OpenAI and Broadcom Partner on Custom AI Accelerators

In a major development for the AI infrastructure sector, OpenAI has announced a partnership with Broadcom to design and deploy 10GW of custom AI accelerators.

The collaboration aims to embed insights from models at the level of GPT-4 directly into silicon for use across global data centres.

This partnership brings together OpenAI's work in building large language models with Broadcom's extensive semiconductor manufacturing capabilities.

Under the agreement, OpenAI is set to handle the chip design, while Broadcom will manage the production and integration of the hardware.

The decision for OpenAI to design its own hardware marks a change in its strategy, as OpenAI has historically relied on external suppliers to power its infrastructure.

The scale of custom AI accelerators

The first systems from the OpenAI and Broadcom partnership are expected in the latter half of 2026, with a full rollout planned for completion by the end of 2029.

The 10GW figure represents a substantial amount of computing power.

For context, 1GW is enough energy to power approximately 700,000 homes in the US.

These accelerators, which are specialised processors designed to handle the mathematical operations fundamental to AI models, will be distributed across OpenAI’s own facilities and the data centres of its partners worldwide.

The move comes as many companies in the AI field explore designing their own chips to enhance performance and reduce dependence on traditional suppliers.

For OpenAI, whose ChatGPT and enterprise products now serve more than 800 million people weekly, the demand for computing resources is enormous.

Developing custom chips is viewed as a key part of the solution to meet this demand.

Embedding model insights into silicon

By taking chip design in-house, OpenAI believes it can achieve superior performance by integrating what it has learned from building advanced models directly into the hardware.

Sam Altman, OpenAI’s Co-founder and CEO, explains the strategic importance of the collaboration.

“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,” Sam says.

“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, further elaborates on the technical advantages.

“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.

The strategic choice of ethernet networking

This partnership also highlights a key technical decision in AI infrastructure design.

The new rack systems will utilise Ethernet networking, a widely adopted standard for connecting computers, rather than alternatives like InfiniBand, which has often been used in high-performance computing clusters.

Hock Tan, Broadcom’s President and CEO, sees the agreement as a validation of this approach.

“OpenAI has been a leader in 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 incorporate Broadcom’s complete suite of connectivity technology, including Ethernet switches and optical links that facilitate high-speed data transfer between racks.

Charlie Kawwas, President of Broadcom’s Semiconductor Solutions Group, says: “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. ”

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

OpenAI and Broadcom have already finalised agreements for co-development and supply and have now added a term sheet that covers the deployment of these systems into production.

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