BREAKING
OpenAI Unveils Jalapeño Inference Chip
Built by a Multi-Partner Stack
Design
Broadcom
●
Custom ASIC for LLM inference
●
Uses Tomahawk networking silicon
Build
TSMC
●
Manufacturing
●
Celestica handles systems
0
mo
design to tape-out
0
GW
Project Nexus scale
Inference-First Strategy
1
Specialize for LLM inference
↓
2
Boost performance per watt
↓
3
Cut inference costs
↓
4
Reduce NVIDIA dependence
Deployment From Late 2026
A Step Toward Full-Stack AI
AI NEWS BLITZ
OpenAI has revealed Jalapeño, its first in-house AI chip built for inference.