Breaking

OpenAI and Broadcom Unveil 'Jalapeño', OpenAI's First Inference Chip

June 25, 2026 at 23:20 EDT

Infra & Chips
Foundation Models

OpenAI and chipmaker Broadcom on June 24, 2026 jointly unveiled "Jalapeño" (formally OpenAI's first Intelligence Processor), the first AI inference chip designed in-house by OpenAI. The ASIC (application-specific integrated circuit) is optimized for large language model (LLM) inference workloads spanning ChatGPT, Codex, the OpenAI API and future agentic products, with the two companies splitting design and manufacturing.

June 24, 2026 · OpenAI × Broadcom

'Jalapeño' — OpenAI's First In-House AI Chip

A reticle-sized ASIC built purely for LLM inference — powering ChatGPT, Codex and the API. Designed in just nine months, with OpenAI using its own models to optimize the silicon.

9 mo

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

~50%

Reported cost savings versus GPUs for inference workloads

Deployment in gigawatt-scale data centers from end of 2026

Inference cost: GPU vs custom ASIC

Reported ~50% cost saving — illustrative relative cost per unit of inference

100%

NVIDIA GPU

~50%

Jalapeño ASIC

The full-stack supply chain

OpenAI

Chip architecture & model-driven design

→

Broadcom

Silicon & Tomahawk networking

→

Celestica

Board, rack & system integration

Why it matters

Cheaper, more scalable AI from full-stack control
Better performance per watt vs current state-of-the-art
Reduced dependence on NVIDIA GPUs

The caveats

Performance claims lack independent verification
Still lab-stage — no real user evaluation yet
Benchmarks & specs undisclosed; deploys end of 2026

Continue reading

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