Breaking

OpenAI and Broadcom Unveil First In-House Inference Chip 'Jalapeño'

June 24, 2026 at 09:22 EDT

Infra & Chips
Foundation Models

On June 24, 2026, OpenAI unveiled Jalapeño, its first in-house AI chip designed jointly with Broadcom and optimized for LLM inference. Billed by the company as its first "Intelligence Processor," the custom silicon is purpose-built for OpenAI's own workloads, including ChatGPT, Codex, the API and future agentic products. The announcement came via OpenAI's official blog and a Broadcom press release.

June 24, 2026 · OpenAI

OpenAI Unveils Jalapeño, Its First In-House AI Chip

A custom ASIC built with Broadcom and made by TSMC, designed from scratch for LLM inference — aiming to cut inference costs and reduce reliance on NVIDIA GPUs.

9 mo

From design to tape-out — among the fastest for a high-performance ASIC

10 GW

Scale targeted under Project Nexus with Broadcom

2026

Deployment begins late this year in gigawatt-scale data centers

The Efficiency Goal

"A substantial advantage in performance per watt versus the current state of the art."

Reduced data movement plus balanced optimization of compute, memory and networking allow effective utilization close to theoretical peak — though detailed benchmarks await a technical report in the coming months.

The Full-Stack Pipeline

Specialized for inference, not training · mainly internal use, no external sales

Design

OpenAI + Broadcom

→

Manufacture

TSMC

→

Integration

Celestica

The Promise

Step toward reducing NVIDIA dependence
Strengthens a full-stack strategy
Samples already running GPT-5.3-Codex-Spark in the lab at target frequency and power

The Open Questions

$18B Project Nexus financing hit a snag in May 2026
Broadcom reportedly asked Microsoft to buy ~40%
Heat, compatibility, software optimization — and price/power figures undisclosed

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.