Meta signed a multiyear deal for hundreds of thousands of Amazon Graviton5 processors to power agentic AI workloads, signalling that the AI infrastructure race has expanded beyond GPUs into CPUs.
Meta will access tens of millions of Graviton cores through AWS in a deal worth multiple billions, targeting real-time reasoning, code generation and multi-step task orchestration.
The shift from GPU-only AI training to CPU-heavy agentic inference changes the economics and architecture of AI at scale, which will flow through to ad platform performance and capability.
Australian businesses relying on Meta and AWS infrastructure will benefit from expanded compute capacity, but the concentration of AI infrastructure among three players (Google, Amazon, Meta) raises long-term dependency questions.
CTOs and marketing technologists building AI workflows on cloud infrastructure, plus agencies running AI-powered campaign optimisation at scale.
Assuming GPUs are the only AI hardware that matters will leave teams unprepared as agentic AI workloads shift cost structures toward inference-optimised compute.
Understand how your AI tools bill for inference versus training to anticipate cost shifts as agentic workloads grow.
Ask your cloud provider about inference-optimised compute options for production AI workflows.
