AI
OpenAI Unveils Jalapeño: Custom AI Chip for Faster, Cheaper Inference
aihardwareopenai
OpenAI and Broadcom announce Jalapeño, a custom Intelligence Processor for LLM inference.
OpenAI Unveils Jalapeño: Custom AI Chip for Faster, Cheaper Inference
In a significant move toward vertical integration, OpenAI and Broadcom announced the Jalapeño, OpenAI's first custom "Intelligence Processor" optimized specifically for large language model (LLM) inference.
The Challenge
As AI inference scales to millions of users, costs dominate. Traditional GPU-based approaches waste energy. Jalapeño targets this inefficiency directly.
Design
Built with Broadcom and manufactured by Celestica, the chip prioritizes:
- Cost efficiency: Lower per-inference expense than equivalent Nvidia setups
- Throughput optimization: Batched inference optimized for LLMs
- Memory bandwidth: Tight integration reduces data movement
Timeline
OpenAI has initial samples and is testing internally. Deployment to API users planned for late 2026.
Impact
- Broadcom shares rose ~3% on the announcement
- Signals OpenAI's commitment to hardware control
- Part of a multi-generation platform strategy
If successful, Jalapeño reshapes AI services economics worldwide.
Comments
Loading comments...