Meta Deploys AWS Graviton Chips to Power Agentic AI Workloads

Meta has entered into a significant agreement with Amazon Web Services (AWS) to deploy AWS Graviton processors at scale, marking a major expansion of the two companies' long-standing partnership. The deployment, which begins with tens of millions of Graviton cores, is designed to support the massive infrastructure requirements behind Meta’s next generation of artificial intelligence. By integrating these custom chips, Meta aims to power the CPU-intensive workloads essential for its agentic AI efforts.

Powering Agentic AI Workloads

While GPUs remain essential for training large-scale models, the rise of agentic AI—autonomous systems capable of reasoning, planning, and executing multi-step tasks—has created a surge in demand for high-performance, CPU-intensive compute. Meta’s AI initiatives require infrastructure capable of handling billions of interactions, including real-time reasoning, code generation, and search orchestration.
Graviton5 processors are purpose-built to handle these complex, multi-step workflows efficiently. Featuring 192 cores and a cache five times larger than previous generations, the chip reduces communication delays between cores by up to 33%. This architecture provides the high bandwidth and fast data processing necessary for AI systems that must continuously reason through and coordinate complex tasks.

Strategic Infrastructure Expansion

For Meta, diversifying compute sources is a strategic imperative as it scales its AI ambitions. Santosh Janardhan, head of infrastructure at Meta, noted that expanding to Graviton allows the company to run the CPU-intensive workloads behind agentic AI with the performance and efficiency required at its scale. The deployment leverages the AWS Nitro System, which provides dedicated hardware and software to deliver high performance, availability, and security, while enabling bare-metal instances that allow Meta to run its own virtual machines without performance compromises.
The partnership also utilizes the Elastic Fabric Adapter (EFA), which enables low-latency, high-bandwidth communication between instances. This capability is critical for Meta’s agentic AI workloads, where large-scale tasks must be distributed across many processors working in coordination.

Efficiency and Sustainability

AWS Graviton5 is built on 3-nanometer chip technology, a manufacturing process that enables smaller, more efficient processors. Because Amazon designs its chips from the ground up, it can optimize performance and server architecture in ways that off-the-shelf processors cannot match. Graviton5 delivers up to 25% better performance than the previous generation, helping Meta pursue its AI goals while maintaining its sustainability targets.
Nafea Bshara, vice president and distinguished engineer at Amazon, emphasized that the collaboration provides the infrastructure foundation, data, and inference services necessary to build AI that scales to billions of people worldwide. By combining purpose-built silicon with the full AWS AI stack, the partnership signals a new chapter in how large-scale AI infrastructure is constructed to deliver smarter, more personalized experiences.