Amazon agentic CPUs gain traction as Meta taps Graviton to run AI agents
Investing.com -- Amazon shares rose 1.2% Friday morning after Meta announced a partnership with Amazon Web Services to deploy tens of millions of AWS Graviton cores for agentic AI workloads. The company and some industry experts believe this is just the start.
Meta said it will become one of the largest Graviton customers globally, with the first deployment starting with tens of millions of Graviton cores. The partnership expands Meta's existing relationship with AWS as the social media company diversifies its compute infrastructure to meet demands of autonomous AI systems that reason, plan, and execute complex tasks.
Amazon published an update Friday on its blog explaining how central processing units are becoming critical for the next phase of AI development, which is focused on autonomous agents rather than large language models.
The company said agentic AI operates differently from standard LLMs. While language models function like calculators that process prompts through parallel math, AI agents act more like managers, breaking down goals into sequential steps and executing tasks autonomously.
Amazon explained that graphics processing units excel at training large language models through parallel processing, but AI agents require sustained computing power with fast communication between processing cores for tasks like logic, file management, network calls and code execution.
AWS Graviton processors are designed for these continuous, low-latency workloads. Amazon said the chips minimize the time different parts of the processor spend communicating with each other, which matters for AI systems constantly exchanging information during reasoning processes.
Vital Knowledge analyst Adam Crisafulli said that Amazon sentiment has "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 noted that the Graviton update underscores how critical the CPU market is becoming for the next phase of AI, one dominated by agents.
