Under the agreement, Meta is expected to deploy computing capacity equivalent to “tens of millions of cores” of Graviton chips. Each Graviton5 processor contains 192 cores, which can be allocated flexibly across different workloads depending on processing demands. While such chips are not primarily used to train advanced AI models, they play a critical role in running deployed systems efficiently once those models are operational.
The arrangement highlights an important distinction in modern AI computing infrastructure. Graphics processing units (GPUs)—produced by companies such as Nvidia—remain the dominant hardware for training large AI models due to their parallel processing strength. However, once models are trained and moved into production environments, CPUs are increasingly used to handle inference tasks, routing, and general-purpose computing more cost-effectively.
AWS, which began developing its proprietary CPU line in 2018, is now on its fifth iteration of the Graviton series. The chips are manufactured by Taiwan Semiconductor Manufacturing Co., and Amazon positions them as a lower-cost alternative to traditional third-party processors. According to AWS vice president and distinguished engineer Nafea Bshara, the company’s strategy is to pass those cost efficiencies directly on to customers, with the Meta agreement reflecting the scale of demand for AI-ready infrastructure.
The CPU market itself is experiencing renewed momentum as artificial intelligence workloads reshape demand patterns across the semiconductor industry. Industry players, including Intel, have noted rising CPU pricing pressures driven by increased demand. The trend reflects a broader diversification of compute requirements, where CPUs, GPUs, and custom accelerators all play distinct but interconnected roles.
Meta has already established itself as a major buyer of advanced computing hardware, with prior large-scale commitments to Nvidia and Advanced Micro Devices, as well as collaboration with Arm Holdings on its newer CPU designs. The company views this diversification as strategically important as it scales its AI systems across products and services.
Santosh Janardhan, Meta’s head of infrastructure, emphasized this approach, noting that broadening its compute sources is essential to supporting the company’s expanding AI ambitions.
The AWS partnership underscores a wider industry shift: rather than relying on a single type of processor or supplier, major technology firms are increasingly building heterogeneous computing stacks designed to balance performance, cost, and scalability in the rapidly evolving AI landscape.