Breaking

OpenAI Unveils Jalapeño, Its First In-House AI Chip Built With Broadcom

June 24, 2026 at 09:11 EDT

Infra & Chips
Foundation Models

On June 24, 2026, OpenAI unveiled Jalapeño, its first in-house designed AI chip, co-developed with Broadcom. Purpose-built for the LLM inference workloads powering ChatGPT, Codex, the API and future agentic products, the chip targets better performance per watt and cost efficiency.

June 24, 2026 · OpenAI

Jalapeño: OpenAI's First In-House AI Chip

Built with Broadcom and purpose-designed for LLM inference, the chip extends OpenAI's full-stack push — from products and models down to its own silicon and infrastructure.

10GW

of custom accelerators across the collaboration

$18B

Broadcom financing arrangement now facing a snag

~40%

of chips Microsoft must commit to buy as a condition

From Partnership to Production

OCT 13, 2025

Broadcom collaboration announced — 10GW plan

→

JUN 24, 2026

Chip named: Jalapeño, built for inference

→

LATE 2026 – 2027

Mass production at TSMC; deploy via Microsoft / Oracle

What's Known So Far

Purpose

LLM inference only — not training

Manufacturing

Fabricated by TSMC

Networking

Ethernet-based fabric

No benchmarks, pricing or availability disclosed. Real-world performance awaits future technical detail.

Optimists See

Less reliance on scarce, costly NVIDIA GPUs
Cost & power efficiency from inference-specialized silicon
Deeper vertical integration of the full stack

Skeptics Note

Large-scale financing remains an open hurdle
Heavy dependence on a Microsoft purchase commitment
Timeline could slip to 2027 or later

Continue reading

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