Hardware1h ago
OpenAI Unveils Jalapeño Inference Chip
OpenAI revealed Jalapeño, a custom inference chip designed to outperform Nvidia's Blackwell on performance per watt. Unlike a standard ASIC, this hardware targets the specific bottlenecks of the current inference stack. The move reduces reliance on third-party silicon. Practitioners should monitor how this vertical integration affects future model latency and deployment costs.