Zhitong Finance APP has learned that Zhongtai Securities Co.,Ltd. (600918) has released a research report stating that the global AI arms race is driving a surge in computing power capital expenditure, and power availability has become the primary bottleneck for AIDC construction. Although traditional gas turbines have lower unit costs, the delivery of main units and the construction cycle of power stations cannot match the time window for AI deployment. Self-provided power sources have shifted from optional to mandatory. SOFC, with its standardized modular stacking, can be rapidly replicated at hundreds of MW scale, bypassing grid connection and transmission bottlenecks, and is upgrading from an alternative technology to a mainstream option for AI power supply.

The main points of Zhongtai Securities Co.,Ltd. are as follows:

The global AI arms race is driving a surge in computing power capital expenditure, and power availability has become the primary bottleneck for AIDC construction. With the rapid development of global artificial intelligence, tech giants are engaging in an arms race around computing power, data center capital expenditure has risen sharply, and electricity demand has surged accordingly. However, aging US power grids, grid connection queues, and slow transmission expansion, combined with continuously lengthening delivery cycles for large power generation equipment, mean that AIDC waits for grid connection and electrical equipment arrival often take years. Although traditional gas turbines have lower unit costs, the delivery of main units and the construction cycle of power stations cannot match the time window for AI deployment. Electricity has become the first constraint on AIDC deployment. Self-provided power sources have shifted from optional to mandatory. Whoever can supply power faster can lock in computing power capacity first, and is thus better positioned in the AI era.

The data center launch window is narrowing, and SOFC solutions with immediate delivery capacity are the first to benefit. Gas turbines, with mature technology and large-scale production capacity, have become the preferred route for on-site self-provided power at AIDC, but product delivery cycles exceed 36 months, and orders through 2028 are locked in. Other on-site energy forms such as reciprocating internal combustion engines, aeroderivative gas turbines, and retired aviation engine conversions have limited capacity and cannot fill the demand gap. In contrast, SOFC, with standardized modular stacking, can be rapidly replicated at hundreds of MW scale, bypassing grid connection and transmission bottlenecks. Technologically, it can also match the future 800VDC AIDC power architecture. SOFC is upgrading from an alternative technology to a mainstream option for AI power supply.

Attention should be paid to the BE company supply chain, and suppliers with better positioning are expected to benefit significantly. The global SOFC player landscape is relatively concentrated. US-based Bloom Energy has cumulatively deployed approximately 1.8 GW, with a backlog of approximately US$20 billion, making it the most core integrator in the SOFC supply chain. Based on Bloom Energy successively winning GW-level North American AIDC power supply orders, including up to 1 GW from AEP and up to 2.8 GW from Oracle, it is expected to become a core catalyst for the SOFC supply chain in the future. At present, multiple companies have clearly established direct or indirect product supply for BE's SOFC systems. As BE's 2027 procurement scale gradually becomes clear in the future, more companies with positioning advantages are expected to enter BE's supply chain system and thereby benefit significantly from the rapid increase in SOFC demand.

Investment recommendations focus on Bloom Energy, Sanhuan Group, Xiandao Jidian, Chunhui Instruments, Zhenhua Co., Ltd., Jingquanhua, Yilian Technology, Johnson Electric Holdings, Weichai Power, Yishitong, Delta Electronics, Ceres Power, Fuelcell Energy, Doosan Group, Yunyi Electric, Kaizhong Precision, Corun, Zhongyuan Internal Combustion Engine Parts, Binglun Environment, and others.

Risk warnings: risks of AI computing power demand and capital expenditure falling short of expectations; technology route competition risk; BE order execution and revenue conversion risk; US policy and tax credit risk; geopolitical relations and tariff uncertainty risk; and risk of untimely updates to information used in the research report, among others.