OpenAI & Broadcom's 10GW AI Accelerator Partnership

OpenAI has announced a major partnership with Broadcom to design and roll out 10GW of custom AI accelerators, a move that signals a major change in how AI infrastructure is developed for global data centres.

The collaboration merges OpenAI’s deep expertise in creating large language models (LLMs) with Broadcom’s extensive capabilities in semiconductor manufacturing and integration.

Under the terms of the deal, OpenAI will lead the chip design, embedding insights from its powerful models directly into the hardware.

Broadcom will be responsible for the manufacturing and integration of the new systems.

This venture places OpenAI among a growing number of AI firms seeking to develop their own chips to enhance performance and reduce reliance on traditional hardware suppliers.

The scale of AI infrastructure

The deployment is set on an ambitious timeline with the first systems scheduled to arrive in the second half of 2026 and a full rollout across OpenAI’s own facilities and partner data centres expected by the end of 2029.

The 10GW figure represents a vast amount of computing power. For scale, 1 gigawatt is enough energy to power approximately 700,000 homes in the US.

These accelerators, which are specialised processors designed to efficiently handle the complex mathematical calculations required by AI models, are fundamental to advancing AI capabilities.

The global distribution of these chips highlights the immense computational demand caused by the widespread adoption of AI technologies.

OpenAI’s in-house chip design

For OpenAI, this decision marks a strategic pivot. Having previously relied on external vendors for its hardware needs, OpenAI is now taking control of its silicon design.

The stated goal is to create more efficient and powerful hardware by tailoring it to the specific demands of its AI models, such as GPT-4.

“Partnering with Broadcom is a critical step in building the infrastructure needed to unlock AI’s potential and deliver real benefits for people and businesses,” says Sam Altman, OpenAI’s Co-founder and CEO.

“Developing our own accelerators adds to the broader ecosystem of partners all building the capacity required to push the frontier of AI to provide benefits to all humanity.”

Greg Brockman, President of OpenAI, explains that the move is vital for innovation.

“By building our own chip, we can embed what we’ve learned from creating frontier models and products directly into the hardware, unlocking new levels of capability and intelligence,” he says.

With a user base now exceeding 800 million people per week through its various products, the demand for computing resources at OpenAI has increased and custom-designed chips could be a key part of meeting that demand.

Broadcom’s Ethernet-based approach

The partnership also highlights a key technical decision in the AI infrastructure landscape. The new rack systems will utilise Ethernet networking, a widely adopted standard for connecting computers.

This choice comes amid a debate over the best networking solution for high-performance computing, with some favouring alternatives like InfiniBand.

Hock Tan, Broadcom’s President and CEO, sees the agreement as a strong endorsement of Ethernet’s role in the future of AI.

“OpenAI has been at the head of the AI revolution since the ChatGPT moment – and we are thrilled to co-develop and deploy 10GW of next-generation accelerators and network systems to pave the way for the future of AI,” he says.

The systems will feature Broadcom’s complete suite of connectivity technologies, including Ethernet switches, PCIe and optical links that enable high-speed data transfer between racks.

Charlie Kawwas, President of Broadcom’s Semiconductor Solutions Group, explains how this integration provides an optimised solution.

“Custom accelerators combine remarkably well with standards-based Ethernet scale-up and scale-out networking solutions to provide cost and performance optimised next generation AI infrastructure,” Charlie says.

He adds: “The racks include Broadcom’s end-to-end portfolio of Ethernet, PCIe and optical connectivity solutions, reaffirming our AI infrastructure portfolio leadership.”

The two companies have already formalised agreements for co-development and supply and have now added a term sheet that covers the deployment of these systems into production.

“Our partnership with OpenAI continues to set new industry benchmarks for the design and deployment of open, scalable and power-efficient AI clusters,” Charlie says.