The multiyear partnership will deliver 10 gigawatts of custom accelerator racks, signaling OpenAI’s growing ambition to control its AI hardware stack.
A Major Hardware Move for OpenAI
OpenAI has officially partnered with Broadcom to build 10 gigawatts’ worth of custom AI accelerator hardware, set to roll out between 2026 and 2029. This collaboration marks a significant step in OpenAI’s strategy to vertically integrate its AI infrastructure, embedding model learnings directly into the chips that run them.
- The hardware will be deployed across OpenAI’s own data centers and those of its partners.
- By designing custom chips, OpenAI aims to optimize performance, reduce latency, and push the boundaries of model intelligence and efficiency.
Embedding AI Knowledge Into Silicon
According to OpenAI, this partnership will allow the company to translate its deep insights from building frontier AI models directly into hardware design.
- This approach enables tighter integration between model architecture and the infrastructure that powers it.
- It reflects a trend among AI leaders—like Google with its TPUs and Amazon with Trainium—to develop application-specific hardware for machine learning tasks.
The scale of the deal suggests OpenAI is planning for significant model growth and data center expansion over the next half-decade.
A Price Tag in the Hundreds of Billions
Though neither company disclosed the financial terms, the Financial Times estimates the Broadcom deal could cost OpenAI between $350 billion and $500 billion over the multi-year rollout.
- If accurate, this would make it one of the largest infrastructure investments in AI history.
- The massive figure underscores the capital intensity of training and deploying next-generation AI models at scale.
Part of a Larger Infrastructure Offensive
This isn’t OpenAI’s only big bet on hardware. In recent weeks, the company has:
- Secured 6 gigawatts of chips from AMD, worth tens of billions of dollars.
- Signed a $300 billion cloud infrastructure deal (reportedly) with Oracle, though not confirmed by either party.
- Received support from Nvidia, which announced a $100 billion investment and a letter of intent to provide 10 gigawatts of hardware.
These moves show OpenAI is hedging across multiple vendors while building out an AI supercomputing ecosystem robust enough to support AGI-scale workloads.
What This Means for the AI Industry
OpenAI’s hardware push signals a new phase of AI development where compute is no longer just rented—it’s engineered from the ground up.
- It positions OpenAI as not just a software innovator but a full-stack AI platform, from chips to models to applications.
- The partnership could accelerate development of more capable, efficient, and secure AI systems, while reshaping the hardware supply chain.
For Broadcom, the deal affirms its increasing relevance in the AI accelerator market, traditionally dominated by Nvidia and, more recently, AMD.
