A detailed look at Jalapeño, OpenAI's ASIC developed with Broadcom in 16 months, which beat Nvidia, AMD, and Google chips on multiple top open-source models (SemiAnalysis)
SemiAnalysis says OpenAI’s first custom inference ASIC is already ahead of Blackwell-class systems on key efficiency tests.
The note says Jalapeño was developed with Broadcom in 16 months and benchmarked against Nvidia, AMD, and Google chips on top open-weight models. It highlights comparisons with Rubin, total cost of ownership, and throughput per megawatt. The surrounding Techmeme entry points to OpenAI’s own claim of higher AI work per watt and lower latency across GPT-OSS, DeepSeek R1, and Kimi K2.5. Techmeme's note
The note says Jalapeño was developed with Broadcom in 16 months and benchmarked against Nvidia, AMD, and Google chips on top open-weight models. It highlights comparisons with Rubin, total cost of ownership, and throughput per megawatt. The surrounding Techmeme entry points to OpenAI’s own claim of higher AI work per watt and lower latency across GPT-OSS, DeepSeek R1, and Kimi K2.5. Techmeme's note
score 8