When two of the world’s largest technology companies sign a multibillion-dollar, multiyear chip deal, it is worth paying attention. On April 24, 2026, Meta and Amazon Web Services announced that Meta would adopt hundreds of thousands of AWS Graviton chips to power its growing AI agent workloads. The deal — worth billions of dollars and spanning multiple years — signals something bigger than a procurement contract. It tells you where enterprise AI is actually going.

What Was Announced

Meta will become one of the largest users of Amazon’s Graviton chips, with AWS initially providing tens of millions of Graviton cores. The latest generation, Graviton5, carries 192 cores per chip, a cache five times larger than its predecessor, and delivers up to 25% better performance while reducing inter-core communication delays by up to 33%.

The deal is not about training large AI models. Meta has plenty of GPU infrastructure for that. This agreement is specifically aimed at agentic AI workloads: real-time reasoning, code generation, search, and the orchestration of multi-step tasks that AI agents run continuously at scale.

Why CPU, Not GPU?

This is the part that matters most for anyone thinking about AI strategy.

Training large AI models is a GPU problem. It requires massive parallel computation to process enormous datasets. But once a model is trained, running it in production — especially for agents that need to respond quickly, coordinate with other systems, and complete tasks in real time — looks very different. That workload is CPU-intensive, not GPU-intensive.

As AI agents have moved from demo to deployment, the industry has hit this reality hard. GPU clouds are expensive and optimised for batch training. Agents need fast, always-on, low-latency compute. Graviton5 was designed with exactly that in mind.

Meta’s deal follows roughly $48 billion in AI infrastructure commitments the company has made in recent weeks with CoreWeave and Nebius for GPU capacity. Combined, these moves sketch a clear picture: Meta is building a two-track infrastructure. GPUs for training and development. Optimised CPUs for running agents in production.

The Signal for the Broader Market

Meta and AWS are not alone. Across the industry, the shift from model training to model deployment is driving new infrastructure investment. Google, Microsoft, and Oracle have all made significant moves to build out inference-optimised capacity. The common thread is agentic AI — the idea that AI systems will run continuously, take actions, and complete complex tasks on behalf of businesses and individuals.

This is not a distant prediction. It is happening now. The infrastructure race underway is the technology industry’s bet that millions of businesses will be running AI agents as a core part of their operations within the next few years.

What This Means for Business

The gap between what big tech is building and what most businesses have deployed is significant. While Meta and Amazon are finalising billion-dollar chip agreements to run agents at scale, the majority of small and mid-sized businesses are still evaluating whether to move beyond basic AI tools.

That gap is not permanent. When the infrastructure gets built, costs come down. The same economics that made cloud computing accessible to businesses of every size will play out for AI agents. The Graviton deal is part of what makes that possible.

The practical implication is this: businesses that are building their AI agent strategy now will be well-positioned when the infrastructure makes deployment cheaper and more reliable. Those that wait until the economics are obvious will be catching up to competitors who moved earlier.

The questions every business owner should be asking are straightforward. What repetitive, high-volume work could an AI agent handle? Where does your team spend time on tasks that follow a pattern? What would change if those tasks ran automatically, around the clock, without headcount?

The infrastructure to support those answers is being built at a scale and speed that the Meta-AWS deal makes concrete. The question is whether your business will be ready to use it.

---

For a deeper walkthrough of tools like this and how they fit together, the free Working With Claude field guide covers the ecosystem end to end. Get the guide.