GoKawiilOpenAI claims custom Jalapeño chip beats Nvidia GPUs on AI inference speed
OpenAI detailed benchmark results for its Jalapeño chip, an inference-focused ASIC built with Broadcom, claiming it outperforms Nvidia's GB200 and GB300 chips on efficiency and response speed. Using its InferenceX benchmark across models like GPT-OSS 120B, DeepSeek R1, and Kimi K2.5, OpenAI reported 1.5-1.9x more work per watt and up to 3.6x lower latency. Small-scale deployment is planned by year-end, with volume scaling into 2027.
GoKawiil's interpretation of the reporting above, not reported fact.
This signals OpenAI's push to reduce reliance on Nvidia hardware by designing custom silicon tailored to its own inference workloads, potentially cutting costs and improving user-facing response times for products like ChatGPT. If the claimed gains hold up in production, it could reshape competitive dynamics in the AI chip market, though independent verification of these self-reported benchmarks remains absent.
- Jalapeño is an ASIC co-developed with Broadcom, purpose-built for AI inference rather than training.
- OpenAI claims it beats Nvidia's GB200/GB300 chips on both energy efficiency and latency, avoiding the usual trade-off.
- Deployment begins in small volumes by end of 2025, scaling up through 2027, with no specific chip counts disclosed.
NVIDIA GeForce RTX 4060 Graphics Card — Custom AI inference chips like OpenAI's Jalapeño live in data centers, but if you want to run models at home, a dedicated GPU is the practical entry point. The RTX 4060 handles local LLM and image-generation workloads alongside gaming, letting you experiment with inference on your own hardware.
See NVIDIA GeForce RTX 4060 Graphics Card on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: theverge.com — Emma Roth, 2026-08-25
Published there as: “OpenAI says its Jalapeño chip can power faster AI responses than the competition”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.