OpenAI and Broadcom Unveil Jalapeño, Their First AI Inference Processor

OpenAI  and Broadcom  have jointly introduced Jalapeño, OpenAI’s first Intelligence Processor, representing a significant milestone in the company’s long-term infrastructure strategy. The accelerator has been specifically designed for large language model (LLM) inference and serves as the first component in a planned multi-generation compute platform intended to make advanced artificial intelligence faster, more reliable, and more accessible. The announcement highlights OpenAI’s expansion from developing AI products and models to designing specialized hardware, reinforcing its commitment to controlling the full technology stack that powers modern AI systems.

Jalapeño was developed through a collaboration between OpenAI, Broadcom, and Celestica. OpenAI contributed the architectural design based on its deep understanding of LLM workloads and future model requirements, while Broadcom provided expertise in silicon implementation, networking technologies, and production scaling. Celestica supported board, rack, and system integration. The processor has been designed not only for OpenAI’s own models but also for current and future LLMs across the broader AI industry. Engineering samples are already running machine-learning workloads at target frequencies and power levels, including workloads associated with GPT-5.3-Codex-Spark.

Early testing results indicate that Jalapeño delivers substantially better performance per watt than current state-of-the-art AI accelerators. Although final performance evaluations are still underway, OpenAI expects the architecture to achieve higher efficiency by reducing unnecessary data movement and balancing computing, memory, and networking resources. This design enables the hardware to operate much closer to its theoretical peak performance. Broadcom’s advanced networking technologies, including its Tomahawk networking silicon, further support large-scale deployment and high-performance operation.

According to OpenAI President and Co-Founder Greg Brockman, the world is increasingly becoming a compute-powered economy, making efficient infrastructure essential for future AI systems. He emphasized that Jalapeño forms part of OpenAI’s long-term strategy to increase the availability of computing resources, resulting in AI systems that are faster, more affordable, and capable of addressing increasingly important challenges. By designing more components of the technology stack internally, OpenAI aims to improve efficiency and broaden access to advanced artificial intelligence.

Richard Ho, who leads OpenAI’s hardware program, stated that the processor was specifically optimized for the kernels, memory movement patterns, networking requirements, and serving demands associated with frontier AI models. He noted that early testing suggests Jalapeño can execute key workloads while operating close to the hardware’s theoretical performance limits. Broadcom President and CEO Hock Tan described the collaboration as a long-term commitment to building the physical infrastructure required for the next decade of AI development. He emphasized that Jalapeño represents only the first stage of a multi-generation roadmap intended to support gigawatt-scale data centers beginning in 2026.

Unlike general-purpose accelerators adapted from earlier computing workloads, Jalapeño was designed specifically for modern LLM inference. The architecture is informed by the operational requirements of services such as ChatGPT, Codex, the OpenAI API, and future AI agents. The goal is to combine the throughput and computational capabilities of leading AI accelerators with the low latency needed for interactive AI applications. This specialization makes Jalapeño particularly suitable for large-scale AI services that require both high performance and rapid response times.

The development of Jalapeño demonstrates OpenAI’s full-stack approach to artificial intelligence. The company is not only creating advanced models and user-facing products but also designing the underlying infrastructure, including chip architectures, memory systems, networking technologies, deployment platforms, and serving systems. By optimizing all layers of the technology stack simultaneously, OpenAI aims to improve the speed, reliability, and affordability of its AI services.

Jalapeño also strengthens OpenAI’s infrastructure and innovation cycle. Improved computing efficiency enables more effective training and inference, which in turn supports the development of more capable AI models. Better models lead to stronger products, increased user adoption, and greater revenue, allowing further investment in future infrastructure. This cycle is intended to make artificial intelligence increasingly capable, dependable, and affordable over time.

One of the most notable achievements associated with the project is the rapid development timeline. Jalapeño progressed from initial design to manufacturing tape-out in just nine months, which OpenAI and Broadcom believe represents one of the fastest application-specific integrated circuit (ASIC) development cycles in advanced semiconductor manufacturing. This accelerated schedule was made possible through close software-hardware collaboration and the use of OpenAI’s own AI models to assist in design and optimization processes. The company argues that AI-assisted chip design could reduce computing costs and expand access to advanced technologies across the industry.

Jalapeño serves as the foundation of a broader multi-generation compute platform that will combine OpenAI-designed accelerators with Broadcom’s networking and silicon technologies and Celestica’s systems expertise. Initial deployments are planned by the end of 2026, with further expansion expected in subsequent years. These deployments are intended to support large-scale data center operations and provide the computing capacity necessary for future AI systems.

Ultimately, the purpose of Jalapeño is to improve AI inference, which is the stage at which artificial intelligence directly serves users. Improvements in speed, reliability, and cost efficiency can translate into faster ChatGPT responses, more capable coding assistants, lower-cost API services, and more dependable access during periods of high demand. By reducing infrastructure costs and increasing computing efficiency, OpenAI aims to make advanced AI technologies more accessible to students, developers, researchers, businesses, and organizations worldwide.