Meta and Amazon reach a multibillion-dollar, multiyear deal for Meta to rent hundreds of thousands of Amazon's Graviton chips for its AI inference needs
BloombergMatt Day
Context & Ripple Effects
Meta’s Amazon arrangement follows its separate multiyear commitment to Nvidia Blackwell and Rubin GPUs, where related coverage said its in-house AI-chip effort faced technical challenges. The combined record points to Meta securing external capacity across different chip suppliers rather than relying on a single internal or merchant source.
The deal also sits alongside Meta’s debt, SPV, and data-center financing activity, making the choice to rent long-duration inference capacity notable: AI infrastructure is being funded and procured through a mix of owned facilities and contracted supply.
First-order effects
Meta gains a large, multiyear source of Graviton capacity for serving AI inference, reducing its near-term dependence on scarce Nvidia chips for that workload.
Amazon converts its custom-chip and infrastructure capacity into a major external commitment from a hyperscale peer, with Meta becoming a significant customer rather than solely a rival platform operator.
Second-order effects
Meta’s split procurement—Nvidia GPUs alongside Amazon Graviton—raises the value of software and model stacks that can operate across heterogeneous accelerators, rather than being optimized for one chip family.
The agreement gives other large AI operators a concrete example of renting specialized capacity from a cloud rival, increasing pressure on cloud and chip providers to offer credible alternatives for inference workloads.
Third-order effects
If such contracts recur, inference capacity could develop into a more explicit market for long-term, committed supply, separate from the race to own training clusters.
The pattern would shift AI infrastructure competition toward a hybrid model of proprietary data centers, external capacity contracts, and increasingly financialized project structures; its durability depends on whether sustained inference demand justifies those commitments.
The trend: AI companies are treating inference compute as strategic, long-duration capacity and diversifying supply beyond Nvidia-centric training infrastructure.
Meta is deploying tens of millions of Arm-based Graviton cores with AWS. As AI shifts toward agentic systems and real-time reasoning, CPU performance is becoming critical at scale—and Arm is built for it.
Another exciting development in our chips business as Meta has decided to bet big on Graviton, our leading CPU chip—committing to tens of millions of Graviton cores. Agentic AI is becoming almost as big a CPU story as a GPU story. Complex multi-step orchestration, real-time rea…
Meta Platforms $META and Amazon Web Services $AMZN just announced a new agreement to “bring tens of millions of AWS Graviton cores into Meta's compute portfolio” “This agreement builds on Meta's long-standing relationship with AWS, and expanding to include Amazon's custom [image]
Today we're announcing an agreement with @awscloud to bring tens of millions of AWS Graviton cores into @Meta 's compute portfolio to support next-generation agentic AI workloads. https://about.fb.com/...
AI has fundamentally changed the equation around compute. Compute is now a competitive differentiator — a make-or-break capability for any company serious about building AI at scale. …
Our Graviton and Nitro devices in the headlines again.... “The deployment starts with tens of millions of Graviton cores, with the potential to expand. …