AI server demand surges on hyperscale and inference workloads, lifting CSP rack-scale buildout amid grid constraints.

Key Highlights

• Demand: AI server growth sustains; inference overtakes training.
• CSP Capex: Top-5 hyperscalers ramp rack-scale procurement.
• Rack Power: Liquid cooling and TDP rise lift per-rack wattage.
• ASIC Rise: Google and Amazon ASIC power tops NVIDIA solutions.
• Build Risk: Grid limits and politics challenge data center rollout.

Table of Contents

1. Global AI Servers to Maintain Robust CAGR between 2026 and 2029 under Actuation of Demand from Hyperscalers and AI Inference Applications
   • Projected Shipment Growth from AI Server Market (2023-2029)
2. Top 5 North American CSPs Combined at Over US$770 Billion of Capex for 2026, with Compute Power of AI Training and Inference Ramping Up to Respective YoY of 56% and 122% under Focus on GB/VR Racks
   • Computing Power of AI Training from Top 5 CSPs' Purchased GB/VR Racks to Approach a Growth of 56% in 2026
   • Computing Power of AI Inference from Top 5 CSPs' Purchased GB/VR Racks to Approach a Growth of 122% in 2026
3. Rack Power Expected to Rise from Increment of Liquid-Cooling and TDP; NVIDIA's Rack-Scale Evolves from 120kW to 225kW, Followed by High Power Integration for ASIC Platforms
   • Anticipated TDP and Power Consumption Specs for Chips Belonging to Major AI Server Platforms
4. Server Power Demand from Top Five CSPs Is Projected at 18GW in 2026; Google's and Amazon’s In-House ASICs Will Surpass NVIDIA’s Solutions, While Meta, Microsoft, and Oracle Remain Reliant on NVIDIA's Racks
5. Top 5 North American CSPs to Add 15GW of Data Center Capacity in 2026; Severe Grid Bottlenecks and Political Risks May Postpone Actual Deployment

<Total Pages: 8>