![\"Jalape\u00f1o](\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/08\/openai_chip_benchmark.png\")
At matched decoding speed, Jalape\u00f1o achieves 54x to 104x the token throughput per kilowatt compared to the best available accelerator, depending on the model. | Image: OpenAI<\/p>\n
Jalape\u00f1o posted these numbers without using techniques like multi-token prediction<\/a> or speculative decoding<\/a>, while some of the comparison systems did rely on those optimizations, so there’s still room to improve.<\/p>\n In its headline performance-per-watt comparison, “Jalape\u00f1o smokes every other chip,” SemiAnalysis writes<\/a>. SemiAnalysis CEO Dylan Patel added<\/a>,\u00a0“Usually first generation chips aren’t competitive, but OpenAI is beating Nvidia Blackwell and even Rubin.”<\/p>\n SemiAnalysis points out that the fairer comparison isn’t Blackwell but Nvidia’s newer Vera Rubin platform<\/a>, since both use HBM4 memory. Even here, Jalape\u00f1o squeezes out more output tokens per megawatt than Vera Rubin, even though Nvidia’s accelerator uses the multi-token prediction optimization that Jalape\u00f1o hasn’t adopted yet. On total cost of ownership per token, the two come out roughly even.<\/p>\n There are caveats, though. Nvidia and AMD have already published results with larger models like Deepseek V4 Pro and Kimi K3 that haven’t been tested on Jalape\u00f1o yet. And while Rubin systems are already shipping to customers, Jalape\u00f1o reportedly hasn’t moved beyond engineering samples.<\/p>\n OpenAI built the chip in nine months, partly using its own models<\/p>\n OpenAI developed Jalape\u00f1o with Broadcom. Design work kicked off in mid-2024, and the final design went to fabrication in November 2025. The full cycle took about 16 months, but OpenAI says only nine months passed between the first chip design and the finished blueprint heading to the factory. The company used its own AI models during development, according to OpenAI<\/a>. Older model generations helped with chip design, while newer ones sped up programming and optimization.<\/p>\n SemiAnalysis sees this as a sign that Nvidia’s much-discussed “CUDA moat”<\/a> may not hold anymore. “The CUDA moat is potentially dead given how fast OpenAI can bring up new models on their silicon,” the firm wrote.<\/p>\n OpenAI CFO Sarah Friar says the chip fits into a broader compute strategy<\/a> where data centers, chips, models, the developer platform, products, and devices all work as one integrated system. She claims Jalape\u00f1o complements OpenAI’s existing partnerships with Nvidia, AMD, AWS, Cerebras, CoreWeave, and others rather than replacing them.<\/p>\n OpenAI has deep ties with several of these companies. Nvidia<\/a>, AMD<\/a>, and AWS<\/a> are all investors or compute partners, with Nvidia being one of the largest. Each of them is also building its own AI chips, which makes the relationship both cooperative and competitive. That said, all of these companies keep saying the world can’t have enough compute<\/a>, a claim that conveniently supports their own business models.<\/p>\n \t\t\t\tAI News Without the Hype \u2013 Curated by Humans<\/p>\n \n\t\t\t\t\tSubscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive “AI Radar” frontier report six times a year, full archive access, and access to our comment section.\t\t\t\t<\/p>\n