Amazon's stock ticked up 1.2% on Friday morning following an announcement that Meta has entered a partnership with Amazon Web Services to run agentic AI workloads on tens of millions of AWS Graviton cores. Meta said the first phase of the deployment will involve tens of millions of Graviton cores and that the company will become one of the largest Graviton customers globally.
The agreement expands the companies' existing relationship as Meta diversifies its compute infrastructure to accommodate autonomous AI systems that are designed to reason, plan and execute complex tasks. Company statements and industry commentary framed the move as part of a shift in how AI compute will be provisioned for the next generation of applications.
In an update published Friday, Amazon argued that central processing units are taking on an increasingly important role as the AI industry moves from focusing primarily on large language models to systems built around autonomous agents. The company said agentic AI functions differently from standard LLMs: while language models operate like calculators that run prompts through large-scale parallel math, agents act more like managers, decomposing objectives into sequential steps and autonomously executing actions to meet those objectives.
Amazon emphasized that graphics processing units remain well suited to training large language models because of their parallel processing strengths. However, the company explained that agentic workloads demand sustained compute capacity with rapid communication between processing cores for activities such as logical reasoning, file management, network calls and code execution.
AWS Graviton processors are, according to Amazon, engineered for these continuous, low-latency tasks. The company said the chips reduce the time required for different portions of a processor to exchange information, a characteristic it described as important for AI systems that constantly trade data during multi-step reasoning processes.
Adam Crisafulli, an analyst at Vital Knowledge, said Amazon's market sentiment has "improved markedly over the last few months, thanks in large part to a growing appreciation for the firm’s custom silicon business." He added that the Graviton update highlights how critical the CPU market is becoming for an AI era centered on agentic architectures.
Meta's stated scale of deployment and AWS's positioning of Graviton for low-latency, continuous workloads suggest both firms view CPUs as complementary to GPUs in supporting the evolving computational profile of autonomous AI agents. Company commentary and analyst remarks included in the announcements indicate some market participants see the announcement as an early sign of broader shifts in AI infrastructure demand, though the companies described only the initial deployment plan.
Key points
- Amazon shares rose 1.2% Friday morning after Meta announced plans to deploy tens of millions of AWS Graviton cores for agentic AI workloads.
- Meta will be among the largest global customers of AWS Graviton, with the first deployment starting at tens of millions of cores to support autonomous AI systems.
- The announcement underscores a transition in compute needs from GPU-optimized training of large language models to CPU-focused, low-latency processing for agentic AI tasks.
Risks and uncertainties
- It remains uncertain whether CPU-centric infrastructure will become the dominant approach for agentic AI workloads; companies are currently balancing GPU and CPU roles.
- Meeting the sustained, low-latency inter-core communication needs described for agentic AI presents technical and scaling challenges for processor designers and cloud operators.
- Concentrating large-scale deployments with a single vendor could create operational or supply-side risks if infrastructure needs change.