Hardware4 min read
OpenAI and Broadcom Unveil Jalapeno, a Custom Reticle-Sized AI Inference ASIC
Tags AI · Infrastructure
Tom's Hardware·
OpenAI and Broadcom unveiled Jalapeno, a custom reticle-sized ASIC designed for LLM inference, built in a nine-month development cycle. The chip is designed to run language models faster and cheaper than GPU-based solutions. This represents OpenAI's first custom silicon and part of a broader industry trend toward purpose-built AI inference hardware that reduces dependence on general-purpose GPUs.
Technical significance
Custom AI silicon from major model developers signals a shift away from general-purpose GPU dependence. If Jalapeno delivers on its performance promises, it could reduce inference costs significantly and challenge NVIDIA's dominance in the AI serving market.