![\"Follow](\"https:\/\/chunks-prd.stocktwits-cdn.com\/_next\/static\/media\/google-news-icon.d9c2bfd1.svg\"\/){kind=icon}
<\/a><\/p>\nSubscribe to Chart Art<\/p>\n
The best trade ideas and analysis from the Stocktwits community. Delivered daily by 8 pm ET.<\/p>\n
OpenAI\u2019s custom-designed \u201cJalape\u00f1o\u201d inference silicon beats Nvidia\u2019s Blackwell systems in throughput per watt and latency on industry benchmarks.<\/p>\n
In standard test suites, Jalape\u00f1o delivered up to 1.9 times more output per watt and cut end-to-end latency by up to 3.6 times compared to Nvidia\u2019s Blackwell.\u00a0OpenAI leveraged its own generative AI models to accelerate hardware engineering, reducing design-to-tapeout timelines to just nine months.While establishing a custom silicon foundation to slash operating costs and lower reliance on single vendors, OpenAI maintains it will continue purchasing accelerators from Nvidia and other third-party partners.\u00a0<\/p>\n
OpenAI unveiled the first live performance benchmarks for its maiden custom inference silicon, code-named Jalape\u00f1o, demonstrating marked advantages in speed and energy efficiency over Nvidia Corp.\u2019s flagship Blackwell architecture.<\/p>\n
\u201cJalape\u00f1o beats Blackwell\u2026 across almost all scenarios without being tuned for any specific point in the curve. It excels not only in low-latency scenarios but also in high-throughput scenario,\u201d SemiAnalysis said.<\/p>\n
According to evaluation data, Jalape\u00f1o delivered higher token-generation speeds per user and superior throughput per kilowatt-hour when evaluated alongside AMD and Google chips.\u00a0<\/p>\n
\u201cIn general, first-generation chips are not competitive, but OpenAI bucks the trend by being industry-leading and beating every Nvidia, AMD, and Google chip we have been able to test on multiple top open source models,\u201d SemiAnalysis\u2019 research showed.\u00a0<\/p>\n
Jalape\u00f1o’s ability to pull ahead of Nvidia’s Blackwell setup represents a significant milestone in custom silicon, proving that in-house hardware tailored specifically for modern artificial intelligence workloads can rival or exceed general-purpose graphics processing units (GPUs).<\/p>\n
Breakthrough Performance: SemiAnalysis Findings<\/p>\n
To establish real-world validity, OpenAI evaluated Jalape\u00f1o using\u00a0InferenceX, a public benchmark framework maintained by SemiAnalysis that measures end-to-end AI request processing. The system was benchmarked across three major large-scale models: GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2.5 1T.<\/p>\n
Across the tested operating range, Jalape\u00f1o achieved 1.5 to 1.9 times higher peak work throughput per watt than benchmarked Nvidia Blackwell configurations. Beyond resolving latency bottlenecks, Jalape\u00f1o lowered end-to-end response times by 1.7x to 3.6x, removing the standard trade-off between throughput capacity and output speed.\u00a0<\/p>\n
OpenAI developed Jalape\u00f1o to eliminate key structural bottlenecks in AI serving, balancing the compute-heavy prefill phase of processing input context with the memory-bound decode phase of generating output tokens.<\/p>\n
AI-Assisted Hardware & Software Engineering<\/p>\n
OpenAI revealed that its generative AI models played a vital role in creating Jalape\u00f1o, developed in partnership with Broadcom (AVGO). Using internal models alongside Codex and domain-specific platforms like GPT-Astra, engineering teams moved from initial concept to chip tapeout in just nine months.<\/p>\n
Beyond initial hardware synthesis, OpenAI employed AI to optimize arithmetic logic circuits and generate highly optimized software kernels. For selected architectural blocks within GPT-OSS, AI-written implementations outperformed human-expert kernels by 1.5 to 1.8 times, dramatically shortening the software bring-up phase for external models to less than two months.<\/p>\n
While Jalape\u00f1o’s benchmark gains over Blackwell showcase strong architectural execution, SemiAnalysis cautioned that the hardware environment continues to evolve rapidly. Nvidia has begun initial delivery of its next-generation\u00a0Rubin platform, which features high-bandwidth memory (HBM4) upgrades that will sharpen competition once Jalape\u00f1o reaches mass production.<\/p>\n
OpenAI plans to begin deploying initial low-volume batches of Jalape\u00f1o within its production data centers by late 2026, followed by broader infrastructure scaling into 2027.\u00a0<\/p>\n
For updates and corrections, email newsroom[at]stocktwits[dot]com.\u00a0<\/p>\n