By Belle Lin
Meta Platforms and Amazon.com agreed to a multibillion-dollar deal over several years in which the social-media company will use tens of millions of Amazon Web Services' Graviton chip cores to support its AI agents and other AI initiatives.
The companies declined to disclose the financial terms and exact duration of the deal. Nafea Bshara, an Amazon vice president and distinguished engineer, said the length of the deal is between three and five years.
Bshara, a co-founder of AWS's in-house chip unit Annapurna Labs, said most of the tens of millions of AWS Graviton cores will be located in the U.S.
The news comes as tech giants and artificial intelligence labs are still scrambling to get as much compute capacity as possible to support their AI goals. Meta's deal with AWS is one of several it has announced with chip companies this year, including Nvidia, Advanced Micro Devices and Arm Holdings, and underscores the need for a diversity of chips to support AI, analysts say.
"Meta has, as you can imagine, access to so many options from the supply side. But they chose Graviton5, our 3-nanometer chip, for price performance," Amazon's Bshara said.
Meta signed this agreement for AWS's Graviton central processing units, or CPUs -- a category of chip that had largely been left behind as the AI world focused its attention on graphics processing units, or GPUs.
Now, thanks to the rise of AI agents, CPUs are once again becoming a hot commodity. And companies like Intel, which specialize in producing CPUs, are benefiting from the trend.
Meta said the new deal reflects its diversified approach to infrastructure, and shows that no single chip architecture can efficiently serve every computational task.
CPUs are useful for running certain applications and feeding that activity back to GPUs -- or in other words, the two chips work well together for a variety of tasks that AI agents can accomplish, said Brendan Burke, a research director at Futurum Group focused on semiconductors.
CPUs are also a key part of the "post-training" step for large language models, or the part of the AI model-building process where pretrained models are more actively trained toward certain goals.
This isn't the first time Meta and AWS have worked together. The partnership between the two tech giants dates back to about 2016, Bshara said, but had mostly involved core cloud services, use of Amazon's Bedrock platform and Meta renting GPU clusters from AWS.
Meta, which agreed to buy the AI startup Manus for more than $2 billion in December, clearly has AI agent ambitions -- and those drive demand for CPUs in their data centers, Burke said. Manus develops an AI agent that can carry out sophisticated tasks.
"It's clear that for the largest frontier labs, there is no limit to their demand for CPUs," Burke said.
To streamline its operations and pay for its massive investments in AI, Meta said Thursday that it will lay off 10% of its staff, or roughly 8,000 people, in May. The company released its first new AI model in a year, called Muse Spark, earlier this month, and said it has plans for more releases soon.
For AWS, the deal with Meta serves as a proof point that its homegrown Graviton CPUs are highly useful for powering AI, according to Futurum's Burke. The agreement makes Meta one of AWS's five largest Graviton customers, Bshara said.
Earlier this week, Amazon announced an additional $5 billion investment in Anthropic, which included the AI company using tens of millions of AWS Graviton CPU cores.
AWS has been building its own in-house-designed silicon since well before 2018, when it released the first version of Graviton based on processor architecture from Arm.
But AWS isn't only making bets on its own chips. In March, the company announced a deal with Cerebras where it will deploy the startup's inference chips inside its data centers.
Write to Belle Lin at belle.lin@wsj.com
(END) Dow Jones Newswires
April 24, 2026 08:00 ET (12:00 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments