OpenAI's Jalapeño Chip Matches Nvidia Blackwell with 50% Cost Advantage, Says Broadcom CEO

OpenAI just put Nvidia on notice. The company disclosed benchmark results for its custom-designed AI chip, codenamed Jalapeño, showing it significantly outperformed Nvidia’s GB300 processors across multiple metrics during both internal and public testing on the InferenceX platform.

The numbers are hard to ignore: Jalapeño delivered 1.5 to 1.9 times more AI work per watt compared to Nvidia’s GB300 processors, and reduced end-to-end latency by 1.7 to 3.6 times across several AI models, including the GPT-OSS 120B and DeepSeek R1.

The chip is an application-specific integrated circuit, or ASIC, designed in collaboration with Broadcom and manufactured by TSMC. Unlike Nvidia’s general-purpose GPUs that handle everything from training massive models to running inference at scale, Jalapeño is purpose-built exclusively for AI inference.

The chip’s sustained power consumption sits at or below 550W, though it’s rated for 700W. Broadcom CEO Hock Tan went further, claiming Jalapeño matches the performance of Nvidia’s Blackwell architecture and Google’s TPU while offering roughly a 50% cost advantage on a per-token and per-kilowatt basis.

Perhaps most striking is the development timeline. OpenAI reportedly went from schematic to tape-out in approximately nine months, with the company’s own AI models assisting in the chip design process.

OpenAI isn’t making this move in isolation. It’s joining a well-established trend among the largest AI players to reduce their dependency on Nvidia’s GPUs, particularly for the most cost-sensitive workloads. Google has been iterating on its Tensor Processing Units (TPUs) for years. Amazon has its Trainium and Inferentia chips.

OpenAI has announced plans for a limited deployment of Jalapeño by the end of 2026, with a full-scale rollout expected in 2027. A second-generation version of the chip is already in development. OpenAI will continue purchasing hardware from Nvidia, AMD, and other suppliers to meet its computing needs, particularly for model training, where Nvidia still dominates.

The performance metrics OpenAI reported were particularly strong on its own frontier models. OpenAI specifically emphasized this “full-stack co-design” approach as a key advantage, suggesting the performance gap could widen further as future models are built with Jalapeño’s architecture in mind.