Watchtower Beating AI News Flash, OpenAI has released the initial test data for its first custom inference chip, Jalapeño, stating that it outperforms NVIDIA's GB300. The chip is at the Pareto frontier on three externally released models: GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2.5 1T. Compared to the current best commercial systems, it achieves 1.5 to 1.9 times higher peak AI output per watt and reduces end-to-end latency by 1.7 to 3.6 times. In high-interaction intelligent agent scenarios, the advantage expands to 2.1 to 4.1 times. Jalapeño has a rated power of 700 watts, with a measured sustained power consumption of no more than 550 watts. OpenAI emphasizes that under intelligent agent load, the true cost should be measured in terms of "AI work done per unit of power consumed," rather than just chip performance.

Jalapeño went from design to tape-out in just 9 months, with AI deeply involved in circuit optimization and validation. Leveraging Codex and GPT-Astra, the team optimized three open-weight models that were not originally planned for in just two months to high performance. Some AI-generated modules achieve speeds 1.5 to 1.8 times faster than human-written implementations. OpenAI plans to deploy Jalapeño to its in-house computing infrastructure by the end of the year while continuing to extensively use external accelerators like NVIDIA's. This is the first generation of a multi-generational chip roadmap, with Gen 2 already in deep development, and Gen 3 taking shape.