ThinkingNews
Top Stories
4 outlets·5 reports

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)

ThinkingNews Desk · how this was written

OpenAI’s Jalapeño ASIC, developed with Broadcom in about 16 months, is designed for fast inference at scale and has been benchmarked against leading competitors. Tests show it delivers roughly 1.5-to-1.9 times more AI work per watt and 1.7-to-3.6 times lower latency than Nvidia chips, and it outperforms the GB300 and comparable Nvidia, AMD and Google chips on several top open-source models.

Written from all 4 reports below, not from any single one.

How it was reported

  1. The Verge·
    OpenAI says its Jalapeño chip can power faster AI responses than the competition

    OpenAI’s new Jalapeño ASIC, co-developed with Broadcom, is built for AI inference and delivers lower latency while maintaining higher throughput than competing chips. The hardware vice president said the chip eliminates the usual trade-off between speed and volume, enabling faster response times for deployed models.

  2. TechCrunch·
    OpenAI’s Jalapeño chip is built for fast inference at scale, benchmarks show

    OpenAI’s Jalapeño chip, co-designed with Broadcom, achieved higher token-per-user counts and greater throughput per kilowatt than Nvidia’s Blackwell system on the SemiAnalysis InferenceX benchmark, delivering faster, lower-latency responses. The architecture minimizes prefill and communication delays by keeping KV-cache data local and dynamically orchestrating compute, memory, and networking resources.

  3. TechMeme·
    OpenAI says its Jalapeño chip delivered 1.5x-1.9x more AI work per watt and 1.7x-3.6x lower latency vs. Nvidia chips across GPT-OSS, DeepSeek R1, Kimi K2.5 1T (Emma Roth/The Verge)

    OpenAI reports that its Jalapeño AI inference chip achieves 1.5-to-1.9 times higher work-per-watt efficiency and 1.7-to-3.6 times lower latency compared with Nvidia’s competing GPUs when running GPT-OSS, DeepSeek R1, and Kimi K2.5 1T models. The benchmark results highlight a substantial performance advantage for OpenAI’s custom silicon in power-constrained and latency-sensitive workloads.

  4. The Next Web·
    OpenAI’s Jalapeno chip outperformed the GB300 on power and speed, according to OpenAI

    OpenAI’s Jalapeno inference chip, co-designed with Broadcom, delivered higher AI work-per-watt and faster response times than Nvidia’s GB300 while consuming 700 W, though it was not benchmarked against Nvidia’s newer Vera Rubin. It runs a small OpenAI model and third-party models from DeepSeek, Moonshot and Kimi, showing the greatest advantage on Kimi. A second version is expected within months, and a third-generation design is already planned.

  5. TechMeme·
    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)

    OpenAI’s Jalapeño ASIC, co-designed with Broadcom in just 16 months, outperformed competing Nvidia, AMD and Google chips on several open-source models such as GPT-OSS, DeepSeek R1 and Kimi K2.5 1T, delivering 1.5-1.9× higher AI work per watt and 1.7-3.6× lower inference latency.

Related stories

Share: