OpenAI is collaborating with custom AI chips from Broadcom along with Nvidia and AMD

Weeks later Broadcomstock Shoot After unveiling a $10 billion customer that analysts said was OpenAI, the two companies made the partnership official.

OpenAI and Broadcom said Monday that they are jointly building and deploying 10 gigawatts of dedicated AI accelerators as part of a broader cross-industry effort to scale AI infrastructure.

Broadcom shares rose 7% on news of the deal.

They did not disclose the financial terms.

While the two companies have been working together for 18 months, they have now announced plans to develop and deploy racks of chips designed by OpenAI starting late next year. OpenAI has announced huge deals in recent weeks with… Nvidia, oracle and Advanced micro devicesas it attempts to secure the capital and computing needs needed for its historically ambitious AI construction plans.

“These things have become so complex that you need everything,” OpenAI CEO Sam Altman He said in a podcast with OpenAI and Broadcom executives that the companies released with the news.

The systems include networking, memory and compute – all dedicated to OpenAI workloads and built on Broadcom’s Ethernet stack. By designing its own chips, OpenAI can lower computing costs and scale its infrastructure further. Industry estimates put the cost of a 1 GW data center at about $50 billion, with $35 billion of that typically allocated to chips – based on Nvidia Current pricing.

Altman said the Broadcom deal provides “a tremendous amount of computing infrastructure to serve the world’s needs for advanced intelligence.” “We can get huge gains in efficiency, and that will lead to much better performance, faster models, cheaper models — all of it.”

Broadcom has been one of the biggest beneficiaries of the generative AI boom, with hyperscalers buying up its custom AI chips, which the company calls XPUs. Broadcom doesn’t name its largest web-scale customers, but analysts said dating back to last year that its top three customers were Google, dead And TikTok’s parent ByteDance.

Broadcom shares are up 40% this year after doubling in 2024, and the company’s market value has surpassed $1.5 trillion.

OpenAI President Greg Brockman said the company used its own models to speed up chip design and improve efficiency.

“We’ve been able to achieve tremendous reductions in space,” he said on the podcast. “You take components that humans have already optimized and add computing to them, and the model comes out with its own improvements.”

Broadcom CEO Hock Tan said in the same conversation that OpenAI is the company building the “most advanced” frontier models.

“You still need computational power — the best and most up-to-date computational power — as you move forward on the road map toward a better and better parametric model and toward superintelligence,” he said. “If you make your own chips, you control your own destiny.”

Altman noted that 10 gigawatts is just the beginning.

“Although it is much larger than the world has today, we expect that high-quality intelligence will be delivered very quickly and at a very low price – the world will absorb it very quickly and find amazing new things to use it for,” he said.

Today, OpenAI operates with just over 2 gigawatts of computing power.

This was enough to scale ChatGPT to what it is today, Altman said, as well as develop and launch the video creation service Sora and conduct a lot of AI research. But demand is rising.

OpenAI has announced nearly 33 gigawatts of compute commitments over the past three weeks across partnerships with Nvidia, Oracle, AMD, and Broadcom.

“If we had 30 gigawatts today with the quality of the current models, I think we would continue to saturate that relatively quickly in terms of what people would do,” he added.

He watches: China opens antitrust investigation into Qualcomm Autotalks deals