Why is Meta using AWS Graviton for AI?

Meta turns to Amazon Graviton to power AI

Meta has struck multiple deals with Amazon to obtain more computing capacity for its AI workloads, including renting large numbers of AWS Graviton CPU chips. The move matters because it reflects how chip availability and cost pressures are reshaping the AI hardware stack.

In the reported arrangement, Meta will use “tens of millions” of AWS Graviton cores for AI model development and delivery, aligning with a broader trend of companies diversifying beyond GPUs during periods of constrained supply and rising costs. Separate reporting also frames the Graviton strategy as a response to ongoing Nvidia chip shortages—an issue that has affected many AI startups and enterprise teams.

Meta’s choice of Graviton is notable because it is CPU-based rather than GPU-based. That suggests Meta is looking to optimize parts of its AI pipeline for different compute types, potentially using CPUs for specific inference, training components, or infrastructure needs where Gravitons can offer favorable economics.

The practical implications are:

• Reduced dependence on GPU supply for at least some workload categories
• Cost and scalability benefits tied to AWS’s capacity and Meta’s scale
• More heterogeneity in how major AI systems are provisioned

While these deals focus on compute, the wider story is that AI execution has become a procurement and infrastructure game as much as a model-development one. Even companies with strong in-house research and engineering capabilities still need reliable hardware access to ship new products and keep inference costs under control.

In short, Meta’s Graviton plan is an infrastructure response—aimed at meeting demand for AI at a time when the most visible accelerators remain scarce and expensive.