AIDC power supply, relying on gas turbines for emergency in the short term, and on these three approaches in the long term
```
Power supply capacity is becoming the invisible ceiling for data center construction.
On April 16, Shenwan Hongyuan Securities released an industry in-depth report "The Growth of Computing Power Demand Intensifies the Power Gap, Opening Growth Space for the Gas Turbine Market", which points out that the acceleration of North American AI Data Center (AIDC) construction is pushing the contradiction of power supply to a critical point.
In the short term, gas turbines and reciprocating internal combustion engines have become the core solutions for AIDC current power supply due to their fast start/stop and flexible deployment advantages; aero-derived gas turbines can start within only 5-10 minutes, and heavy-duty gas turbines in combined-cycle can achieve more than 60% efficiency. In the medium and long term, SMR (small modular nuclear reactors), controllable nuclear fusion, and SOFC (solid oxide fuel cell) are seen as ideal long-term power supply directions for AIDC.
Gas Turbines: The Most Effective "Firefighting" Solution at Present
Faced with AIDC's immediate and rigid power supply demand, Shenwan Hongyuan believes that gas turbines and reciprocating internal combustion engines are the most feasible core solutions at this stage.
This is due to a high degree of technical matching. Photovoltaics and wind power are intermittent sources, and can hardly meet 24/7 stable power supply demand. Construction periods for nuclear and large hydropower plants are generally longer than 4-5 years, which do not match AIDC's construction pace. Gas turbines, however, respond quickly, are flexible to regulate, and have short construction cycles, making them "the mainstream solution at the current stage to support AIDC's continuous load and emergency power backup."
Within gas turbines, light and heavy models each have their own focus:
Aero-derived gas turbines are modified from aviation engines, require only 5-10 minutes to start up, provide typical output of 5-60MW, occupy little space, and can be deployed flexibly—suitable for distributed power, emergency peak shaving, and other scenarios. Representative models include GE's LM2500+ series, LM6000 series, Siemens Energy's SGT-A35, Mitsubishi Heavy Industries' FT4000 (based on Pratt & Whitney PW4000 engine), etc.
Heavy-duty gas turbines have higher power, typical output of 50-500MW, startup time of 30-60 minutes, suitable for base load power supply of large AIDCs. With combined cycle systems, efficiency can reach more than 60%, representing the highest standard for fossil fuel efficiency. Based on turbine inlet temperature, heavy-duty gas turbines range from class D (<100MW) to class J (>1600℃, 300-400MW), with stepwise improvements.
Reciprocating internal combustion engines complement gas turbines. They have even faster start/stop (5-10 minutes to full load), single-unit capacity of 3-20MW, construction cycle of 15-24 months, and controllable costs. They mainly cover power for small to medium AIDC and extreme backup scenarios. For instance, Meta's data center power plant in Ohio deployed both Solar Titan250, Siemens Energy SGT 400 gas turbines, and CAT 3520 gas internal combustion engines.
Medium and Long-Term: Three Technical Routes Open Long-term Space
According to Shenwan Hongyuan's report, gas turbines are a "firefighting" solution, but AIDC's long-term power supply demand points to three cutting-edge technical routes: SMR, controllable nuclear fusion, and SOFC.
SMR (Small Modular Nuclear Reactor)
The International Atomic Energy Agency (IAEA) defines SMR as nuclear reactors with a single-module rated output of 10-300 MWe. Compared with traditional large nuclear power plants, SMRs adopt a modular design for "lego-like" construction, greatly reducing construction risks and costs. According to the IAEA's latest forecast, by 2050, nuclear power generation capacity will reach 2.5 times the current level, with SMR accounting for 25%.
Tech giants are intensively entering this field. The report notes: Amazon cooperates with X-energy, planning to deploy over 5GW of SMR in the US by 2039; Google signed a power purchase agreement with Kairos Power, aiming to complete its first SMR by 2030; Microsoft and Constellation signed a 20-year PPA and restarted Three Mile Island Nuclear Plant; Meta funded two TerraPower reactor projects (690MW), and signed an agreement with Oklo to develop a 1.2GW nuclear park in Ohio; Oracle is designing a data center expected to be powered by three SMRs providing over 1GW of electricity.
The report believes that the economics of SMR plants are expected to rival gas plants after 2030.
Controllable Nuclear Fusion
Nuclear fusion is seen as the "ultimate energy". Currently, its main technical paths are magnetic confinement and inertial confinement. Shenwan Hongyuan believes that magnetic confinement fusion "has more engineering feasibility and long-term industrialization potential", with representative facilities including ITER (global) and EAST (China).
Tech giants are also making proactive layouts. Microsoft is working with Helion to promote pulsed field reversed configuration, signing the world’s first fusion electricity PPA, expecting completion by 2028; Google is collaborating with CFS and TAE in high-temperature superconducting tokamak and proton-boron fusion, respectively; Nvidia invests in CFS, focusing on AI+fusion simulation.
SOFC (Solid Oxide Fuel Cell)
SOFCs operate at 600-1000°C, can directly use various fuels such as natural gas, coal gas, biogas, etc., and have power generation efficiency of 45%-60%. When combined with cogeneration, comprehensive energy efficiency can exceed 90%. US-based Bloom Energy has achieved large-scale commercial application, deploying multiple distributed generation systems for Apple, Walmart, Google, etc., with power efficiency near 60% and CHP efficiency close to 90%.
All three routes have common characteristics: environmental protection, high energy density, long-cycle stable power supply, highly matching AIDC's long-term needs. Shenwan Hongyuan believes that 2030 is expected to be a key node for commercial implementation.
Structural Power Supply Mismatch
The report judges that the current data center construction is not facing an overall power shortage, but a structural mismatch. According to EIA (US Energy Information Administration) data, in 2024 the total US power generation is 4.31 trillion kWh, total electricity consumption is 3.98 trillion kWh—not overall scarce. But 70% of US transmission lines and transformers have been operating for over 25 years, and 60% of circuit breakers for over 30 years.
Meanwhile, AIDC's power demand is highly concentrated in tech hubs such as Silicon Valley in California, Texas, and Virginia, producing dual mismatches in time and location between grid aging and demand explosion.
Shenwan Hongyuan points out in its report: "Power supply capacity greatly influences the scale and rollout speed of AIDC construction. The importance of power supply solutions may even surpass computing hardware, becoming an underlying constraint that cannot be ignored in the training and commercialization of large AI models, rather than a simple auxiliary supporting factor."
The US Department of Energy’s (DOE) "Resource Adequacy Report" also clearly warns that, due to plant retirements and increasing power loads, the number of blackouts in the US may increase by 100% by 2030.
Shenwan Hongyuan suggests two main investment directions in its report.
The first direction focuses on mature current-generation power equipment, including manufacturing of complete gas turbines, reciprocating engines, their core components, system integration, and O&M services. The report mentions Aero Engine Corporation, China Power, Dongfang Electric, and, within industry core supply chains, Yingliu, Wanze, Aerospace Technology, Longda, and others.
The second direction is forward-looking, focusing on advanced power generation technologies, including SMR nuclear reactor design and manufacturing, nuclear fusion core equipment, SOFC fuel cell R&D and industrialization. Nuclear fusion-related companies include HFM, Union Optech, Western Superconducting, and so on.
The report also warns of three types of risk: AIDC construction progress falls short of expectations, technology route substitution risks, and market expansion below expectations (including impact of geopolitical factors on overseas markets).
Risk Warning & DisclaimerThe market involves risk and investments should be made with caution. This document does not constitute personal investment advice and does not take into account the individual investment objectives, financial situation or needs of any particular user. Users should consider whether any opinions, views, or conclusions in this document are suitable for their individual circumstances. Investment decisions are at your own risk. ```