Sinolink: Computing Power Reshaping the Power Boundary Power Supply, Liquid Cooling, and On-site Energy Resonance Upgrade.

Sinolink released a research report stating that the main focus of AIDC investment is undergoing a change, shifting from the shipment of single AI servers to gradually supporting the underlying infrastructure capabilities for the operation of large-scale AI clusters. The investment opportunity should upgrade from "server components volume increase" to "systematic restructuring of power infrastructure". As the scale of AI clusters expands, the intensity of infrastructure investment behind unit computing power continues to increase. Sinolink's main points are as follows: The bottleneck for the expansion of AI computing power is shifting from chip supply to power and thermal management supply. The bank believes that by 2026, the main focus of AIDC investment is changing: the market's focus is shifting from the shipment of single AI servers to gradually supporting the underlying infrastructure capabilities for the operation of large-scale AI clusters. With the increase in GPU/ASIC power consumption, the rise in power density per cabinet, and the continuous expansion of CSP capital expenditure, the constraints on data center construction are no longer just coming from chips and servers, but are further spreading to power supply, distribution, cooling, and on-site energy supply. Therefore, the investment opportunities in the AIDC industry chain should upgrade from "server component volume increase" to "systematic restructuring of power infrastructure". Corresponding to the industry links, power supplies solve "how to efficiently deliver power to chips", power equipment solves "how to access and expand capacity", liquid cooling solves "how to remove high-density heat", and SOFC solves "how to quickly provide on-site power when the grid cannot support it". AICapex continues to increase, and the intensity of infrastructure investment is increasing in sync. The capital expenditure of leading cloud providers worldwide is still in an upward cycle, driven by the demand for AI training, inference, and self-developed ASICs, accelerating the construction of intelligent computing centers. Unlike traditional cloud computing cycles, AIDC construction not only requires investment in servers and network equipment, but also synchronous configuration of higher-spec power distribution systems, thermal management systems, and backup/independent power systems. As the scale of AI clusters expands, the intensity of infrastructure investment behind unit computing power continues to increase: on one hand, high-power GPUs/ASICs are driving rapid increases in cabinet power density; on the other hand, large-scale cluster construction requires higher standards for power access, substations, distribution equipment, liquid cooling systems, and on-site power supply solutions. The spillover effect of AIDCCapex is becoming the core demand base for the prosperity of power supply, power equipment, liquid cooling, and SOFC. Risk warning AI data center power demand falls short of expectations; intensification of competition in SOFC technology; risks of changes in technology direction and product reliability verification falling short of expectations; geopolitical risks.