- OpenAI partners with Broadcom to unveil “Jalapeño,” its first custom AI inference chip, designed in a record nine months to slash computing costs.
In a milestone shift from software developer to hardware architect, OpenAI has officially stepped into the physical semiconductor arena. Partnering with networking and silicon giant Broadcom, the ChatGPT creator unveiled Jalapeño, its first custom-designed “Intelligence Processor.”
The announcement signals a major escalation in OpenAI’s strategy to verticalize its computing stack. By creating proprietary Application-Specific Integrated Circuits (ASICs), OpenAI aims to systematically reduce its reliance on general-purpose GPUs and lower the skyrocketing costs associated with running frontier artificial intelligence models at scale.
Architected from a “Blank Slate”
Unlike traditional graphics processors that have been retrofitted to handle AI workloads, Jalapeño was built from scratch. According to technical details released in an official OpenAI Blog Post, the chip is a “blank-slate design” engineered specifically for Large Language Model (LLM) inference, the computational process of executing responses after an AI model has already been trained.
The processor’s core architecture zeroes in on the biggest bottleneck in modern AI data centers: data movement. By heavily optimizing the physical layout to minimize data transfer between logic components and off-chip memory, Jalapeño balances compute, memory, and networking resources. This tailored engineering allows the silicon to operate at utilization rates remarkably close to its theoretical peak performance.
According to certain reports, the inference stack natively integrates Broadcom’s ultra-high-speed Tomahawk 6 networking chip, boasting an impressive 1.6 Terabits per second (Tbps) throughput. Lab samples are already operating at production target frequency, successfully running internal next-generation workloads like GPT-5.3-Codex-Spark.
The Nine-Month Design Loop
Perhaps the most disruptive aspect of the announcement is the speed at which the chip was developed. OpenAI and Broadcom compressed a traditional multi-year advanced semiconductor design cycle into just nine months from initial concept to manufacturing tape-out.
This record-shattering timeline was achieved through tight software-hardware co-design and a self-referential engineering loop. OpenAI utilized its own advanced language models to automate and accelerate complex parts of the silicon layout, optimization, and testing phases. Representatives noted that using current AI tools to design the physical infrastructure for future AI represents a major paradigm shift that could drastically lower engineering barriers across the chip industry.
Slashing Infrastructure Costs
The financial implications of the partnership are massive. Broadcom CEO Hock Tan indicated that the custom chip framework could yield roughly 50% cost savings compared to current commercial AI graphics processors. This economic relief is vital for OpenAI as it scales interactive user features, complex coding agents, and enterprise API layers globally.
The manufacturing phase of the project highlights a highly collaborative, multi-company infrastructure play. While Broadcom contributes silicon implementation and networking technologies, electronic manufacturer Celestica will oversee the complex board, rack, and system integration. Mass production is expected to be handled by Taiwan Semiconductor Manufacturing Company (TSMC), as detailed in reports from semiconductor tracker AnySilicon.
Scaling for a Gigawatt Future
The collaboration marks the beginning of a multi-generation hardware roadmap. OpenAI plans to deploy the chips at an unprecedented gigawatt scale across major data centers, leaning on its primary cloud partner, Microsoft, with initial field deployments scheduled to begin by the end of 2026.
By designing its own hardware underbelly, spanning kernels, memory architectures, and silicon layouts, OpenAI is solidifying its footprint alongside hyperscale giants like Google, Amazon, and Meta. Jalapeño represents a definitive step toward making advanced, real-time AI agents structurally sustainable and financially accessible to hundreds of millions of users worldwide.
About the Author
Jennifer Sakmufuwo Baba is a tech analyst and writer covering artificial intelligence, fintech, and emerging technologies at TechRegard. Based in Nigeria, she's passionate about translating complex tech developments into compelling, accessible stories for diverse audiences. Her work focuses on how technology shapes innovation across Africa and globally.
