{
  "id": 4910056,
  "title": "OpenAI lifts the lid on its in-house Jalapeño chip - with benchmarks claiming it beats Nvidia's GB300",
  "url": "https://urgent.news/2026/09/01/openai-lifts-the-lid-on-its-in-house-jalapeno-chip-with-benchmarks",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-01T17:50:00.000Z",
  "source": {
    "name": "TechRadar",
    "slug": "techradar",
    "url": "https://www.techradar.com/pro/openai-lifts-the-lid-on-its-in-house-jalapeno-chip-with-benchmarks-claiming-it-beats-nvidias-gb300"
  },
  "original_language": "en",
  "account": "On August 25, OpenAI unveiled the first performance benchmarks for its in-house chip, Jalapeño, at the Hot Chips conference. Developed in collaboration with Broadcom, Jalapeño is an application-specific integrated circuit (ASIC) tailored for AI inference workloads. OpenAI claims the chip offers 1.5 to 1.9 times more throughput per kilowatt and 1.7 to 3.6 times lower end-to-end latency compared to Nvidia's GB200 and GB300 systems. The chip operates at a power draw of 700W, which is lower than Nvidia's comparable chips rated at 1200W and 1400W.\n\nOpenAI tested Jalapeño on three open models - GPT-OSS 120B, DeepSeek R1 670B, and Moonshot AI's trillion-parameter Kimi K2.5 - using the SemiAnalysis InferenceX suite, with a nominal 8,000-token input and 1,000-token output. The chip showed a 1.9x efficiency gain over the GB200 on GPT-OSS, and 1.7x and 1.5x gains on DeepSeek and Moonshot models, respectively, compared to a GB300 system. However, the company did not publish results comparing Jalapeño and GB300 on the GPT-OSS 120B model.\n\nNvidia's GB300, which is currently more powerful in absolute throughput, maintains its lead in terms of raw performance. However, the efficiency advantage of Jalapeño could have significant implications for OpenAI's AI infrastructure. The chip's design focuses on minimizing data movement and communication delays, with cores and HBM (high-bandwidth memory) divided into slices for local model state management. The A0 silicon version of Jalapeño, which underwent testing, is expected to be replaced by a more efficient B0 stepping by 2027. While Jalapeño may not immediately challenge Nvidia's Vera Rubin, it poses a substantial competitive threat to Nvidia's inference business.",
  "summary": "OpenAI says its 700W Jalapeño chip beats Nvidia's 1,400W GB300 in efficiency, while skipping a crucial head-to-head against the latter's Vera Rubin architecture.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}