Meta Bets on Amazon's Graviton5 to Power the Agentic AI Era
News Summary
Meta and Amazon Sign Multibillion-Dollar Deal to Power Agentic AI with Graviton5 Chips
On April 24, 2026 (Eastern Time), Meta announced a landmark multibillion-dollar, multi-year agreement with Amazon Web Services (AWS) to deploy tens of millions of Amazon's Graviton5 ARM CPU cores across AWS data centers. The partnership makes Meta one of the largest Graviton customers in the world and marks a significant milestone in the infrastructure evolution powering the next generation of artificial intelligence.
What Is Being Built and Why It Matters
While graphics processing units (GPUs) have long been the backbone of training large AI models, a new category of AI workloads is rapidly emerging: agentic AI. These are systems that can reason in real time, generate code, conduct searches, and coordinate complex multi-step tasks autonomously. Such workloads are highly CPU-intensive rather than GPU-intensive, requiring massive parallel processing power with low-latency communication between compute nodes.
AWS Graviton5 was engineered precisely for this type of demand. Built on cutting-edge 3-nanometer chip technology, Graviton5 features 192 cores per chip and a cache five times larger than its predecessor. These architectural advances reduce inter-core communication delays by up to 33% and deliver up to 25% better overall performance compared to the previous Graviton generation. The chips also support the Elastic Fabric Adapter (EFA), a networking technology that enables low-latency, high-bandwidth communication between instances — critical for distributing large-scale agentic AI tasks across thousands of processors working in tight coordination.
Meta's Strategic Infrastructure Push
Meta has been aggressively scaling its AI infrastructure as part of a broader strategic pivot toward AI-first products and services. The company has disclosed a capital expenditure plan exceeding $135 billion for 2026, underscoring the scale of its ambitions. This AWS agreement complements Meta's existing partnerships with Nvidia and AMD, expanding the diversity of its compute portfolio to match different workload profiles.
Santosh Janardhan, Meta's head of infrastructure, described the rationale clearly: "As we scale the infrastructure behind Meta's AI ambitions, diversifying our compute sources is a strategic imperative." By integrating Graviton5 into its stack alongside traditional GPU clusters, Meta can route CPU-bound agentic workloads to the most efficient hardware available, reducing cost per inference and improving latency for end users.
The majority of the Graviton5 capacity will be deployed in United States-based AWS data centers. This domestic deployment focus aligns with Meta's efforts to ensure resilience and regulatory compliance for its AI infrastructure.
The Rise of Agentic AI as a Compute Driver
The Meta-AWS deal reflects a broader industry shift in how AI compute is being purchased and deployed. The first wave of AI infrastructure spending was dominated by GPU procurement for model training. The second wave, now underway, is centered on inference and agentic AI — applications where models are not just trained once but run continuously, handling real-time requests, reasoning through multi-step problems, and interacting with users and systems dynamically.
This shift creates a more heterogeneous compute landscape. Companies like Meta need not only GPU farms but also vast pools of high-performance CPU capacity capable of handling the orchestration layers of agentic systems. CPU-based cloud compute, particularly with chips as capable as Graviton5, is becoming a first-class citizen in AI infrastructure planning.
Amazon's Growing Role in AI Infrastructure
For Amazon Web Services, the deal represents a significant validation of the Graviton chip line as a serious contender in AI workloads. AWS has invested heavily in custom silicon — including the Graviton series for general-purpose compute and Trainium/Inferentia chips for AI-specific tasks — as part of its strategy to offer differentiated, cost-effective cloud services.
Securing Meta, one of the world's largest AI spenders, as a major Graviton customer strengthens AWS's position in the competitive cloud market. It also signals to the broader industry that CPU-optimized custom silicon has a defined and growing role in modern AI deployments, not merely as a support role behind GPUs but as a primary compute tier for entire classes of production workloads.
Looking Ahead
The Meta-AWS Graviton5 partnership is expected to shape how other large AI developers approach infrastructure planning. As agentic AI capabilities advance and production deployments scale, the demand for CPU-dense, low-latency cloud compute will only grow. The deal sets a precedent for how hyperscalers and AI platform companies can collaborate at scale, combining custom silicon innovation with the operational demands of frontier AI systems.
Both companies characterized the agreement as a long-term, strategic partnership — a signal that the collaboration is designed to evolve alongside the technology rather than serve a single fixed deployment milestone.