AI computing power construction shifts from “lack of electricity” to “lack of people”: the rise of modularity, Schneider, Weidi, and Eaton welcome new opportunities

The Zhitong Finance App learned that Bernstein recently released a research report. In the context of the rapid expansion of AI computing power construction, the main contradictions limiting the development of data centers are changing. The problem of connecting the power grid to the grid was gradually alleviated through self-provided power supplies after the meter, but the shortage of on-site construction and mechanical and electrical manpower became a new obstacle to development. The report points out that the modular construction model will break the bottleneck in computing power construction and reshape the market pattern and profit distribution logic of electrical equipment tracks.

Construction restrictions and iteration: shortage of manpower forces modular transformation

In the past few years, the biggest pain point of US data centers was that the grid connection cycle was too long, and the waiting time to be connected to the grid was extended to more than 5 years. The industry generally uses self-contained power supplies such as fuel cells and reciprocating engines (BTM) to bypass power grid restrictions. Currently, the proportion of off-meter power supplies in planned pipeline projects has reached 40%.

After the power supply bottleneck was eased, on-site civil construction and MEP mechanical and electrical labor became new hard constraints. 70% of data center projects are located in areas where mechanical and electrical workers are scarce. The 800VDC high-voltage architecture further enhances construction difficulty and increases on-site installation time by 50%. The maximum construction limit for data centers in the US in 2030 due to manpower is only 35 GW, which is far below the potential demand of 70 GW corresponding to GPU computing power.

Modularization migrates a large number of construction processes to factory prefabrication, and can only complete assembly on site. It can shorten the deployment cycle by 30-60%, reduce comprehensive labor costs by about 37%, shift employment demand from rural construction sites to industrial manufacturing centers, and break through the labor supply ceiling.

Changes in procurement models: integrated modules bring share dividends

Modularization is not only a change in engineering processes, but also a disruption in procurement logic. Under the traditional model, operators or EPC general contractors purchase parts separately; in the modular era, customers prefer to purchase Power+IT integrated prefabricated cabins. OEMs control product specifications, system integration and supply chains, and can complete the internal supply of 80 to 90% of components.

Research shows that 52% of customers want to purchase power supplies and IT modules at the same time for future AI projects, and modularization will significantly increase customers' willingness to purchase from a single source. According to estimates, the increase in modular penetration rate can bring about a net market share increase of nearly 3 percentage points for leading manufacturers with vertical integration capabilities. In terms of market space, the power module market size is 1.9 billion US dollars/GW, and IT modules reach 1.8 billion US dollars/GW; halving the deployment cycle can bring an incremental current value of 900 million US dollars per GW, and equipment vendors can obtain 25% of the revenue.

In the short term, the expansion of order scale and the acceleration of delivery turnover will improve the visibility of corporate orders, but climbing production capacity and increasing sales expenses will suppress profits; in the long term, depending on standardization and factory scale effects, EBITA profit per megawatt will continue to rise.

Circuit differentiation: IT module capabilities determine the competitive position of enterprises

Power module technology is relatively mature, and most electrical equipment manufacturers can supply it; IT computing modules are a scarce capability and have become the core yardstick for dividing the competitive echelons.

Schneider, VRT.US (VRT.US), and Eaton (ETN.US) form the first echelon, and also have complete product capabilities for power and IT cabins. The upper limit of Verdi products is high, and the OneCore architecture can support the construction of up to 1 GW park; Schneider has extensive experience in implementing projects, and prefabricated production capacity is expanding rapidly; Eaton relies on mergers and acquisitions and external cooperation to complete IT modules, and product maturity is slightly weak.

Legrand, ABB, and Siemens are in the second tier. Although they have power modules, they lack self-developed mature IT modules and can only output discrete components. Unless end-to-end integration capabilities are added, it will be passive in the wave of integrated module procurement. Non-hyperscale cloud vendors and managed cloud enterprises are the main driving force for modularity. From 2026 to 2030, this group will contribute 60% of new data centers and will also become the core customers of leading integrated OEMs.

Overall, modularization is an inevitable choice for AI computing power construction, but there are also potential risks: the turnkey model will amplify the OEM supply chain and project execution risks, and if the industry has excess capacity, the price war will erode the industry's profits.