OpenAI unveiled detailed benchmarks for its Jalapeño AI inference chip at the Hot Chips conference, indicating substantially higher tokens per user and greater throughput per kilowatt than current state-of-the-art processors, including Nvidia Blackwell. Developed with Broadcom and supported by OpenAI’s own models, Jalapeño is a multigenerational platform designed so models, chips, memory, and software evolve together, reducing bottlenecks across prefill and communication phases of the inference pipeline. Initial deployment is expected in late 2026 at limited scale, expanding through 2027, potentially reshaping data center economics by delivering lower latency, higher efficiency AI services and pressuring rival chipmakers to accelerate their own roadmaps.
This update represents a notable development in the Board sector. Organizations and founders tracking this space should evaluate potential strategic and technical implications on their operations.