Amazon (AMZN) Tariff Bets on CPUs as AI Agent Demand Surges

TLDR

• Meta and Amazon have struck a multi-billion-dollar deal for Meta to use AWS Graviton CPU cores for AI.
• The deal covers tens of millions of Graviton cores, most of which are located in the United States, over three to five years.
• Meta stock rose 0.6% and Amazon stock rose 1.4% in pre-market trading after the announcement.
• The deal adds Amazon to Meta’s existing list of chip suppliers, which includes Nvidia, Broadcom and AMD.
• The agreement makes Meta one of AWS’ top five Graviton customers.

---

Meta and Amazon have agreed to a multibillion-dollar deal that will see Meta use tens of millions of Amazon Web Services’ Graviton CPU cores to support AI agent efforts.

$dead It is expanding its partnership with AWS to deploy tens of millions of Graviton cores, making it one of Amazon’s partnerships $AMZN Graviton’s largest customer. The chips will help run CPU-heavy AI workloads such as reasoning, searching, and multi-step task coordination. pic.twitter.com/kygaILDGRh

– Wall Street Engine (@wallstengine) April 24, 2026

The term of the deal is between three and five years, according to Nafie Bishara, Amazon vice president and co-founder of Annapurna Labs, AWS’s internal chip unit. Most Graviton cores will be deployed in the United States

Amazon stock rose 1.4% in premarket trading after the news broke, while Meta stock rose 0.6%.

Amazon.com, Inc., AMZN

The agreement is specifically for AWS Graviton CPUs — not GPUs. Central processing units (CPUs) used to take a back seat as AI development focused largely on graphics processing unit (GPU) computation, but that’s changing.

The emergence of artificial intelligence agents has brought a refocus on central processing units (CPUs). They handle specific tasks and feed the activity of the GPUs, making the two types of chips complementary for a wide range of AI workloads.

Central processing units (CPUs) also play a key role in the “post-training” phase of large language models, where pre-trained models are optimized to achieve specific goals.

He chose dead Amazon Graviton5, a 3nm chip, citing the price-performance ratio. “Meta has access to many options on the supply side. But they chose Graviton5,” Bishara said.

Meta strategy for diversified chips

This deal adds Amazon to a list of suppliers that already includes Nvidia, Broadcom, and AMDAnd the holding arm. Meta has made it clear that it doesn’t want to rely on any single chip architecture.

“No single chip architecture can efficiently serve every computational task,” Meta said in a statement linked to the announcement.

The first deployment will start with tens of millions of Graviton cores, with room to expand as Meta’s AI needs grow.

Meta’s AI Push and cost reductions

dead AI ambitions run deep. The company acquired artificial intelligence startup Manus for more than $2 billion in December. Manus builds AI agents capable of performing complex tasks, creating increased demand on the CPU.

To help fund the AI ​​buildout, Meta also announced Thursday that it will cut nearly 10% of its workforce — about 8,000 employees — in May.

Meta released its first new AI model in a year, the Muse Spark, earlier this month, and has hinted at more releases to come.

The Meta-AWS relationship goes back to about 2016, but has mostly involved cloud services, the Bedrock platform, and GPU cluster leasing. This deal represents a clear expansion into custom silicon.

For AWS, having Meta as a Graviton customer is a useful endorsement. Earlier this week, Amazon also announced a $5 billion investment in Anthropic, which similarly includes tens of millions of Graviton CPU cores.

AWS has been designing its own chips since before 2018, when the first Graviton was released on the Arm architecture.

Bishara confirmed that the deal places Meta among AWS’s five largest Graviton customers.