Meta Partners With AWS on Graviton Chips to Power Agentic AI

Meta Partners With AWS on Graviton Chips to Power Agentic AI

Summary (EN)

Meta announced on April 24 that it has signed an agreement with Amazon Web Services to add tens of millions of AWS Graviton cores to its compute portfolio, with room to expand as its AI systems scale. The company said the deal extends a long-running AWS relationship and reflects a broader infrastructure strategy centered on diversification rather than reliance on a single hardware path. In Meta’s framing, the key issue is that agentic AI changes the mix of compute required: GPUs remain central for training large models, but many workloads behind autonomous, multi-step systems are highly CPU intensive. Meta highlighted reasoning, planning, orchestration, and execution-heavy tasks as the kinds of operations that require large volumes of general-purpose processing at high efficiency. AWS and Meta both described Graviton5 as well suited to these workloads because of its faster data processing, larger cache, and greater bandwidth. Meta said the deployment will help support the infrastructure behind Meta AI and other agentic experiences serving billions of users, while AWS positioned the agreement as evidence that purpose-built silicon and cloud AI services are becoming a foundational layer for large-scale agent systems. The announcement is significant because it is not a routine cloud contract. It is a concrete industry signal that CPU architecture, not only GPU supply, is becoming strategically important as frontier companies move from model training toward sustained, production-grade agentic inference and orchestration.

Summary (ZH)

Meta 于 4 月 24 日宣布，已与 Amazon Web Services 达成协议，将把数千万个 AWS Graviton 核心纳入自身算力组合，并保留未来继续扩容的空间，以支撑其不断扩大的 AI 系统需求。Meta 表示，这项合作建立在双方长期关系之上，也体现了其更广泛的基础设施策略，即通过多元化算力来源来降低单一路径依赖。按照 Meta 的说法，agentic AI 正在改变算力结构需求。GPU 仍然是大模型训练的核心，但在自主、多步骤 AI 系统的实际运行中，大量工作负载更依赖 CPU，例如实时推理、任务规划、流程编排与执行控制等。Meta 与 AWS 都强调，Graviton5 由于具备更快的数据处理能力、更大的缓存和更高带宽，适合承担这类高强度 CPU 型任务。Meta 称，这批 Graviton 核心将用于支撑 Meta AI 及面向数十亿用户的 agentic 体验背后的基础设施；AWS 则把这项合作描述为一个信号，说明定制芯片与完整云端 AI 栈正成为大规模代理系统的基础层。此次公告的重要性不在于普通云采购，而在于它明确显示，随着前沿公司从模型训练转向持续性的生产级 agentic 推理与编排，CPU 架构本身也正在成为与 GPU 供应同样关键的战略资源。

Source

https://about.fb.com/news/2026/04/meta-partners-with-aws-on-graviton-chips-to-power-agentic-ai/