The useful question is what changes for users, developers or buyers, and whether the announcement stays industry context or becomes something people can actually use.

Microsoft and Amazon’s latest earnings calls highlighted the growing role of Arm-based CPUs in hyperscale cloud infrastructure, marking a shift from experimental use to core operational backbone. Microsoft’s Cobalt 200 CPUs, now deployed in over 25 data centers globally, are designed for agentic AI workloads, with performance improvements of up to 50% over its predecessor. The company’s Azure revenue grew 43% for the quarter, partly attributed to efficiency gains from its CPU and GPU fleets, demonstrating how optimized silicon can directly impact financial performance.

Amazon’s chips business, including Graviton, Trainium, and Nitro, surpassed a $25 billion annual revenue run rate, with Graviton5 adoption accelerating nearly twice as fast as Graviton4. Graviton5-powered instances deliver up to 25% better compute performance and 35% faster machine learning inference compared to Graviton4. The rapid uptake underscores the demand for CPUs tailored to agentic AI tasks such as real-time reasoning and multi-step orchestration, which require efficient handling of routing, data preparation, and tool management.

The broader industry trend reflects a move beyond GPUs as the primary AI infrastructure, with Google Cloud’s Axion and NVIDIA’s Vera CPUs serving as orchestration layers for AI systems. IDC projects global AI infrastructure spending to reach $497 billion in 2026, with demand expanding beyond GPU systems to include CPU-only inference clusters and data pipelines. This shift is driven by the need for efficiency in power and rack utilization, particularly as electricity use from AI-focused data centers surged 50% in 2025, according to the IEA.

Microsoft and AWS are leading this architectural transition, but the trend extends across major hyperscalers. Google’s Axion CPUs underpin its cloud infrastructure and AI head nodes, while NVIDIA’s Vera CPU coordinates memory, networking, and accelerators in rack-scale deployments. The common thread is Arm-based CPUs serving as the foundational compute layer for modern AI data centers, enabling scalable, efficient, and flexible infrastructure to meet the demands of agentic AI workloads.