Analysis of the Impact of the 2030 U.S. Power Crisis on AI Data Center Industry Chain Investments

I will now conduct a systematic analysis based on the collected data.

---

Goldman Sachs issued a stark warning in its 2025 annual research report: the U.S. power grid is facing an unprecedented power crisis, which is expected to enter a state of full strain by 2030 [1]. According to Samantha Dart, co-head of Global Commodities Research at Goldman Sachs, the explosive growth in power demand from U.S. data centers is triggering a severe power crisis; if power constraints continue to intensify, China is expected to overtake in the AI race [2].

From a core data perspective, the situation is particularly grim. A July 2025 Goldman Sachs report shows that the U.S. data center power gap has remained at around 10GW since 2023 (a 1GW computing cluster can accommodate 200,000 NVIDIA GB200 chips), while the U.S. only adds 12GW of new power generation capacity annually [3]. October 2025 statistics from Barclays indicate that the computing power investment plans of tech companies and data center service providers including Amazon, Microsoft, Google, Oracle, and OpenAI amount to at least 45GW [3]. More worryingly, according to Financial Times analysis, AI data centers are expected to add approximately 44GW of power demand by 2028, but only 25GW can be met due to grid infrastructure bottlenecks, resulting in a power gap of about 19GW (accounting for 40%) [4].

According to industry standards, power grids need to maintain a reserve margin of at least 15% (the difference between peak load and available generation capacity), but currently 8 out of 13 regional power markets in the U.S. have reserve capacities at or below critical levels [5]. Data from the U.S. Energy Information Administration shows that data centers accounted for 4% of U.S. social power consumption in 2024, and this is expected to increase to 9% by 2030 [3]. The IEA’s baseline scenario shows that global data center power consumption will double to 945 TWh by 2030, and may even exceed 1200 TWh in the aggressive scenario [1].

Despite tight power supply, capital expenditure of overseas cloud providers continues to grow strongly. In Q3 2025, the total capital expenditure of major overseas players reached USD 99.617 billion, representing an 80.39% year-on-year increase and a 9.54% quarter-on-quarter increase [6]. Specifically:

Company	2025 Q3 Capital Expenditure (USD 100 million)	Full-Year Guidance (USD 100 million)	2026 Outlook
Amazon	342.28	~1250	Continue to increase
Google	239.53	910-930	Will increase significantly
Meta	188.29	700-720	Growth rate significantly higher than 2025
Microsoft	167.45	Growth rate higher than FY2025	Double data center scale within two years
Oracle	58.62	~350	Substantial upward revision

Microsoft CEO Satya Nadella stated frankly in November 2025: “If we cannot build data centers in places with power quickly, there may be a large number of chips sitting in warehouses unable to be powered on. This is the problem I face today.” [3] This statement profoundly reveals that power bottlenecks have become the core factor restricting the implementation of computing power investments.

SemiAnalysis revealed the essence of the problem in its in-depth report: the construction cycle of AI data centers has been compressed to 12-24 months, while the typical cycle for grid expansion, transmission line construction, and grid connection approval remains 3-5 years [7]. Take Texas ERCOT as an example: between 2024-2025, the new load applications submitted by data centers reached dozens of GW, but only about 1GW of new load was actually approved and successfully connected to the grid during the same period [7]. This ‘time mismatch’ makes ‘waiting for the grid’ an unbearable risk in itself.

From an investment return perspective, a 1GW-scale AI data center can generate potential annual revenue of up to USD 10 billion [7]. Even for a medium-scale cluster, the commercial value brought by launching a few months ahead is sufficient to cover higher power costs. In this context, power has evolved from a mere operating cost to the primary constraint determining whether AI projects can go online on schedule.

Facing grid bottlenecks, the ‘Bring Your Own Generation’ (BYOG) model is rapidly gaining popularity. The goal of this model is not to permanently decouple from the grid, but to quickly put capacity into operation in an off-grid manner in the early stage, then gradually connect to the grid later, with the on-site power plant converted to backup and redundancy [7].

xAI in Memphis is a model example. Elon Musk built a 100,000-GPU-scale cluster in less than four months, with the core of its energy strategy being completely bypassing the grid, using truck-mounted gas turbines and engines to generate power on-site, having deployed over 500 MW of turbines [7]. In October 2025, OpenAI and Oracle ordered the largest on-site natural gas power plant in history in Texas, with a scale of 2.3 GW [7]. The on-site natural gas power generation market is entering an era of three-digit annual growth.

SMRs are regarded as a key solution to the AI power crisis. Unlike traditional large-scale nuclear power projects, SMRs attempt to transform nuclear energy from a ‘large-scale project’ to an ‘industrial product’ by shortening the construction cycle to 3-5 years and lowering the initial capital threshold [8].

Tech giants have been actively laying out in the nuclear energy sector:

• Microsoft
  : Signed a 20-year power purchase agreement (PPA) aimed at restarting the Three Mile Island Unit 1 [8]
• Google
  : Ordered 6-7 reactors from Kairos Power to obtain 500 MW of clean energy [8]
• Amazon
  : Took a stake in X-energy and signed a memorandum of understanding on SMR site selection with Dominion [8]
• Oracle
  : Announced a large-scale data park powered by three modular reactors [8]

The International Energy Agency (IEA) predicts that annual investment in SMRs will reach USD 25 billion by 2030 [8]. However, large-scale deployment of SMRs still faces severe challenges, including fuel supply chain issues—Russia controls 40% of the global uranium enrichment capacity, and Kazakhstan supplies 43% of global uranium [8].

U.S. regulated utility companies have announced their five-year investment plans, expecting to invest USD 1.1 trillion to meet the growth in power demand driven by AI development. This cost will ultimately be passed on to power users, and the scale is twice the average annual expenditure level over the past decade [9]. In addition:

• Planned new natural gas pipeline projects: USD 50 billion
• Private transmission projects: tens of billions of USD
• Capacity expansion by equipment manufacturers: Eaton, GE Energy, Schneider Electric, Vertiv, etc. have all announced plans to invest billions of USD to expand factories [9]

Data centers have become the core incremental application scenario for the power equipment industry. Global new installed capacity of data centers reached 14GW in 2024, and the power density per cabinet has increased significantly, which puts higher requirements on the stability and energy efficiency of power supply [6]. McKinsey estimates that global investment in artificial intelligence and cloud computing will reach USD 6.7 trillion between 2025 and 2030, broken down as follows:

• Information technology equipment (chips, servers, and storage devices): USD 4.4 trillion
• Construction labor, materials, and land costs for data centers: USD 1 trillion
• Electrical machinery and equipment, and fiber optic networks: nearly USD 1 trillion [9]

From an industry chain perspective, the following sub-sectors will benefit significantly:

• SST (Static Switch Transfer)
  : AIDC construction drives growth in SST demand, recommend Sungrow Power Supply (300274.SZ)
• UPS and Power Supply & Distribution Systems
  : Kehua Data (002335.SZ), Kstar (002518.SZ), Clou Electronics (002121.SZ)
• Industrial Power Supplies
  : Megmeet (002851.SZ), Jinpan Technology (688676.SH), Sifang Electric (601126.SH) [6]

The natural gas industry is viewing the AI race as a once-in-a-generation opportunity. EQT Corporation’s natural gas power plant supporting the proposed large off-grid AI data center in Homer City, Pennsylvania, can accommodate up to 4.4 GW of artificial intelligence computing power [9]. Williams Companies has planned a USD 5.1 billion portfolio of power generation and transportation projects to directly power data centers [9]. Clean energy producers are also seizing the opportunity to build solar and energy storage power stations equipped with natural gas backup power sources.

Goldman Sachs’ comparative analysis shows that China currently has sufficient reserve capacity, and this figure is expected to grow further. By 2030, Goldman Sachs predicts that China’s effective reserve power capacity will be more than three times the global expected data center power demand (approximately 400 GW vs. approximately 120 GW) [5]. The most notable feature of China’s continuous growth in power capacity over the past five years is the rapid expansion of all energy supply sources—China’s solar installed capacity is currently about six times that of the U.S. [5].

The Chinese government also provides power subsidies for data centers, offering up to 50% power subsidies for data centers using domestic artificial intelligence chips [5]. This policy support further strengthens China’s competitive advantage in the field of AI infrastructure.

For investors, the following core investment logics should be noted:

: Power equipment and power supply & distribution system suppliers will directly benefit from the growth in data center construction demand, focusing on leading enterprises with technical barriers and scale advantages.

: The SMR industry chain has huge development potential, including nuclear fuel supply chains, equipment manufacturers, and operation service providers, but it is necessary to closely monitor technology verification progress and regulatory policy changes.

: Smart grid transformation, energy storage systems, and distributed energy will become key investment directions. Integrated energy service providers with both ‘power + computing power’ capabilities are expected to obtain excess returns.

: Cloud providers’ capital expenditure falls short of expectations/rhythm fluctuations; AI application implementation falls short of expectations; raw material and trade policy risks; SMR technology verification failure risks; grid approval progress is slower than expected [6].

---

[1] Wang Wei. 2026-2035 New Era of World Economy: Nine Major Trends Reshape the Future Landscape. (December 2025). https://blog.leowang.net/content/files/2025/12/2026-2035-------------------------------------------------------.pdf

[2] Translated by MoneyDJ. AI Data Centers Trigger Power Crisis! Goldman Sachs: U.S. Reserve Capacity Insufficient, China May Overtake. Yahoo Finance Hong Kong. (January 2026). https://hk.finance.yahoo.com/news/ai資料中心引爆電力危機-高盛-美備用不足-陸恐後來居上-015054049.html

[3] Caijing. How Can China’s Computing Power Become Stronger? Wall Street CN. (January 4, 2026). https://wallstreetcn.com/articles/3762510

[4] AIBase. U.S. Power Grid Crisis: The Future of the AI Industry Hinges on Power Supply! (December 9, 2025). https://news.aibase.com/zh/news/24416

[5] Goldman Sachs Report. Sino-U.S. Tech War, China’s Advantages. Tencent News. (December 18, 2025). https://view.inews.qq.com/a/20251218A05JXV00

[6] AJ Securities. AI and Data Centers Remain in High Boom, Strong Demand for Power Equipment — Tracking Report on Data Center Power Supply and Distribution Equipment Industry. (December 16, 2025). https://pdf.dfcfw.com/pdf/H3_AP202512161801712716_1.pdf

[7] SemiAnalysis. How AI Labs Solve the Power Crisis: In-depth Analysis of On-site Natural Gas Power Generation. Wall Street CN. (January 1, 2026). https://news.futunn.com/hk/post/66856259/semianalysis-in-depth-report-us-power-grid-lags-behind-as

[8] Long Yue. AI Devours Power, Small Modular Reactors (SMRs) Become the Key Solution. Wall Street CN. (January 5, 2026). https://wallstreetcn.com/articles/3762540

[9] Sina Finance. AI Boom Spawns New Growth Driver in the Energy Industry. (January 6, 2026). https://finance.sina.com.cn/stock/usstock/c/2026-01-06/doc-inhfkfur5671575.shtml