The host CPU layer of custom AI server infrastructure is undergoing a quiet but structural transformation, with x86 architectures gradually giving way to proprietary Arm-based designs, according to the latest Data Center AI Server Compute ASICs Shipment Forecast and Tracker from Counterpoint Research's HPC Service.

While public discourse on AI infrastructure has focused almost entirely on accelerator shipments, our bottom-up tracking of host CPU attach rates for Google, AWS, Microsoft, Meta and other major hyperscalers reveals a separate, underappreciated demand transition unfolding in parallel.

Over the last three years, x86-based CPUs have dominated the CPU attach rates for AI infrastructure accelerator deployments. Leading hyperscalers have relied on Intel and AMD processors as host CPUs for their ASIC server deployments through the current generation, reflecting software compatibility requirements and existing data center infrastructure. However, in this growing market, and due to the need for heterogeneous architecture driven by semi-custom AI accelerators or XPUs, we are seeing even a bigger push in the adoption of Arm-based server CPUs based on Arm Neoverse cores.

This move is fundamentally cost and efficiency-driven for hyperscalers. By designing their own host processors alongside their AI accelerators, hyperscalers aim to diversify, reduce dependency on merchant silicon vendors, recapture margin at scale, and bring down the token costs. As a result, proprietary Arm CPUs were initially directed toward general-purpose cloud workloads before being designed for AI server programs. Arm-based CPUs have demonstrated up to twice the performance-per-watt of comparable x86 rack configurations, a critical advantage as hyperscalers look to maximize compute density within fixed power envelopes.

Highlighting this trend, Research Associate David Wu said, “While x86 architectures currently maintain a significant presence in AI server infrastructure, our generation-by-generation analysis suggests this established stronghold is swiftly transitioning toward proprietary Arm-based designs. We expect a more pronounced acceleration in this shift in the second half of 2026. Understanding which specific hyperscaler and which ASIC generation is transitioning from x86 to Arm is where the actionable insight lives.”

Proliferation of Arm-based CPUs across AI hyperscalers

For Google, the ramp-up of its Axion Arm-based CPU in its next-generation TPU infrastructure stands out as the most significant single event in this transition, a signal that Arm-based CPUs are now ready for large-scale AI infrastructure deployment.

AWS has similarly been shifting its approach across successive Trainium generations, with Arm-based Graviton processors playing a growing role in higher-density configurations, even as x86 remains in certain deployments for backward compatibility.

Beyond Google and AWS, the Arm shift is gaining traction across the hyperscaler landscape. Microsoft has paired its Azure Cobalt ARM CPU with the Maia AI accelerator family from the start.

Enter Arm AGI CPU

Meta’s recent confirmation of Arm as a strategic CPU partner for its next-generation MTIA infrastructure, with Meta named as the launch customer for Arm’s first-ever AGI CPU, further shows that the move away from merchant x86 is a deliberate, industry-wide direction rather than an isolated design decision.

Research VP Neil Shah said, “The transition from x86 to Arm in AI servers is not a single switch. It has played out generation by generation, configuration by configuration. Hyperscalers are making deliberate choices based on their specific deployment needs, writing compatible and interoperable software, and the economics are very encouraging. The transition is expected to accelerate meaningfully in the second half of 2026, driven by the broad deployment of in-house Arm CPUs alongside next-generation ASIC platforms across major hyperscalers.”

Shah added, “Our analysis projects Arm-based CPUs will account for at least 90% of host CPU deployments in custom AI ASIC servers by 2029, up from around 25% in 2025, a structural shift driven by the accelerating rollout of in-house Arm CPU programs across major hyperscalers.”

This shift also has meaningful implications for the broader semiconductor supply chain. As hyperscalers move toward in-house Arm CPUs manufactured on advanced process nodes, the AI server build-out will increasingly drive demand across both the AI server compute ASICs and CPU layers of the TSMC supply chain simultaneously.

For detailed research, refer to the following reports available for subscribers:

• Global AI Server Compute ASIC CPU Attach Shipments by Chip Model, Hyperscaler, Design Partner and Foundry Forecast, March 2026
• HBM Demand for AI Server Compute ASICs to Grow 35x by 2028
• AI Server Compute ASIC Shipments to Triple by 2027 as Custom Silicon Enters Hyper-Growth Phase
• Foundry Monthly Intelligence Report, February 2026

About Counterpoint Research

Counterpoint Research is a global market research firm specializing in products across the technology ecosystem. We advise a diverse range of clients – from smartphone OEMs to chipmakers and channel players to Big Tech – through our offices located in the world's major innovation hubs, manufacturing clusters and commercial centers. Our analyst team, led by seasoned experts, engages with stakeholders across the enterprise – from the C-suite to professionals in strategy, analyst relations (AR), market intelligence (MI), business intelligence (BI), product and marketing – to deliver services spanning market data, industry thought leadership and consulting. Our core areas of coverage include AI, Automotive, Consumer Electronics, Displays, eSIM, IoT, Location Platforms, Macroeconomics, Manufacturing, Networks and Infrastructure, Semiconductors, Smartphones and Wearables. Visit our Insights page to explore our publicly available market data, insights and thought leadership, and to understand our focus, meet our analysts and start a conversation.