OpenAI and Broadcom announce chip designed for LLM inference at scale
Ars Technica AI · Samuel Axon · 2026-06-24
OpenAI and Broadcom jointly announce a custom chip called Jalapeño built specifically for large language model inference in data centers, with both companies describing it as the first in a planned multi-generation silicon partnership.
Appears in
Extraction
Topics: custom-siliconllm-inferenceai-hardwaredata-center-infrastructure
Claims
- OpenAI and Broadcom have jointly announced a chip named Jalapeño designed specifically for LLM inference at data center scale.
- The Jalapeño chip is intended for deployment in large data centers rather than consumer or edge hardware.
- Both companies describe this as only the first generation of a long-term collaboration to refine and improve chips over time.