OpenAI, in collaboration with semiconductor giant Broadcom, has unveiled its first «intelligent processor,» Jalapeño, designed for AI servers. This specialized chip (ASIC) is designed from the ground up for inference—the process of processing user queries and real-time LLM operations.
OpenAI, in collaboration with semiconductor giant Broadcom, has unveiled its first «intelligent processor,» Jalapeño, designed for AI servers. This specialized chip (ASIC) is designed from the ground up for inference—the process of processing user queries and real-time LLM operations.
As noted in the OpenAI press release, the new chip should be the first step in creating a large-scale computing platform, which the company, together with partners, plans to deploy by the end of 2026.
Creating its own «silicon» will allow ChatGPT to significantly reduce its dependence on Nvidia’s GPUs, which are currently in short supply and expensive. According to the developers, Jalapeño is optimized for the architecture of current and future models, including experimental systems like GPT-5.3-Codex-Spark, which are already being tested on the new chips in the company’s labs.
Broadcom CEO Hock Tan said in a comment to Reuters that the new chips are not inferior in performance to Nvidia’s flagship Blackwell solution and Google’s TPU processors.
«Our collaboration with OpenAI is a fundamental commitment to scaling the physical infrastructure needed for the next decade of AI. This is just the beginning of a roadmap that spans multiple generations. By co-developing our industry-leading chips directly with OpenAI, we are enabling the deployment of gigawatt-scale data centers with Microsoft and other partners starting in 2026,» said Hock Tan.
Interestingly, the entire Jalapeño design cycle, from initial sketches to finished chip, took just nine months. OpenAI claims that this is a record-breaking time for the semiconductor industry, achieved in part by using its own artificial intelligence models to automate and optimize individual stages of chip development.
«Jalapeño is part of our long-term strategy for a fully-featured infrastructure to increase the availability of computing resources. By developing more of the stack ourselves, we can deliver more intelligence with greater efficiency,» said Greg Brockman, president and co-founder of OpenAI.
While final performance measurements are still ongoing, the company promises that Jalapeño’s power efficiency (performance per watt) will significantly exceed current market leading performance through architecturally reducing latency and optimizing data movement between memory and the processor. A detailed technical report is planned to be released in the coming months.
OpenAI announced a partnership with Broadcom to create its own hardware last spring. Thus, Sam Altman’s company joined other technology giants such as Microsoft, Meta and Amazon, which already have their own lines of AI chips (such as Azure Maia or Meta Training and Inference Accelerator) to reduce the operational costs of cloud infrastructure.
