OpenAI and Broadcom unveil Jalapeño, their first custom LLM inference chip

breakthroughDeveloperCompute

Thursday, June 25, 2026

Confidence

Medium · — cost claim pending independent benchmarking

Evidence

company announcement + multi-outlet reporting (CNBC, TechCrunch, VentureBeat)

What happened

On Wednesday, OpenAI unveiled its first custom-built inference processor, designed and manufactured in collaboration with Broadcom; the chip, named Jalapeño, was designed specifically for the unique needs of OpenAI's inference systems. It is the first step in a multi-generation compute platform designed for initial deployment by the end of 2026. The companies say the chip reached tape-out in just nine months. It reportedly cuts inference costs by about 50%, according to Bloomberg. Coverage in TechCrunch, CNBC, and VentureBeat; see also the OpenAI announcement.

What it means

Jalapeño changes the unit economics of building on OpenAI before it changes anyone else's. If the reported 50% inference cost reduction lands, OpenAI gets the option to either pocket margin or undercut Anthropic and Google on API pricing at the moment competitors are still paying Nvidia retail. The strategic read is darker for everyone above the API: the companies are aiming to build enough silicon to require 10 gigawatts of power , locking customers into a stack OpenAI now owns end to end.

Counterview

Engineering samples are running in the lab, not in production. ASIC roadmaps slip, Nvidia is not standing still, and the headline cost number traces back to a single Bloomberg report OpenAI has not confirmed.

Sources