OpenAI and Broadcom have unveiled Jalapeño, OpenAI’s first Intelligence Processor designed to accelerate large language model (LLM) inference, marking a significant expansion of OpenAI’s strategy to control the full stack of AI development. The chip progressed from initial design to production in just nine months, a fast turnaround aided by OpenAI’s own models. The companies plan to deploy the accelerator at a gigawatt scale with data center partners, signaling substantial investment and anticipated demand for the new technology. “Jalapeño is part of our long-term full-stack infrastructure strategy to make compute more abundant,” said Greg Brockman, President and Co-Founder of OpenAI, “resulting in AI which is faster, more reliable, and more affordable for people and businesses.”
Jalapeño Chip: LLM-Optimized Inference Architecture
The unveiling of the Jalapeño chip marks a decisive move by OpenAI beyond model development and into custom hardware design, signaling a full-stack strategy to control the entire AI infrastructure. Developed in nine months, from initial design to manufacturing tape-out, this first Intelligence Processor represents a fast advanced semiconductor development cycle. This accelerated timeline was aided by the application of OpenAI’s own models to optimize aspects of the chip’s design process, demonstrating a potential feedback loop where AI refines the tools used to create it. Early testing indicates that Jalapeño delivers improved performance per watt compared to current accelerators, a critical metric for the energy-intensive task of running large language models.
The chip’s architecture prioritizes minimizing data movement and balancing compute resources, aiming for realized performance closer to theoretical peaks. Broadcom’s silicon implementation and Tomahawk networking technologies are central to achieving this at scale. Richard Ho, who leads OpenAI’s hardware program, explained that the chip was “designed from the ground up for LLM inference using detailed insights from our close collaboration with OpenAI researchers,” optimizing for the specific needs of frontier AI models. This is not an adaptation of existing hardware, but a blank-slate design focused on the demands of modern LLM inference, promising lower latency and improved efficiency for interactive AI products.
Jalapeño was designed from the ground up for LLM inference using detailed insights from our close collaboration with OpenAI researchers.
Richard Ho, who leads OpenAI’s hardware program
Nine-Month Tape-Out Driven by OpenAI Models
The artificial intelligence hardware market is currently dominated by established players adapting existing chip architectures for the demands of large language models. OpenAI is disrupting this with a vertically integrated approach, extending its control from model development to silicon design. This strategy culminated in the swift production of Jalapeño, OpenAI’s first Intelligence Processor, delivered to company leadership in nine months, a timeline that significantly reduces typical advanced semiconductor development cycles. This co-development effort leveraged Broadcom’s silicon implementation expertise and Celestica’s system integration capabilities, but was fundamentally driven by OpenAI’s deep understanding of LLM requirements.
The world is moving to a compute-powered economy.
Greg Brockman, President and Co-Founder of OpenAI
Broadcom Silicon Enables Gigawatt-Scale Deployment
Broadcom is playing a pivotal role in translating OpenAI’s AI strategy into physical infrastructure, evidenced by the rapid development and impending deployment of the Jalapeño chip. The collaboration isn’t simply about manufacturing; it’s about building a cohesive, vertically integrated system, from model design to silicon implementation. This speed demonstrates a new paradigm where AI assists in the creation of the hardware that powers it, potentially lowering compute costs industry-wide. The scale of this undertaking is significant; OpenAI intends to deploy Jalapeño with data center partners, signaling a substantial investment and expectation for robust demand.
Broadcom’s contribution extends beyond chip fabrication to encompass networking technologies, including Tomahawk silicon, essential for large-scale production and data transfer. “Our collaboration with OpenAI represents a fundamental commitment to scaling the physical infrastructure required for the next decade of AI,” stated Hock Tan, President and CEO of Broadcom, highlighting the long-term nature of the partnership. OpenAI’s approach focuses on optimizing every layer of the stack, from chip architecture to serving systems, with the goal of creating AI that is. The architecture prioritizes minimizing data movement and maximizing resource utilization, aiming for performance levels approaching theoretical peaks and ultimately, broader access to advanced AI capabilities.
Our collaboration with OpenAI represents a fundamental commitment to scaling the physical infrastructure required for the next decade of AI.
Hock Tan, President and CEO, Broadcom
Full-Stack Integration Improves AI Efficiency & Access
OpenAI’s unveiling of the Jalapeño Intelligence Processor signals a decisive move toward vertically integrated AI infrastructure, with implications for both performance and accessibility. Beyond designing models, the company is now directly involved in chip fabrication, a strategy intended to optimize the entire AI pipeline from algorithm to hardware. This full-stack approach, developed in partnership with Broadcom and Celestica, aims to address limitations in existing accelerator technology and deliver gains in efficiency. Engineering samples are already demonstrating promising results, with early testing indicating performance per watt significantly exceeding current accelerators while running ML workloads at production target frequency and power, including GPT‑5, 3‑Codex, and Spark.
