Power supply capacity is increasingly becoming an invisible bottleneck for data center construction. An industry deep-dive report from Shenwan Hongyuan Securities, published on April 16th, highlights that the accelerated construction of AI data centers (AIDC) in North America is pushing power supply constraints to a critical point. In the short term, gas turbines and reciprocating internal combustion engines, leveraging their fast start-stop capabilities and flexible deployment advantages, have become the core power supply solution for AIDCs currently. Aeroderivative gas turbines can start up in just 5-10 minutes, while heavy-duty gas turbines in combined cycle configurations can achieve efficiencies exceeding 60%. For the medium to long term, Small Modular Reactors (SMRs), controlled nuclear fusion, and Solid Oxide Fuel Cells (SOFCs) are viewed as the ideal long-term power solutions for AIDCs.

Gas Turbines: The Current "Firefighting" Solution Confronted with the immediate, rigid power demand of AIDCs, Shenwan Hongyuan identifies gas turbines and reciprocating internal combustion engines as the most feasible core solutions at this stage. This is due to a high degree of compatibility in technical characteristics. Solar and wind power are intermittent energy sources, struggling to meet round-the-clock stable power needs. Nuclear power plants and large-scale hydroelectric projects typically have construction cycles of 4-5 years or more, which cannot align with the rapid pace of AIDC development. In contrast, gas turbines offer rapid response, flexible regulation, and short construction cycles, making them the "mainstream solution for supporting AIDC continuous loads and emergency backup power at this stage."

Within the gas turbine category, light and heavy models serve different purposes. Aeroderivative gas turbines, derived from modified aircraft engines, require only 5-10 minutes to start, typically generate 5-60MW, have a small footprint, allow for mobile deployment, and are suited for distributed power generation and emergency peak shaving. Representative models include GE's LM2500+ and LM6000 series, Siemens Energy's SGT-A35, and Mitsubishi Heavy Industries' FT4000 based on the Pratt & Whitney PW4000 engine. Heavy-duty gas turbines offer higher power, typically 50-500MW, with start-up times of 30-60 minutes, making them suitable for base-load power requirements of large AIDCs. When equipped with combined cycle systems, their power generation efficiency can surpass 60%, representing a top-tier technological pathway for fossil fuel power generation efficiency. Classified by turbine inlet temperature, heavy-duty gas turbines range from Class D (power <100MW) to Class J (>1600°C, power 300-400MW).

Reciprocating internal combustion engines complement gas turbines. They start and stop even faster (reaching full load in 5-10 minutes), have unit capacities of 3-20MW, construction cycles of 15-24 months, and controllable costs, primarily covering power supply for small-to-medium AIDCs and extreme backup scenarios. For instance, Meta's actual deployment at its Ohio data center power plant utilizes a combination of gas turbines like the Solar Titan 250 and Siemens Energy SGT 400, alongside CAT 3520 gas internal combustion engines.

Medium to Long Term: Three Technological Pathways Unlock Future Potential The Shenwan Hongyuan report states that while gas turbines are a "firefighting" solution, the long-term power needs of AIDCs point towards three cutting-edge technological pathways: SMRs, controlled nuclear fusion, and SOFCs.

Small Modular Reactors (SMRs) The International Atomic Energy Agency (IAEA) defines SMRs as nuclear reactors with a single-unit rated output between 10-300 MWe. Compared to traditional large nuclear power plants, SMRs employ modular design, allowing for "building block" style construction, significantly reducing construction risks and costs. The latest IAEA forecast predicts global nuclear power generation capacity will increase to 2.5 times current levels by 2050, with SMRs accounting for 25% of this. Tech giants are actively involved. According to the report: Amazon is collaborating with X-energy, planning to deploy over 5 GW of SMRs in the US by 2039; Google has signed a power purchase agreement with Kairos Power, targeting the first SMR by 2030; Microsoft has a 20-year agreement with Constellation to restart the Three Mile Island nuclear station; Meta is funding two TerraPower reactor projects totaling 690 MW and has an agreement with Oklo to develop a 1.2 GW nuclear technology park in Ohio; Oracle is designing a data center expected to be powered by three small reactors providing over 1 GW of electricity. The report suggests SMR power plants could achieve cost-competitiveness with natural gas plants after 2030.

Controlled Nuclear Fusion Nuclear fusion is regarded as the "ultimate energy source," currently pursued mainly through magnetic confinement and inertial confinement pathways. Shenwan Hongyuan believes magnetic confinement fusion "holds greater potential for engineering feasibility and medium-to-long-term industrialization," with representative projects including ITER (global) and EAST (China). Tech giants are also actively investing. Microsoft is partnering with Helion on pulsed magnetic field technology, signing the world's first fusion power purchase agreement, targeting operation by 2028; Google is involved with CFS and TAE, focusing on high-temperature superconducting tokamaks and hydrogen-boron fusion respectively; NVIDIA has invested in CFS, concentrating on AI applications for fusion simulation.

Solid Oxide Fuel Cells (SOFCs) SOFCs operate at 600-1000°C, can directly utilize various fuels like natural gas, coal gas, and biogas, achieve electrical efficiencies of 45%-60%, and can surpass 90% overall energy efficiency when combined with cogeneration. US-based Bloom Energy has achieved commercial scale, deploying multiple distributed generation systems for companies like Apple, Walmart, and Google, with electrical efficiency around 60% and cogeneration efficiency nearing 90%. The common characteristics of these three pathways are: environmental friendliness, high energy density, and long-term stable power supply, which align well with the long-term needs of AIDCs. Shenwan Hongyuan anticipates a critical juncture for commercial deployment around 2030.

Structural Mismatch in Power Supply The report concludes that the current challenge for data center construction is not a general power shortage, but a structural mismatch. According to EIA data, the US generated 43.1 trillion kWh and consumed 39.8 trillion kWh in 2024, indicating no overall deficit. However, 70% of US transmission lines and transformers are over 25 years old, and 60% of circuit breakers are over 30 years old. Simultaneously, AIDC power demand is highly concentrated in tech hubs like California's Silicon Valley, Texas, and Virginia, creating a dual mismatch in both time and space between grid aging and surging demand. The report makes a notable assertion: "Power supply capability significantly influences the scale and pace of AIDC construction. The importance of power solutions may even surpass that of computing hardware, becoming a fundamental constraint in AI model training and commercialization that cannot be overlooked, rather than a mere auxiliary component." The US Department of Energy's Resource Adequacy Report also warns that, affected by both power plant retirements and growing electricity demand, power outages in the US could increase by 100% by 2030.

Shenwan Hongyuan proposes two main investment themes. The first theme focuses on mature power generation equipment for the current stage, including manufacturing, core components, system integration, and maintenance services for gas turbines and reciprocating internal combustion engines. The report mentions companies with final assembly capabilities and key players in the supply chain. The second theme involves forward-looking investment in advanced power generation technologies, focusing on the R&D and industrialization of SMR design/manufacturing, core equipment for nuclear fusion, and SOFC fuel cells. The report also highlights three categories of risk: slower-than-expected AIDC construction progress, technological substitution risks, and market expansion falling short of expectations, including potential impacts from geopolitical factors on overseas markets.