Meta, AWS sign billion-dollar cloud contracts… Invest in 'Graviton5' for AI agents

Meta Platforms has signed a multi-year, multi-billion-dollar contract with Amazon Web Services ( AWS ). The core of the contract is AWS’s self-developed CPU “Graviton” series, with Meta obtaining cloud infrastructure needed to operate large-scale AI agents.

According to Bloomberg, through this contract, Meta can acquire tens of millions of Graviton cores and has the option to further increase the quantity based on future demand. Meta plans to use these resources to drive “agent-based AI.” AI agents refer to forms of artificial intelligence that go beyond simple chatbots, capable of connecting with external applications and automatically executing tasks.

Features and Performance of Graviton5

The most talked-about chip in this contract is AWS’s latest CPU “Graviton5,” released last December. The chip is manufactured using 3-nanometer process technology, equipped with 192 cores, and based on Arm Holdings’ instruction set architecture. Instruction set architecture is essentially the “language” that defines how semiconductors perform calculations. Arm architecture has advantages in vector and matrix computations, which are considered beneficial for AI tasks.

AWS states that Graviton5 offers up to a 25% performance improvement over the previous generation of its own CPUs. The performance boost is backed by a large-scale expansion of the L3 cache. L3 cache is a memory space located near processor cores that improves processing speed by shortening data transfer distances. According to AWS, Graviton5’s L3 cache is five times larger than that of the previous generation.

The Role of CPUs in AI Infrastructure

CPUs still play an important role in AI servers. While large-scale neural network computations are primarily handled by graphics processing units ( GPU ), CPUs coordinate these GPUs and perform various auxiliary tasks. Especially since Meta’s AI agents often need to connect to external tools and services, processing demands around CPUs are significant. This contract is seen as a signal that Meta is becoming more sophisticated not only in GPU deployment but also in overall AI infrastructure design.

Graviton5 is designed to work in conjunction with AWS’s “Nitro System.” Nitro is an architecture that offloads part of infrastructure management to independent accelerators, allowing CPUs to dedicate more resources to customer applications. AWS also explains that the “Nitro Isolation Engine” can verify whether workloads from different customers are separated, ensuring security and efficiency in multi-tenant environments.

Meta’s Long-term AI Strategy

Meta Infrastructure Head Santosh Janardan stated, “AWS is our long-trusted cloud partner. By expanding the use of Graviton, we can operate CPU-intensive AI agent workloads with performance and efficiency suitable for Meta’s scale.”

This announcement is notable as it is the second major CPU contract Meta has announced in the past month. Previously, Meta decided to adopt Arm’s new AGI CPU and announced plans to collaborate on various next-generation chip designs. Market analysts believe Meta is diversifying its AI infrastructure strategy beyond reliance on GPUs, expanding into CPUs, custom chips, and cloud resources, preparing for long-term competition.

TP AI Notes: Summarized using TokenPost.ai-based language model. The article may omit key content or be inconsistent with facts.