OpenAI and Broadcom Unveil Jalapeno Custom Inference Chip in Nine-Month Design Cycle
OpenAI released its first custom AI accelerator designed for LLM inference in partnership with Broadcom, moving from design to tape-out in nine months using its own models to speed development. The chip targets deployment by end of 2026 with substantially better performance per watt than current state-of-the-art.
· OpenAI
View the original source
Read this Signal briefing