Meta Partners With AWS to Deploy Tens of Millions of Graviton Cores for Agentic AI

# Meta Partners With AWS to Deploy Tens of Millions of Graviton Cores for Agentic AI

# Meta Partners With AWS to Deploy Tens of Millions of Graviton Cores for Agentic AI

Meta Platforms has signed a multibillion-dollar agreement with Amazon Web Services to deploy tens of millions of Graviton processor cores, making the social media giant one of the largest Graviton customers in the world. The deal, announced on April 24, 2026, signals a fundamental shift in how the AI industry thinks about compute: the rise of agentic AI is driving explosive demand not just for GPUs, but for the CPU-intensive workloads that power autonomous reasoning, code generation, and multi-step task orchestration.

The agreement centers on AWS Graviton5, Amazon's latest Arm-based processor built on a 3-nanometer process. Each Graviton5 chip packs 192 cores and features a cache five times larger than its predecessor, reducing inter-core communication latency by up to 33 percent. For agentic AI systems that must continuously reason through and execute complex multi-step tasks in real time, that combination of bandwidth and low latency is critical. The initial deployment starts with tens of millions of cores across hundreds of thousands of chips, with flexibility to scale further. The deal runs for at least three years.

A Strategic Imperative for Meta

Santosh Janardhan, Meta's Head of Infrastructure, framed the partnership as a matter of strategic necessity. "AWS has been a trusted cloud partner for years, and expanding to Graviton allows us to run the CPU-intensive workloads behind agentic AI with the performance and efficiency we need at our scale," Janardhan said in a statement accompanying the announcement.

The deal reflects Meta's recognition that agentic AI workloads have a fundamentally different compute profile than the large language model training that has dominated infrastructure spending to date. While Nvidia GPUs remain indispensable for training frontier models, the inference and orchestration layers powering autonomous agents demand massive CPU throughput. Meta is betting that Graviton's combination of price-performance and energy efficiency, with AWS claiming 60 percent lower energy consumption than comparable alternatives, gives it an edge at the scale required to serve billions of users.

Nafea Bshara, Amazon Vice President and Distinguished Engineer who helped architect the Graviton line, emphasized the competitive dynamics behind Meta's choice. "Meta has, as you can imagine, access to so many options from the supply side. But they chose Graviton5, our 3-nanometer chip, for price performance," Bshara said, underscoring that the deal was won on technical merit rather than relationship alone.

Part of a Massive Infrastructure Spending Spree

The AWS partnership is the latest in a string of infrastructure commitments that underscore Meta's determination to lead in agentic AI. In recent weeks alone, the company committed $21 billion to CoreWeave and up to $27 billion to Nebius, a combined $48 billion directed at specialized AI cloud providers. Meta's total 2026 capital expenditure guidance stands at a staggering $115 billion to $135 billion, nearly double the $72 billion it spent in 2025.

While the precise dollar value of the AWS Graviton deal has not been disclosed, multiple reports describe it as a multibillion-dollar arrangement. Combined with its GPU investments including a reported $50 billion commitment to Nvidia and $60 billion to AMD, Meta's overall AI infrastructure procurement now exceeds $200 billion across multiple vendors and chip architectures.

That diversification is deliberate. By spreading workloads across Nvidia GPUs for training, custom MTIA silicon for inference, and now AWS Graviton for CPU-intensive agentic tasks, Meta is building a heterogeneous compute stack designed to optimize both performance and cost at unprecedented scale.

Why This Matters: The Agentic Compute Shift

The Meta-AWS deal is significant beyond its raw numbers because it crystallizes an industry-wide inflection point. The first wave of AI infrastructure spending was almost entirely GPU-centric, driven by the race to train ever-larger language models. But as the industry pivots from building models to deploying agents, the compute bottleneck is shifting.

Agentic AI systems do not simply generate text; they reason, plan, execute code, query databases, call APIs, and coordinate across tools in real time. Those workloads are CPU-intensive and latency-sensitive in ways that differ fundamentally from batch training on GPU clusters. Meta's willingness to commit tens of millions of CPU cores specifically for agentic workloads suggests the company sees agent deployment, not model training, as the defining infrastructure challenge of the next several years.

AWS CEO Matt Garman has argued that "task-accomplishing agents deliver more than just content generation, and enterprises will see massive returns in 2026." Meta's Graviton bet is a concrete wager on exactly that thesis.

What to Watch Next

The deal raises several questions worth tracking. First, whether other hyperscalers and AI companies follow Meta's lead in building dedicated CPU capacity for agentic workloads, or whether GPU-first architectures continue to dominate. Second, how the Graviton5's real-world performance compares to alternatives from Intel and AMD as these deployments scale. And third, whether Meta's total infrastructure spending, now approaching a quarter-trillion dollars across all partners, begins to generate the revenue returns that justify such extraordinary capital commitment. With the three-year deal window extending into 2029, the answers will unfold over the coming quarters as Meta's agentic AI products move from development to deployment at global scale.