Amazon says Meta to use millions of AWS Graviton chips

Amazon said on April 24 that Meta will use millions of AWS Graviton chips for AI workloads, giving AWS a large customer for its in-house ARM-based processors, which are used for general computing and AI inference rather than model training.

The announcement follows Meta’s six-year US$10 billion cloud deal with Google Cloud in August 2025, although the company had previously relied mainly on AWS and also used Microsoft Azure.

Amazon is also expanding its custom AI chip business through Trainium. Earlier in April, Anthropic agreed to spend US$100 billion over 10 years on AWS workloads focused on those chips as Amazon added another US$5 billion to its investment in the startup.

🔗 Source: Amazon

🧠 Food for thought

Implications, context, and why it matters.

AI workloads are changing how cloud chips get used

The deal follows the rise of “agentic AI,” where AI systems handle multi-step work such as real-time reasoning and code generation 1.
Model training leans on graphics processing units (GPUs). Agentic workloads can lean harder on central processing units (CPUs), which has led some executives to call this a CPU “renaissance” in AI data centers 2, 1.
Meta plans to deploy “tens of millions” of Graviton5 processor cores in a multiyear deal worth billions of dollars 3, 4.
Meta has also committed to Nvidia’s Vera CPUs 3. It also partnered with Arm, the chip architecture company behind many mobile and cloud processors, on Arm’s artificial general intelligence (AGI) CPU for agentic AI infrastructure 3.

CPU competition is growing alongside GPUs

Demand for AI inference and agentic tasks is pushing CPUs, including Arm-based designs, into a larger role beside GPUs 2.
Blackwell-era GPUs were mostly paired with Nvidia’s Grace CPU, based on Arm Neoverse V2 5. Google may also design Axion CPUs as head nodes in tensor processing unit (TPU) clusters that power Gemini 5.
Cloud providers now compete with traditional chipmakers such as Nvidia while also buying from them 6. They sell custom CPUs like Graviton against chips sold broadly to outside customers, such as Nvidia’s Vera CPU 6.
Amazon Web Services can design chips in-house 4. It can also use manufacturers such as Taiwan Semiconductor Manufacturing Company (TSMC) to exert more control over performance, cost, and supply 4.