an LLM-optimized inference ASIC that Hock Tan, Broadcom’s president and CEO, told Bloomberg is delivering roughly 50% cost savings versus typical AI GPUs. The chip went from initial design to tape-out in nine months, a cycle the two companies describe as the fastest ever achieved in high-performance advanced semiconductors.

That nine-month figure is the headline, and it’s the part the rest of the industry will read closely. OpenAI president Greg Brockman told CNBC’s David Faber that the company’s own models did meaningful work inside the design loop, and that “the degree to which our models have been able to accelerate it was very surprising to us.” Richard Ho, who leads OpenAI’s hardware program, said the team “optimized the architecture around the kernels, memory movement, networking, and serving patterns that matter most for frontier AI models,” and that engineering samples running GPT-5.3-Codex-Spark in the lab are already operating “close to the hardware’s theoretical limits.”

The partnership was announced last October. Celestica is handling board, rack, and system integration; Broadcom’s Tomahawk networking silicon stitches the platform together at scale. Initial deployment is targeted for the end of 2026, what Tan called “small prototype development,” followed by a real ramp in 2027 and full volume in the first half of 2028. He framed the timeline around “the deployment of gigawatt scale data centers with Microsoft and other partners beginning in 2026.”

ASICs trade Nvidia’s general-purpose flexibility for task-specific economics, and that’s the structural bet here: Sam Altman’s OpenAI is migrating inference, the largest and most predictable slice of its compute bill, onto silicon it co-designed. Broadcom shares are up 10% in 2026 and nearly sevenfold since the end of 2022, a re-rating that tracks the rise of custom-chip programs across hyperscalers, not unlike the way Cisco was repriced during the 1999 buildout of internet backbones.

Tan now counts six custom-chip customers whose demand he called “simply insatiable.” His framing was blunter still: “It’s just much more than we can address, and this is not just ‘26, not ‘27, we’re seeing that same and even elevated demand in ‘28 as well.” Brockman’s gloss from the OpenAI side was shorter. The company, he said, “cannot get compute fast enough.”

Sources