OpenAI has taken a major step toward building its own AI infrastructure with the launch of Jalapeño, the company’s first custom-designed artificial intelligence chip. Developed in partnership with semiconductor giant Broadcom, the new processor has been created specifically to run AI models more efficiently rather than train them. The announcement signals OpenAI’s growing ambition to reduce its dependence on third-party hardware while preparing for the next generation of AI services.
Unlike traditional AI accelerators that are designed to handle both training and inference, Jalapeño is focused on inference—the stage where an already trained model generates responses for users. Every time someone asks ChatGPT a question, generates an image or uses an AI coding assistant, inference is taking place. OpenAI believes building dedicated hardware for this task will make its services faster, more efficient and significantly less expensive to operate at scale.
Jalapeño is purpose-built for AI inference
OpenAI describes Jalapeño as an “Intelligence Processor” engineered specifically for large language model (LLM) inference. According to the company, the chip has been optimized around the three biggest challenges facing modern AI systems—compute performance, memory bandwidth and networking efficiency.
Instead of adapting a general-purpose processor for AI workloads, OpenAI says Jalapeño was designed from the ground up for the specific demands of models powering products such as ChatGPT, Codex, the OpenAI API and the company’s future generation of AI agents. By tailoring the hardware to these workloads, the company expects the processor to deliver higher performance while consuming less power than today’s leading AI inference solutions.
OpenAI claims Jalapeño can significantly reduce unnecessary data movement between components, helping the hardware operate much closer to its theoretical maximum efficiency. Although the company has not yet released benchmark numbers comparing the processor against competing AI chips, it says the architecture has been designed to maximise performance per watt—a metric that has become increasingly important as AI data centres continue expanding worldwide.
The project extends beyond the processor itself. OpenAI says it collaborated closely with Broadcom on every layer of development, including chip architecture, server board design, networking systems, rack integration and deployment infrastructure. Manufacturing partner Celestica is also involved in bringing the hardware to production.
Built with Broadcom in record time
One of the most notable aspects of the announcement is how quickly the processor reached manufacturing. According to OpenAI and Broadcom, Jalapeño completed the journey from initial design to manufacturing tape-out in just nine months, making it one of the fastest development cycles reported for a high-performance AI application-specific integrated circuit (ASIC).
Broadcom contributed its expertise in custom silicon engineering alongside high-speed networking technologies, including its Tomahawk networking platform. These networking capabilities are expected to play an important role as AI clusters continue growing into thousands of interconnected accelerators inside modern data centres.
Engineering samples of Jalapeño are already running machine learning workloads inside OpenAI’s laboratories. The company says the chips are operating at their intended production clock speeds and power targets while testing workloads that include GPT-5.3-Codex-Spark, demonstrating that the hardware is already moving beyond the prototype stage.
Greg Brockman, President and Co-Founder of OpenAI, described Jalapeño as part of the company’s broader effort to make artificial intelligence faster, more reliable and more affordable. He also said the world is rapidly moving toward what he called a “compute-powered economy,” where access to efficient computing resources will increasingly determine how quickly AI products can evolve.
Why OpenAI building its own AI chip matters
The introduction of Jalapeño reflects a much larger shift happening across the AI industry. Companies developing advanced language models are no longer focused solely on building better software—they are increasingly investing in the hardware that powers those systems. Custom AI chips allow companies to optimise performance, reduce operating costs and avoid relying entirely on external suppliers for critical infrastructure.
For OpenAI, that strategy could become particularly important as products like ChatGPT continue attracting hundreds of millions of users while new AI agents and enterprise services demand even greater computing resources. Running large language models at global scale requires enormous amounts of processing power, and specialised chips can significantly improve efficiency compared with general-purpose accelerators.
OpenAI says Jalapeño will become the foundation of a multi-generation AI computing platform beginning next year. While the company will likely continue using third-party hardware for many workloads, developing its own silicon gives it greater control over future AI infrastructure and product development.
The launch also places OpenAI alongside other major technology companies that have begun designing their own AI processors to support increasingly demanding machine learning workloads. As competition intensifies across the artificial intelligence industry, custom silicon is rapidly becoming a strategic advantage rather than simply another piece of hardware. With Jalapeño, OpenAI is making it clear that its long-term vision extends beyond AI models and software—it also wants to shape the technology powering the next generation of intelligent computing.
