China Great Wall: Iterative Upgrade of Distribution Architecture Focuses on High Growth Performance and New Technological Aspects

China Great Wall released a research report stating that the artificial intelligence revolution represented by large-scale model training and reasoning is driving a transformation in the form and infrastructure requirements of data centers. The Intelligent Computing Center (AIDC) centered around xPU is becoming the trend for the upgrade and development of general-purpose computing centers. According to their calculations, assuming a data center with a GPU power of 1000MW, the demand for distribution equipment is approximately 4.3 times the GPU power. In addition, the increase in power density raises the requirements for data center cooling, leading to a shift towards liquid cooling. The calculated growth rates for the liquid cooling market from 2026-2028 are 104.8%/65.4%/59.1%. During the industry's upward trend, attention is focused on high growth performance and new technological components. Key points from China Great Wall include: - Computing power demand drives data center construction. - The AI technology revolution, represented by large-scale model training and reasoning, is reshaping the form and infrastructure needs of data centers. From an industry perspective, domestic internet companies are entering a new round of capital expansion, while overseas cloud providers continue to increase their capital expenditures. In 2025 Q4, overseas CSP providers such as Microsoft, Google, Amazon, and Meta had a combined capital expenditure of 117.58 billion USD, up 62.5% year-on-year. - The CAGR of AI data center scale from 2025-2030 is estimated to be 29%. - According to McKinsey & Company data, by 2030, the total data center scale will be 220GW, with the AI data center scale reaching 156GW. The estimated CAGR from 2025-2030 is about 29%. Based on Alpha Matica data, the CapEx for a 100MW data center ranges from 3.4-5.5 billion USD, with IT equipment expenditure accounting for 73% and construction expenditure for 27%. - In a redundant configuration scenario, the demand for distribution equipment is about 4.3 times the GPU power. - High power drives the iteration of power and cooling architecture. - During the upward trend, attention is focused on high-growth performance and new technological components. Risk factors: 1) Data center construction progress falls short of expectations, 2) Progress in the application of technologies such as HVDC and SST is slower than expected, 3) Risks of competition in data center product pricing, 4) Risks of calculation data not aligning with actual circumstances.