- 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 execution of multi-step tasks.
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 one key difference: Meta doesn’t just buy chips; is buying up all the infrastructure around them. It’s a wholesale deal, 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.
Article continues below.
Why agent AI is driving massive CPU demand
While GPUs remain essential for training large AI models, the rise of agent AI is creating massive demand for CPU-intensive workloads.
Agent systems perform real-time reasoning, code generation, lookup operations, and multi-step task orchestration, all of which rely heavily on CPU power.
Graviton5 is designed specifically for these workloads and offers faster data processing and higher bandwidth than general-purpose alternatives.
The chip features 192 Arm Neoverse V3 cores, 600 megabytes of total cache, and support for DDR5 8800 and PCIe Gen6 memory.
Meta will use these chips to support its AI tools that require handling billions of interactions while coordinating complex, multi-step agent workflows.
This deal could be seen as a blow to AMD and Intel, two companies that have traditionally dominated the large-scale infrastructure CPU market.
Meta is not only purchasing tens of millions of Graviton cores, but also the power, data center space, networking, and AWS management tools that surround those cores.
This means that Meta is choosing AWS’s vertically integrated infrastructure instead of purchasing off-the-shelf chips and integrating them into its own data centers.
How the Graviton infrastructure works
Graviton5 is based 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 storage devices and family networks.
This allows Meta to run its own virtual machines without compromising performance.
AWS designs its chips from scratch and controls the entire process, from chip design to server architecture.
Therefore, it can optimize performance and efficiency in ways that regular processors cannot match.
Meta appreciates that capability because it means chips can be tuned to achieve specific performance levels.
As Meta’s AI capability grows, deployment will likely extend beyond the initial tens of millions of cores.
The deal between Meta and AWS is clearly important, but it’s important to distinguish between buying infrastructure and buying chips.
Meta will not accept Graviton processors for installation in its own data centers; is leasing compute capacity hosted on AWS on a large scale.
“It’s not just about chips; it’s about providing customers with the infrastructure foundation, as well as data and inference services, to build AI that efficiently understands, anticipates, and scales billions of people around the world,” said Nafea Bshara, vice president and distinguished engineer at Amazon.
Follow TechRadar on Google News and add us as a preferred source to receive news, reviews and opinions from our experts in your feeds.
