OpenAI serves us a hot chip: The Jalapeño AI processor is coming!

Specialized čip for the new era of artificial intelligence

Unlike the general-purpose graphics processors (GPUs) offered by Nvidia, Jalapeño is an ASIC (Application-Specific Integrated Circuit), an IC for a precisely defined task. Its architecture is designed from the ground up to make the most of resources when running models like those behind ChatGPT and Codex. According to OpenAI, the first tests show that Jalapeño will offer significantly better performance per watt of electricity than existing solutions. Broadcom CEO Hock Tan said the new čip’s performance is comparable to Nvidia’s Blackwell GPUs and Google’s TPU units. The goal is to combine power and throughput with lower latency, making it ideal for interactive AI products. The architecture reduces unnecessary data movement and balances computing, memory and network resources.

Record-breaking development with the help ofć own models

The speed of development is particularly impressive. The entire process, from initial design to production readiness of the chip, took only nine months, which is probably the fastest development cycle ever achieved in the advanced semiconductor segment. The company emphasizes that such a pace is made possible by deep cooperation between software and hardware teams, but also by using its own AI models to speed up parts of the design and optimization process. Models that serve millions of users every day have thus helped create the infrastructure on which future generations of AI will work.

A strategic shift in the global race for čips

The launch of Jalapeño ranks OpenAI alongside tech giants like Google, Amazon, Microsoft and Meta, which alreadyć develop their own silicon to reduce costs and optimize performance. The move is part of a broader trend of vertical integration, where leading AI companies seek to take control of the entire ecosystem. The partnership with Broadcom and Celestico, in charge of integration, is the basis for the implementation of “gigawatt-scale” data centers, operational from 2026. This profiles OpenAI not only as a software laboratory, but and as a key game in the global hardware arena.