Meta signs agreement with AWS to power agentic AI on Amazon's Graviton chips

Meta announced an expanded partnership with Amazon Web Services under which Meta will deploy "tens of millions" of AWS Graviton processor cores to run agentic AI workloads, including code generation, real-time reasoning, and other CPU-intensive inference tasks. The announcement was made via an Amazon press release on April 30, 2026, highlighting that Graviton custom chips offer cost-performance advantages for large-scale AI deployments where GPU use can be reduced.

The partnership builds on existing ties between Meta and AWS: Meta continues to use AWS as a key cloud provider for its AI model serving alongside its own data centre infrastructure. Meta's VP of AI Infrastructure said the collaboration accelerates cost-efficient inference scaling, noting Graviton improves throughput-per-watt for "emerging single-core deterministic overlays segment." No specific financial terms nor milestone figures for launch milestones were given except the phrase 'tens of millions' reiterating core volume scale-up timeline undisclosed number-base commit.

The new deal marks another instance of cloud providers offering non-GPU silicon alternatives to meet increasing demand for fast token compute that AI app-server burdens currently demand — especially amid broader shortage pathways and geopricing arcs. DeepSeek’s sudden leap (April 24) reinforces server arm counting as seen in both closed & controlled weight race spirals calling varied chipset performances comparisons that industry reckons matter for competitive distribution cost before upcoming relevant policy rolls.