OpenAI's Jalapeño chip beats Nvidia's Blackwell on AI work per watt
OpenAI's first custom inference chip reportedly delivers 1.5–1.9x more AI work per watt at peak throughput and up to 3.6x lower latency than the comparison systems. The results matter more than the specs: a model maker now controls silicon, serving software, and systems in one stack, squeezing Nvidia's margins in inference. OpenAI says the architecture also works across third-party models like DeepSeek R1 and Kimi K2.5, not just its own.