Skip to content

Thursday, August 27, 2026

Gigantum.net
Artificial intelligence

Amazon to Buy 2 Million Nvidia Chips for Data Center Build-Out

Amazon.com Inc. will add an additional 2 million Nvidia Corp. graphics processing units to its data center fleet in the next two years, a sign that the compa...

· 265 words

(Bloomberg) -- Amazon.com Inc. will add an additional 2 million Nvidia Corp. graphics processing units to its data center fleet in the next two years, a sign that the company remains committed to the AI hardware leader's products despite its own competitive chipmaking efforts.

The two companies said on Wednesday that Amazon would deploy some 2 million of Nvidia's high-end chips, which are designed to train and run artificial intelligence models, in 2027 and 2028. That's in addition to the 1 million Nvidia chips Amazon in March said that it would be installing in Amazon Web Services data centers, beginning this year.

The chips in the latest deal include Nvidia Blackwell Ultra, Rubin and Rubin Ultra GPUs. Those products will likely carry a list price of at least tens of thousands of dollars a unit, though big customers can get discounts. Amazon and Nvidia didn't announce financial terms of their latest deal.

Nvidia's GPUs are the essential currency of the artificial intelligence boom, powering the assembly and use of most large language models.

Nvidia's largest and best customers, Amazon, Microsoft Corp. and Alphabet Inc.'s Google, are also building alternative products. The cloud giants are betting that they can better tailor hardware to their own data centers and pass efficiency gains on to their customers.

Amazon in recent years has invested heavily in its own chipmaking arm, Annapurna Labs, which produces a central processing unit, as well as an AI accelerator designed to rival Nvidia's GPUs. The company has said the upcoming edition of its Trainium AI chip will make use of an Nvidia networking technology.

Gathered from external sources. Rights to this text belong to whoever originally published it.