OpenAI's new Jalapeno chip beats Nvidia GB300 in efficiency
thesaraharminta · x · 2026-08-25
OpenAI announced its Broadcom-designed inference chip, Jalapeno, outperformed Nvidia's GB300 in power efficiency and response speed during internal tests. Running at 700W, the chip targets large model inference to reduce data center costs and is set to deploy later this year. However, it was not tested against Nvidia's newer Vera Rubin generation. A second-generation chip is approaching tape-out.
More from Infra
- OpenAI to Deploy Jalapeño by Year-End, Gen 2 and 3 in Development — OpenAI · 2026-08-26
- Sail Research CEO on inference efficiency for long-running AI agents, from chips to engines — agihouse_org · 2026-08-26
- Verne Robotics trained a 3x larger model on ~100x more data via the YC cluster on Together AI — togethercompute · 2026-08-26
- Dwarkesh x Dylan Patel: Anthropic and OpenAI on track to control most usable FLOPs — scaling01 · 2026-08-25
- Google TPU beats NVIDIA Vera Rubin NVL72 on a third-party model without MTP or disaggregation — YouJiacheng · 2026-08-25
- Jensen gifts DGX Station to Perplexity CEO after local frontier-model demo — AravSrinivas · 2026-08-25