Hyperscalers Are Building Their Own CPUs

Hyperscalers like AWS, Google, and Microsoft are increasingly designing their own Arm-based CPUs, driven by the need for tailored data center solutions. Arm's Mohamed Awad emphasizes that this move goes beyond cost reduction; these companies aim to create fully integrated systems optimized for specific workloads.

Instead of building data centers around existing CPUs, hyperscalers develop processors from the ground up, focusing on their unique requirements. This allows them to prioritize performance, efficiency, or both, and eliminate unnecessary components. This approach extends beyond CPUs to encompass AI accelerators and networking chips, enabling holistic system-level optimization.

Hyperscalers anticipate lower total ownership costs through reduced margins and improved power efficiency. Awad highlights Arm's ecosystem, which offers performance, power efficiency, a mature software and hardware base, and design flexibility. While companies like Ampere provide Arm-based CPUs, hyperscalers often require custom solutions built specifically for their infrastructure.

Arm's initiatives, such as Compute Subsystems and Arm Total Design, aim to democratize custom silicon development for firms with fewer resources. These programs lower design expenses and ease the support burden on Arm.

Awad projects Arm-based CPUs to capture a significant portion of the data center market, displacing x86 processors. He stresses the importance of a robust software ecosystem, a key advantage for Arm. The surge in AI workloads and the adoption of Arm-based CPUs in conjunction with GPUs further fuels Arm's growth in data centers.