Meta taps AWS Graviton at scale, deploying tens of millions of cores

Meta Platforms has signed a multi-year agreement with Amazon Web Services (AWS) to deploy AWS Graviton processors at scale, starting with tens of millions of cores and with scope to expand as its AI workloads grow. The deal is positioned to support agentic AI and inference workloads where highly parallel CPU capacity complements GPU-based training systems.

According to company statements and the official release, the arrangement covers Graviton-class CPU capacity that AWS will allocate to Meta over multiple years and carries a multi-billion-dollar scale profile, though neither party disclosed an exact headline figure in the initial announcement. The move follows Meta’s recent large cloud commitments: roughly $21 billion with CoreWeave and up to $27 billion with Nebius, deals that together total approximately $48 billion of pledged AI infrastructure spending.

Markets reacted with modest optimism for AWS’s commercial trajectory: Amazon shares ticked higher as investors priced in stronger uptake of AWS’s custom silicon stack (Graviton, Trainium) and the potential revenue lift from selling specialized CPU-based capacity to hyperscalers and large AI labs. For Meta, the agreement reduces reliance on a single vendor mix for inference and increases flexibility to route workloads between in-house accelerators, dedicated neocloud capacity and AWS-hosted Graviton instances.

Strategically, the deal underlines a broader industry trend: constrained GPU supply and rising costs are pushing major AI users to diversify compute across CPUs (efficient inference), custom accelerators and specialist GPU clouds. Arm-based Graviton processors are being marketed for energy efficiency and cost-effective scale in inference layers, while GPU providers and custom XPU efforts continue to focus on training performance and high-throughput workloads. The Meta-AWS pact therefore reflects both tactical capacity needs and a longer-term architecture shift in hyperscale AI deployment.

Analysts say the agreement is a pragmatic step to secure near-term capacity and manage TCO for large-scale AI services, but caution that performance benchmarks, workload placement and contract terms will determine the financial payoff. Investors will watch subsequent quarterly disclosures for capex cadence, service-level details and any impact on incumbent partners; in the near term, the meta-story is clear: Meta is diversifying its compute suppliers while accelerating an aggressive AI scaling roadmap.