OpenAI has taken a major step beyond AI models and into chip design with the launch of Jalapeño, its first custom AI inference processor built in partnership with Broadcom. The chip is designed specifically for running large language models and has already been tested internally on OpenAI systems, including GPT-5.3-Codex-Spark. Unlike the GPUs used to train AI models, Jalapeño focuses on inference — the process of generating responses after a model has been trained.
Notably, every ChatGPT conversation, coding request, image-generation task, and AI-agent action relies on inference, making it one of the highest ongoing costs for AI companies. According to the Sam Altman-led firm, its goal is to reduce those costs, improve energy efficiency, secure long-term computing supply, and gain more control over the infrastructure that powers its AI services. This is particularly important as demand for AI chips continues to outpace supply across the industry.
Broadcom played a key role in developing the processor. OpenAI provided the AI-specific requirements and architecture ideas, while Broadcom contributed its expertise in semiconductor design, networking technology, and custom AI accelerators. Broadcom CEO Hock Tan said Jalapeño is competitive with Nvidia’s latest Blackwell chips and Google’s TPU systems. The chip is manufactured by TSMC, the world’s leading advanced semiconductor foundry, while server maker Celestica will help build systems around the processor. OpenAI currently plans to use Jalapeño internally rather than selling it to outside customers.
One of the most surprising details is the development timeline. OpenAI claims that the chip went from design to tape-out in about nine months, much faster than the multi-year timelines that are common in advanced semiconductor projects. The company said AI-assisted engineering tools helped speed up parts of the design process. The chip was built specifically for AI workloads involving massive matrix calculations and high-bandwidth data movement, which are essential for running large language models efficiently at data-center scale.
Importantly, Jalapeño is also part of a much larger infrastructure strategy. Last year, OpenAI and Broadcom announced plans to develop around 10 gigawatts of custom AI accelerator capacity, with deployments expected between 2026 and 2029. For comparison, a gigawatt is a unit normally used to measure the output of large power plants. Estimates suggest OpenAI’s broader infrastructure plans could eventually involve around 26 gigawatts of computing capacity across custom chips, Nvidia hardware, and other accelerators.
The Tech Portal is published by Blue Box Media Private Limited. Our investors have no influence over our reporting. Read our full Ownership and Funding Disclosure →
Ashutosh is a Senior Writer at The Tech Portal, largely reporting on new tech, and intersection of technology and business. Ashutosh’s career spans across nearly a decade of technology writing across multiple platforms and languages.
