Meta Signs Agreement with Amazon Web Services to Operate Artificial Intelligence

SadaNews - Meta is working on developing a new generation of artificial intelligence technologies and creating unprecedented experiences for users. To achieve this, the company requires a new generation of processors, which Amazon Web Services (AWS) provides. An agreement has been signed between the two companies to utilize tens of millions of AWS Graviton processors, reflecting a fundamental shift in how AI infrastructure is built.

Graviton 5 Processor

Previously, Graphics Processing Units (GPUs) were essential for training large models. However, the evolution of Agentic AI (software units capable of thinking, planning, and performing complex tasks as though a user is sitting in front of a computer) means there is an increasing need for intense CPU resource demands for workloads such as code generation, search, and coordinating multi-step tasks. The Graviton5 processor has been specifically designed for these tasks, granting Meta the computing capabilities necessary for efficient and reliable operation.

Faster Data Processing

The new processor features 192 cores and five times larger cache memory compared to the previous generation, reducing the response time of the cores by up to 33 percent, which translates to faster data processing with larger bandwidth—both essential conditions for AI systems that require continuous multi-step task execution. The processor is manufactured with 3-nanometer precision to achieve a smaller size and higher efficiency, resulting in infrastructure performance improvements of up to 25 percent compared to the previous generation, while maintaining superior energy efficiency to minimize environmental impact.

Highest Levels of Performance and Security

The processors are designed on the AWS Nitro system, which employs dedicated electronic circuits and software to achieve the highest levels of performance, availability, and security. This system allows for direct hardware access while providing advanced technologies such as Elastic Network Adapter (ENA) and Amazon Elastic Block Store (EBS), facilitating a private virtual environment with no impact on performance.

The processor also supports Elastic Fabric Adapter (EFA) technology, which enables low-latency, high-capacity connections between multiple virtual environments, a crucial factor for running Meta's agentic AI workloads that require the distribution of broad tasks across a coordinated set of processors working harmoniously toward a single user goal.