The rapid iteration of AI technology has escalated energy demands, while the explosive growth in computing power is driving the swift expansion of new loads such as data centers, thereby creating investment opportunities within the sector. In 2025, numerous individual stocks in the power sector have doubled in value. Indices more closely tied to computing power, specifically those reflecting ultra-high voltage and power grid equipment, surged over 70% for the year, with linked ETFs also posting impressive gains.
Multiple fund managers believe the importance of power supply is becoming increasingly prominent. Coupled with overseas grid modernization needs and energy transition demands, issues like aging power infrastructure and supply constraints in certain regions are highlighting sustained strong demand for power equipment. Chinese companies, leveraging their comprehensive industrial chains, diverse product portfolios, and production capacity advantages, are poised to benefit further from overseas power facility construction needs.
Amid the ongoing computing power boom, the energy sector—as both its driver and potential bottleneck—has delivered solid returns, albeit less dazzling than components like optical modules or PCBs.
For the full year 2025, the "Power Select Index" rose 10.49%. However, the "Hang Seng A-Share Grid Equipment" index, which is more closely linked to computing power and reflects ultra-high voltage and grid equipment, soared over 70%, with many constituent stocks achieving double-digit gains. Two linked products from Guotai Fund and GF Fund also performed remarkably well. Specifically, the Guotai Hang Seng A-Share Grid Equipment ETF saw its shares outstanding increase by over 200 million units for the year, while the GF Hang Seng A-Share Grid Equipment ETF's shares outstanding grew by more than 20 million units.
Several fund managers argue that the rapid development of AI is making power supply increasingly critical.
"The logic supporting power shortages remains quite strong currently; US AI data centers still face significant electricity supply constraints. This shortage rate could persist above 10 percentage points for many years to come, with risks potentially becoming more pronounced around 2027-2028," recently stated a fund manager from North China. The manager noted that on-site power generation appears to be the primary solution, relying mainly on gas turbines, energy storage, repurposing sites like those used for Bitcoin mining, and technologies such as SOFC (Solid Oxide Fuel Cells) to address short-term supply issues.
Furthermore, Chen Yabo, Investment Manager at Zhuque Fund's专户 investment department, indicated that regarding AI training, research by scholars estimates significant consumption. Using OpenAI's 1.3-billion-parameter GPT-3XL model as an example, assuming ChatGPT is trained 50 times annually, the yearly power consumption could reach 1.183 billion kWh. On the inference side, as of the end of June 2025, China's average daily Token consumption surged over 300-fold within a year and a half. The trend towards longer "thought processes" in inference increases Token usage, suggesting inference demands could substantially raise power consumption in the long term.
Ren Fei, Deputy Director of Equity Research and Fund Manager at ZhongOu Fund, stated that unlike previous economic cycles, two trend-driven forces are now creating additional electricity demand: the rapid expansion of AI data centers, and the shift of high-energy-consumption industries and supply chain restructuring driven by re-industrialization. Both are independent of traditional economic fluctuations and exhibit significant growth rates. From a medium-to-long-term perspective, the world may face a massive increase in required electricity consumption over the coming years.
Recently, management from an overseas company commented that despite sustained high market demand for AI chips, the current challenge is not an oversupply of computing power. Instead, data centers are hitting limits on power supply and physical space, causing large quantities of AI chips to remain idle in inventory, unable to be "powered on and operational."
It is reported that at Nvidia's investment summit in December 2025, some institutions pointed out that by 2027, GPU clusters supplied by Nvidia alone could consume 150-200 GW of electricity globally—equivalent to the total power consumption of 1.5 to 2 countries the size of France. This represents just the tip of the iceberg regarding computing power demand; the power gap has become the ultimate bottleneck for AI development. Moreover, if power issues remain unresolved, all advanced chips and powerful models would be merely "inoperable shells," stalling the AI revolution at the energy layer.
"We can also observe that overseas OEMs currently hold relatively high order backlogs, with rapid growth rates and clear expansion plans. Therefore, the integration of Chinese gas turbine components into the global supply chain appears quite certain. Many companies are already seeing strong order growth. This segment's high景气度 is likely to persist and is difficult to disprove, warranting continued attention," added the aforementioned North China fund manager.
Chen Yabo believes that against the backdrop of surging AI electricity consumption, three major areas within the power sector present布局 opportunities. First, on the grid side, factors like data center grid integration, aging grid upgrades, and increased electrification are driving robust demand景气度 for overseas power equipment, creating favorable conditions for Chinese companies expanding abroad. Demand for power transformers is increasing in some overseas regions, leading to supply tightness. Chinese enterprises, with their complete industrial chains, diverse product ranges, and capacity advantages, stand to benefit further from overseas power infrastructure needs. Additionally, integrated "源网荷储" (generation-grid-load-storage) models can improve energy efficiency, facilitate renewable energy integration, and enhance supply reliability, potentially becoming a solution for green power supply to data centers.
Second, the user side is witnessing technological upgrades. In recent years, data center rack power density has continuously increased, with potential for further growth, likely driving a shift in power supply architecture towards higher voltage levels and DC power. 800V DC could become a solution for AI computing center power architectures, and Solid-State Transformers (SST) also hold promise for application.
Finally, on the power generation side, addressing data center power needs and characteristics, generation sources like gas power, nuclear power, Solid Oxide Fuel Cells (SOFC), and integrated wind-PV-storage are expected to be key directions for strengthening data center power capacity.
"The transformation in power demand under the AI wave presents both challenges and opportunities," Chen Yabo said. The overseas opportunities and technological innovations on the grid side, the architectural evolution on the user side, and the diversified upgrades on the generation side collectively form the strategic blueprint for the power industry.
ZhongOu Fund's Ren Fei also identified three key directions worth watching: first, supply constraints, where high-energy-consumption industries like electrolytic aluminum might face direct restrictions; second, demand stimulation, as global power紧张 could boost demand for photovoltaics and energy storage, with the overseas energy storage market potentially growing around 2027-2028; and third, China's advantage, where its relatively ample power supply, coupled with cost and supply chain benefits in high-energy-consumption manufacturing, will become more pronounced.
Comments