Meta Launches One of World’s Largest CPU‑Based AI Expansions with AWS Graviton5

Meta has signed a landmark deal with AWS to deploy tens of millions of Graviton5 cores, marking one of the largest CPU‑based AI infrastructure expansions to date. This partnership reflects a strategic shift toward powering agentic AI workloads—real‑time reasoning, code generation, and multi‑step orchestration—on purpose‑built silicon.

It is considered one of the world’s largest CPU‑based infrastructure expansions because Meta is deploying tens of millions of AWS Graviton5 cores at once—making it one of the biggest single CPU deployments ever announced by a major tech company. This scale places Meta among the largest Graviton customers globally and signals a fundamental shift in AI infrastructure design.

Meta’s initial rollout already involves tens of millions of Graviton cores, with flexibility to expand further. Few companies have ever announced CPU deployments at this magnitude.

The infrastructure is designed to support billions of AI interactions daily, powering agentic AI workloads such as reasoning, orchestration, and code generation.

Meta is now officially one of the largest AWS Graviton customers worldwide, surpassing most other enterprises that use Graviton for cloud workloads.

Why This Deal Matters

• Scale: Meta becomes one of the largest Graviton customers globally, starting with tens of millions of cores and room to expand.
• Shift in AI Infrastructure: While GPUs remain critical for training large models, agentic AI workloads are CPU‑intensive—requiring chips optimized for reasoning and orchestration rather than raw matrix multiplication.
• Energy Efficiency: Graviton5, built on 3‑nanometer technology, delivers up to 25% better performance than its predecessor while reducing environmental impact.

AWS Graviton5: Technical Highlights

AWS Graviton5 is Amazon Web Services’ latest generation of custom‑built ARM‑based CPUs, designed specifically for cloud workloads. It delivers up to 25% better performance than Graviton4, packs 192 cores per chip, and features a cache five times larger—making it one of the most powerful and energy‑efficient CPUs available for large‑scale AI and enterprise applications.   

• 192 cores with a cache 5× larger than the previous generation.
• 33% faster inter‑core communication, enabling higher bandwidth and reduced latency.
• Built on the AWS Nitro System, ensuring high performance, availability, and security.
• Supports Elastic Fabric Adapter (EFA) for low‑latency, high‑bandwidth communication—critical for distributed agentic AI tasks.

Meta’s Strategic Goals

• Diversification of Compute: Expanding to Graviton allows Meta to run CPU‑intensive workloads efficiently at scale.
• Agentic AI at Scale: Infrastructure capable of handling billions of interactions and coordinating multi‑step agent workflows.
• Sustainability: Leveraging Graviton’s efficiency aligns Meta’s AI expansion with sustainability targets.

Industry Implications

• For AWS: Demonstrates AWS’s ability to deliver custom silicon integrated with its full AI stack.
• For Meta: Positions Meta as a leader in agentic AI infrastructure with a hybrid compute strategy.
• For the Market: Signals a broader industry trend—CPU‑optimized chips complement GPU‑driven training.

Comparison: GPU vs. Graviton CPU Workloads

Aspect	GPU (e.g., Nvidia H100)	AWS Graviton5 CPU
Best For	Training large AI models	Agentic AI workloads (reasoning, orchestration, code generation)
Core Count	Thousands of parallel cores	192 high‑performance cores
Latency	Higher for multi‑step tasks	Lower, optimized for reasoning
Energy Efficiency	High but power‑hungry	25% better than previous gen, built for efficiency
Scalability	Clustered GPU farms	Tens of millions of CPU cores, distributed workloads

Expert Voices

• Nafea Bshara, Amazon VP: “This isn’t just about chips; it’s about giving customers the infrastructure foundation to build AI that scales to billions worldwide.”
• Santosh Janardhan, Meta: “Expanding to Graviton allows us to run CPU‑intensive workloads behind agentic AI with the performance and efficiency we need at scale.”

In short: Meta’s massive Graviton5 deployment with AWS signals a new chapter in AI infrastructure—where CPUs, not just GPUs, become central to powering agentic AI systems at global scale.