Meta Secures AWS Graviton CPUs for Agentic AI

Meta signed a three-year AWS contract to run tens of millions of Graviton5 cores for AI. Valued at billions, the deal makes Meta a top Graviton customer and follows a layoff of about 8,000 staff.

CNBC

GPU clusters dominate model training, but AI agents need CPUs for inference. Meta has used Nvidia GPUs via AWS, signed $48 billion of GPU contracts with CoreWeave and Nebius, and a $10 billion Google Cloud pact. The Graviton deal adds CPU capacity for real-time reasoning, code generation and multi-step coordination.

Amazon AI CPUs Power Meta’s Future: Massive Graviton Deal Shifts Chip Landscape

AWS will supply tens of millions of Graviton5 chips, each with 192 cores and a cache five times larger than the previous generation. Amazon says they use up to 60 % less energy per compute unit. AWS’s silicon business earns over $20 billion annually and could reach $50 billion. Meta plans $115-$135 billion AI spending this year; Intel reports Xeon demand outstripping supply.

GeekWire

Santosh Janardhan, Meta’s head of infrastructure, said Graviton provides the performance and efficiency needed for agentic AI. AWS vice-president Nafea Bshara noted Graviton’s broad pre-training adoption. CEO Andy Jassy called agentic AI “almost as big a CPU story as a GPU story.” Intel CEO Lip-Bu Tan observed CPU demand exceeds supply, marking a resurgence.

CNBCGeekWireMeta signs deal to deploy AWS Graviton chips for AI workloadsMeta Becomes One of Largest Customers of Amazon AI Chips

Intel’s Xeon shortage signals a bottleneck for CPU-heavy AI workloads. Nvidia’s Vera CPU, also ARM-based, competes with Graviton for agentic AI, showing a crowded market. Meta’s reliance on a single provider for billions of cores could raise supply-chain concerns, though it also uses Google Cloud and Azure.

Amazon AI CPUs Power Meta’s Future: Massive Graviton Deal Shifts Chip Landscape

The deal confirms CPUs as a primary tier for AI inference and could shift spend from GPU-centric to ARM CPUs. AWS’s chip revenue boost strengthens its stance against Google Cloud and Azure, while pressuring Nvidia and Intel to accelerate AI-focused CPU development. Meta’s diversified compute strategy underscores the growing role of agentic AI in future data-center design.

Amazon AI CPUs Power Meta’s Future: Massive Graviton Deal Shifts Chip Landscape

• “Graviton is one of the most used platforms for pre training by a lot of foundation model companies, and Meta is now one the newest one,” — Nafea Bshara, AWS Vice President & Distinguished Engineer

CNBC

• “Expanding to Graviton allows us to run the CPU-intensive workloads behind agentic AI with the performance and efficiency we need at our scale,” — Santosh Janardhan, Meta Head of Infrastructure

CNBC

• “becoming almost as big a CPU story as a GPU story.” — Andy Jassy, Amazon CEO

GeekWire

• “For the last few years, the story around high performance computing was almost exclusively about GPU and other accelerators,” — Lip-Bu Tan, Intel CEO

CNBC

Meta will expand Graviton use as AI models grow; AWS seeks more customers.

CNBC