Breaking

OpenAI Unveils In-House AI Chip 'Jalapeño' Built With Broadcom

June 24, 2026 at 09:05 EDT

Infra & Chips
Foundation Models

OpenAI and Broadcom have revealed "Jalapeño," a custom AI chip designed in-house by OpenAI and tuned for LLM inference. It is part of the 10-gigawatt strategic collaboration the two announced in October 2025, with OpenAI saying it completed the work from design to tape-out in about nine months.

June 24, 2026 · OpenAI × Broadcom

OpenAI Unveils "Jalapeño," Its First In-House AI Chip

A custom accelerator built from scratch for large language model inference — engineered to power products like ChatGPT, cut costs, and loosen reliance on NVIDIA GPUs.

10 GW

Total scale of the compute platform, across multiple generations

~9 mo

From design to tape-out — a notably fast development cycle

~50%

Cost reduction Broadcom has referenced for the platform

Operating cost — referenced ~50% reduction

100%

Prior state
of the art

~50%

Jalapeño
platform

Pricing and detailed benchmarks remain unpublished — full evaluation awaits the technical report.

Full-stack design loop — the claimed edge over general-purpose GPUs

Frontier models
& kernels

→

Serving systems
& product needs

→

Inference-tuned
silicon

Praised

Full-stack efficiency gains
Lower-cost AI access
Nine-month development speed

Watch points

Financing challenges
Possible delayed volume ramp
Benchmarks still unpublished

Deployment

Full deployment across data centers from late 2026.

Primarily for OpenAI and partner facilities — including Microsoft and Oracle — targeting faster ChatGPT responses, expanded Codex task processing, and low-latency API delivery.

Continue reading

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