OpenAI has inked $10 billion deals to date: Here is why it sidelined Google, Amazon, and Intel

OpenAI has already locked in $10 billion worth of chip and cloud deals. Not one of those puts Intel, Google, or Amazon in the driver’s seat. It didn’t happen by chance. It happened because OpenAI doesn’t want to rely on any of them. It wants power spread out. It wants to scale fast. It wants control.

Back in November, after Nvidia crushed earnings, Jensen Huang told investors, “Everything that OpenAI does runs on Nvidia today.” That’s true for now. But it’s not going to stay that way much longer.

The same week, OpenAI went out and signed a $10 billion deal with Cerebras, a much smaller chipmaker that’s trying to go public. This wasn’t just another deal. It was part of a bigger play: use new players, get more chips, build faster, depend on no one.

OpenAI taps Cerebras, Broadcom, AMD to spread chip bets wide

The Cerebras deal is just one piece. OpenAI said it will use 750 megawatts of Cerebras chips across phases that run through 2028. These chips will help run its large models and heavier workloads.

This comes on top of last year’s $1.4 trillion infrastructure spree, where it teamed up with Nvidia, AMD, and Broadcom. That’s what pushed OpenAI’s valuation to $500 billion in private markets.

In September, Jensen committed $100 billion from Nvidia to help build out 10 gigawatts of systems for OpenAI. That’s the same energy used by 8 million homes in a year. Jensen said that it would need 4 to 5 million GPUs. But OpenAI isn’t betting everything on him. Just hours after that was announced, it revealed another 10 gigawatts worth of chips coming from Broadcom. These aren’t standard chips. These are custom AI accelerators, called XPUs. Broadcom’s been working on them with OpenAI for over a year.

The Broadcom deal blew up on Wall Street. The stock went flying. Broadcom is now worth over $1.6 trillion. That’s what happens when OpenAI gives you a seat at the table.

Google, Amazon, and Intel left out as OpenAI builds its own chip stack

Meanwhile, Amazon, Google, and Intel are barely in the frame.OpenAI did sign a $38 billion cloud deal with Amazon Web Services in November. It will run workloads on AWS data centers. Amazon also said it will build new ones for OpenAI.

And sure, Amazon may put more than $10 billion into the company, but there’s still no commitment to use Inferentia or Trainium, Amazon’s in-house chips. Talks are ongoing, but nothing’s locked.

Google Cloud is also providing capacity under a deal signed last year. But when asked about using Google’s tensor processing units, OpenAI said no. It’s not interested. Not even with Broadcom helping make those chips.

Then there’s Intel. Reuters said the company had the chance years ago to invest in OpenAI and supply chips. It passed. Now it’s trailing everyone.

In October, Intel tried to catch up. It showed off a chip called Crescent Island. It’s meant for AI inference and will offer higher memory and better energy use. But real sampling won’t even start until late 2026.

To stay alive in AI, Intel had to take money from Nvidia and the U.S. government. Wall Street will see next week if that’s made any difference when Intel kicks off tech earnings.

If you’re reading this, you’re already ahead. Stay there with our newsletter.