Meta announced Friday that it has signed an agreement to deploy AWS Graviton processors at scale, a deal that spans multiple years and is worth billions of dollars according to AWS. The partnership makes Meta one of the largest Graviton customers in the world and represents a significant expansion of a long-standing relationship between the two companies.
The deployment begins with tens of millions of Graviton5 cores, with flexibility to expand as Meta's AI capabilities grow. Meta's Head of Infrastructure, Santosh Janardhan, said the partnership reflects the company's strategy of diversifying compute sources. The Graviton5 chip features 192 cores and is built on 3-nanometer chip technology from Taiwan Semiconductor Manufacturing Co.
The Agentic AI Angle
What makes this deal interesting is where those chips are going: agentic AI workloads. While GPUs remain essential for training large models, the rise of autonomous AI systems that reason, plan, and execute multi-step tasks is creating massive demand for CPU-intensive processing. Real-time reasoning, code generation, search orchestration, and agent coordination are all tasks that lean heavily on CPUs rather than GPUs.
Graviton5 is purpose-built for these workloads. AWS designed the chip with a cache five times larger than the previous generation, which reduces inter-core latency by up to 33 percent. That architecture matters for agentic systems that need to continuously process and coordinate tasks across many processors working in tandem.
Amazon's Nafea Bshara, Vice President and Distinguished Engineer, framed the deal as providing Meta with infrastructure to build AI that scales to billions of people worldwide. The Graviton5 instances also support the Elastic Fabric Adapter for low-latency, high-bandwidth communication between instances, which is essential for large-scale distributed agent workloads.
Meta's Multi-Vendor Strategy
This AWS deal is the latest piece in Meta's aggressive infrastructure diversification. As the era of agentic AI scales up, Meta has been locking down compute from every major silicon provider.
In February 2026, Meta expanded its partnership with Nvidia in a multi-year deal covering millions of Blackwell and Vera Rubin GPUs, plus the first large-scale deployment of standalone Grace CPUs. The company plans to deploy Nvidia's Vera CPU-only servers in 2027. A week later, Meta signed a 6-gigawatt agreement with AMD for custom MI450 GPUs across multiple generations, a deal potentially worth over $100 billion that includes a warrant for Meta to acquire roughly 10 percent of AMD's equity.
Meta has been explicit that it needs Nvidia, AMD, AWS, and its own custom silicon to support different workloads. The company is guiding capital expenditures of $115 to $135 billion for 2026, much of it directed toward AI infrastructure. The AWS deal adds another layer to that stack.
The Broader CPU Renaissance
The partnership arrives as CPU demand is surging across the industry. Earlier this month, Amazon CEO Andy Jassy revealed that two large AWS customers had asked to buy out all available 2026 instance capacity for Graviton. He declined to preserve capacity for other customers. Intel reported better-than-expected first quarter earnings this week and noted that CPU prices are rising as demand soars.
Nvidia recognized this shift too. The company is pushing standalone CPU products, with its Vera chip designed specifically for AI agentic workloads. AWS only sells access to Graviton through its cloud service, whereas Nvidia sells chips to enterprises and cloud providers directly. These are competing approaches to a market that barely existed two years ago.
For AWS, landing Meta as a flagship Graviton customer is validation for its custom silicon strategy. Google is also pushing adoption of its own chips, and the hyperscaler chip wars are intensifying. More than 50 percent of new CPU capacity added by AWS now runs on Graviton. Adobe, Epic Games, and Pinterest were early adopters. Meta is the biggest name yet.
Meta operates its own data centers and develops custom hardware while also partnering with cloud providers for specific capabilities. This deal reflects that hybrid approach. The company is not all-in on any single vendor. It is, instead, building the most aggressive multi-vendor compute strategy in the industry.
