Meta Bets on Amazon Graviton Chips for AI Growth: A Multibillion-Dollar Shift in Data Center Strategy

Meta Platforms has signed a multiyear, multibillion-dollar deal with Amazon Web Services to use Amazon’s Graviton processors across parts of its AI infrastructure, according to reporting by Bloomberg.

The agreement covers deliveries totaling tens of millions of CPU cores for Meta’s data centers.

Graviton is AWS’s in-house Arm-based server chip line, designed to deliver strong performance per watt and lower total cost for large-scale cloud workloads. For Meta, the move signals a push to secure dependable compute as AI features expand across its apps and services.

Why CPUs still matter for AI?

While GPUs remain central for training large models, Meta is expected to use Graviton heavily for inference and AI-adjacent tasks such as running agents, orchestration, and supporting services.

These workloads can scale across vast fleets of efficient CPUs, where cost and power use are decisive.

The deal also reflects a broader industry trend: hyperscalers are increasingly designing their own chips to reduce reliance on a single supplier and better tailor hardware to their platforms.

Amazon positions Graviton as a price-performance alternative to conventional server CPUs and a complement to specialized AI accelerators.

A diversification play amid chip constraints

Meta has continued to buy large amounts of AI hardware from companies such as Nvidia and has also worked with AMD as it builds out next-generation capacity. Adding Graviton gives Meta another lever in procurement, especially as demand for AI compute strains supply chains and data center power budgets.

For AWS, landing Meta as a major Graviton customer is a high-profile validation of its custom silicon strategy at a moment when cloud providers are competing on both AI infrastructure and cost efficiency.

The partnership may influence how other large AI developers balance GPUs, custom accelerators, and CPU fleets in production.