Meta strikes multi-billion-dollar Graviton5 deal with AWS

TL;DR:

• Meta will use “tens of millions of cores” of AWS Graviton5 CPU chips in a multi-year, multi-billion-dollar agreement.
• AWS’s Graviton5 is its fifth-generation in-house CPU, with each chip carrying 192 cores; AWS has been building the line since 2018, sourcing directly from TSMC.
• For UK enterprise architects, the deal is another concrete data point that the AI infrastructure stack is widening — not narrowing — beyond Nvidia GPUs as inference workloads scale.

Meta Platforms has signed a multi-year deal worth billions of dollars with Amazon Web Services to use AWS’s in-house Graviton5 central processors, the companies said on Friday. AWS distinguished engineer and vice-president Nafea Bshara told Reuters the deal will cover “tens of millions of cores” worth of chips. Each Graviton5 carries 192 cores that can be assigned to different tasks.

The agreement reinforces the pattern emerging in this week’s other infrastructure stories. Where Nvidia GPUs remain essential for training the biggest AI models, the actual deployment of those models — inference at production scale — increasingly runs on CPUs. Intel’s first-quarter results, also reported on Friday, attributed a 24% share-price jump and a record valuation to exactly the same dynamic, with HSBC pointing specifically to rising demand for Xeon server CPUs in AI data centres.

A more crowded chip stack at Meta

The Graviton5 commitment widens what was already a multi-vendor Meta chip stack. The company has previously signed large deals with Nvidia and AMD, and worked closely with Arm Holdings on Arm’s new CPU. “As we scale the infrastructure behind Meta’s AI ambitions, diversifying our compute sources is a strategic imperative,” Meta head of infrastructure Santosh Janardhan said in a statement.

AWS produces Graviton CPUs through Taiwan Semiconductor Manufacturing Company, having internalised the line since 2018. Bshara said the company passes the cost savings on to customers — which is the part of the announcement most relevant beyond Meta.

Looking forward

For UK enterprise customers, the deal is one of the clearer recent signals that the AI infrastructure market is splitting into more layers. The training-side narrative remains Nvidia-dominated. The inference-side narrative is now visibly multi-vendor: Intel and AMD on x86, Arm and AWS on Arm, with Nvidia entering the CPU market itself. That gives buyers more procurement levers — and more decision complexity — than 12 months ago. Meta’s deal also illustrates that hyperscaler-to-hyperscaler chip purchases at the multi-billion-dollar scale are now part of normal AI infrastructure operations, which has implications for any UK firm modelling its own cloud cost trajectory on assumptions of stable per-core pricing.