Meta has signed a deal with Amazon Web Services (AWS) to deploy tens of millions of Graviton processor cores across its AI infrastructure, in what Bloomberg reports is a multibillion-dollar, multiyear agreement that makes the Facebook owner one of the largest Graviton customers in the world.
The deployment centres on AWS's latest Graviton5 chip, a 192-core Arm-based processor with a cache five times larger than the previous generation and up to 33% lower inter-core latency.
The deal reflects a broader shift in AI infrastructure away from an exclusive focus on GPUs towards the CPU-intensive workloads generated by agentic AI systems, which must reason, plan and execute complex multi-step tasks in real time.
"As we scale the infrastructure behind Meta's AI ambitions, diversifying our compute sources is a strategic imperative," said Santosh Janardhan, Meta's head of infrastructure.
"Expanding to Graviton allows us to run the CPU-intensive workloads behind agentic AI with the performance and efficiency we need at our scale."
While GPUs remain essential for training large models, agentic workloads such as real-time reasoning, code generation, search and orchestrating multi-agent systems are increasingly CPU-bound.
The agreement builds on a long-standing relationship between Meta and AWS, including Meta's use of Amazon Bedrock, the cloud provider's platform for building generative AI applications, where Meta is already one of the largest customers.
Meta said it will blend its own data centres and custom hardware with cloud partners to match chip architectures to specific workloads, rather than relying on a single platform.
The initial deployment starts at tens of millions of cores with flexibility to expand as Meta's AI capabilities grow.
Notably, Meta is buying not just the chips but the full surrounding infrastructure, including power, data centre capacity and networking, hosted by AWS rather than integrated into Meta's own facilities.
Nafea Bshara, vice president and distinguished engineer at Amazon, said Graviton is already widely used for pre-training by foundation model companies and that Meta is now the newest major customer at scale.
"Meta's expanded partnership, deploying tens of millions of Graviton cores, shows what happens when you combine purpose-built silicon with the full AWS AI stack to power the next generation of agentic AI," Bshara said.
Related reading
- Meta breaks ground on Tulsa AI data center
- UK transport department uses Google Cloud AI to cut consultation analysis from months to hours
- AWS chief defends OpenAI investment as routine partner conflict after backing Anthropic
AWS says Graviton delivers the best price-to-performance ratio of any compute option available through its EC2 service while using 60% less energy than comparable instances.
The deal comes in the same week that Anthropic announced plans to use Graviton processors and as Amazon's chip ambitions increasingly compete with Nvidia's new Vera CPU, also Arm-based and designed for agentic workloads.
The recap
- Meta will add AWS Graviton cores to its compute portfolio.
- Initial deployment starts with tens of millions of cores.
- Deployment can expand as Meta's AI capabilities grow.
