Meta bets on Amazon’s 3nm Graviton chips with 192 cores for next-gen AI workloads

Meta has signed a deal to deploy tens of millions of AWS Graviton processor cores as it expands the computing backbone needed for its next generation of artificial intelligence systems. The agreement deepens Meta’s long-running relationship with Amazon Web Services and highlights a growing shift in AI infrastructure. While graphics processors remain central to training large AI models, companies are now seeking more CPU power for inference, real-time reasoning, search, coding tools, and multi-step AI agents. Amazon said the rollout will begin with tens of millions of Graviton cores and can expand further as Meta’s AI demand grows. The chips are expected to power a range of Meta workloads tied to AI services used by billions of people across its platforms. The move also reflects rising demand for custom silicon that can reduce cost and energy use while delivering performance at hyperscale. CPU race begins AWS Graviton chips are Amazon’s in-house processors built on Arm architecture. They are designed to run cloud workloads faster, cheaper, and with lower power consumption than many traditional server chips. The latest Graviton5 chip uses a 3-nanometer manufacturing process and includes 192 cores. Amazon said it offers up to 25 percent better performance than the previous generation. The company also said Graviton5 carries a cache five times larger than the prior version, helping cut delays in communication between cores by up to 33 percent. That matters for AI systems that need to rapidly process data while coordinating many tasks at once. Graviton processors run on the AWS Nitro System, Amazon’s hardware and software stack designed to improve security, networking, and performance. The chips also support Elastic Fabric Adapter technology, which enables low-latency communication across large clusters of servers. AI demand shifts As AI products evolve, the industry is moving beyond model training alone. Newer agentic AI systems are expected to handle planning, coding,…