- Meta commits to tens of millions of hosted Graviton cores
- The deal includes infrastructure, networking, power, and management layers
- Graviton5 is designed for sustained processing and multi-step task execution
Meta has signed an agreement to deploy tens of millions of AWS Graviton Arm cores, making it one of the largest Graviton customers in the world.
The deal marks a major expansion of the long-standing partnership between Meta and AWS, but with a critical difference: Meta is not just buying chips; it is buying the entire infrastructure around them. It is a wholesale agreement, not a hardware purchase.
"As we scale the infrastructure behind Meta's AI ambitions, diversifying our compute sources is a strategic imperative," said Santosh Janardhan, head of infrastructure at Meta.
Why agentic AI is driving massive demand for CPUs
While GPUs remain essential for training large AI models, the rise of agentic AI is creating massive demand for CPU-intensive workloads.
Agentic systems perform real-time reasoning, code generation, search operations, and orchestration of multi-step tasks, all of which rely heavily on CPU power.
Graviton5 is purpose-built for these workloads, offering faster data processing and greater bandwidth than general-purpose alternatives.
The chip features 192 Arm Neoverse V3 cores, 600 megabytes of total cache, and support for DDR5 8800 memory and PCIe Gen6.
Meta will use these chips to support its AI tools that require handling billions of interactions while coordinating complex, multi-step agent workflows.
This agreement could be seen as a major blow to AMD and Intel, two companies that have traditionally dominated the CPU market for large-scale infrastructure.
Meta is not only buying tens of millions of Graviton cores, but also the power, data center space, networking, and AWS management tools surrounding those cores.
This means Meta is choosing AWS's vertically integrated infrastructure over buying off-the-shelf chips and integrating them into its own data centers.
How the Graviton infrastructure works
Graviton5 is built on 3nm chip technology, a manufacturing process that produces smaller, more efficient processors.
It features a Nitro system that enables bare metal instances while providing familiar networking and storage devices.
This allows Meta to run its own virtual machines without performance compromises.
AWS designs its chips from the ground up and controls the full process from chip design through server architecture.
Therefore, it can optimize performance and efficiency in ways that regular processors cannot match.
Meta appreciates such capacity because this means the chips can be tweaked to achieve specific performance levels.
As Meta’s AI capability grows, the deployment will probably extend beyond the initial tens of millions of cores.
The agreement between Meta and AWS is clearly important, but it is important to distinguish between buying infrastructure and buying chips.
Meta is not taking delivery of Graviton processors to install in its own data centers; it is renting AWS-hosted compute capacity on a massive scale.
“This isn’t just about chips; it’s about giving customers the infrastructure foundation, as well as data and inference services, to build AI that understands, anticipates, and scales efficiently to billions of people worldwide,” said Nafea Bshara, Vice President and Distinguished Engineer, Amazon.
