OpenAI publishes first benchmarks for its custom inference chip Jalapeño, outperforming Nvidia's Blackwell on efficiency
OpenAI announced on August 26 the first public benchmark results for Jalapeño, its custom AI inference chip co-developed with Broadcom. Measured on SemiAnalysis's public InferenceX benchmark against an Nvidia Blackwell system, Jalapeño handled 1.5 to 1.9 times more work per watt while cutting end-to-end latency by 1.7 to 3.6 times; on highly interactive workloads resembling ChatGPT's chatty, low-latency traffic, it ran 2.1 to 4.1 times faster than the systems it was measured against. Semiconductor research firm SemiAnalysis said its engineers visited OpenAI's labs and ran the workloads themselves using InferenceX to independently verify the results. Jalapeño was co-developed with Broadcom on silicon and networking and with Celestica on systems integration, a product of the two companies' October 2025 deal to co-develop 10 gigawatts of custom AI accelerators; OpenAI said it plans a small-scale deployment by the end of 2026, with a larger rollout following in 2027.
Verification note: Summarized from CNBC and Forbes reporting and SemiAnalysis's independent verification. Refer to OpenAI's official statements for technical specifications and the deployment timeline.