--- title: "AI 尽头是电力！美股公用事业世纪并购后，AI 数据中心新 “卖铲人” 电力股站到聚光灯下" type: "News" locale: "en" url: "https://longbridge.com/en/news/287004650.md" description: "随着 AI 数据中心对电力需求的激增，电力供应成为其扩张的关键制约因素。新纪元能源以 670 亿美元收购道明尼能源，标志着电力行业的重大变革。预计到 2030 年，全球数据中心用电量将翻倍，北美 AI 数据中心电力需求年复合增长率高达 73%。" datetime: "2026-05-20T04:29:07.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/287004650.md) - [en](https://longbridge.com/en/news/287004650.md) - [zh-HK](https://longbridge.com/zh-HK/news/287004650.md) --- # AI 尽头是电力！美股公用事业世纪并购后，AI 数据中心新 “卖铲人” 电力股站到聚光灯下 As the global capital markets are fiercely competing for GPU computing power and HBM memory bandwidth, a more covert yet disruptive mainline is reshaping the energy landscape in North America and even globally. Electricity has become the most critical constraint in the construction of AI data centers. However, the current issue is not a lack of total electricity supply, but rather a triple mismatch—grid access speed is far behind the pace of data center construction, regional power infrastructure is aging and cannot support explosive growth, and traditional approval processes are nearly ineffective under the time scale of the AI era. ## The "Mismatch Crisis" of Electricity Supply and Demand: An Underestimated Systemic Bottleneck According to Zhitong Finance APP, a significant acquisition this week has revealed this signal. On May 18, Nextera Energy (NEE.US) announced a $67 billion acquisition of competitor Dominion Energy (D.US)—the largest utility acquisition in U.S. history. What drove this transaction was not any traditional energy strategy, but the insatiable thirst for electricity from global AI data centers—Dominion Energy holds over 51 GW of data center contract demand, which Nextera Energy CEO John Ketchum described as "a once-in-a-lifetime demand growth." ![455d34ca654654899498c22f5ba09b0.png](https://imageproxy.pbkrs.com/https://img.zhitongcaijing.com/image/20260520/1779248990260832.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg) This transaction is the latest sign that electricity supply has become a key constraint on the expansion of AI data centers. As IEA data shows, global data center electricity consumption is expected to grow by 17% in 2025, reaching 485 terawatt-hours, and the IEA predicts that this figure will double to approximately 950 terawatt-hours by 2030. This article synthesizes the latest industry dynamics to comprehensively analyze the deep logic of the AI electricity bottleneck from both supply and demand sides, the restructuring path of the industrial chain, and the investment value of related listed companies. ![31e5cc8e8c079a595aba6060bfc947d.png](https://imageproxy.pbkrs.com/https://img.zhitongcaijing.com/image/20260520/1779249223301790.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg) According to industry research reports, the three-year compound annual growth rate of electricity demand for AI data centers in North America is as high as 73%, while the pace of grid access and interconnection approvals stands in stark contrast. On May 16, 2026, a report from Monitoring Analytics revealed the most critical negative signal of this contradiction: the electricity prices in the largest power market in the U.S., PJM Interconnection, experienced an "irreversible" significant increase, soaring by 76%. The report rarely used the term "irreversible," indicating that the electricity supply and demand structure is undergoing fundamental changes This "mismatch crisis" is being transmitted outward from the grid connection end, reshaping the entire energy supply chain. As pointed out at the Data Center World 2026 conference, the speed of expansion of power infrastructure has become a de facto limit to AI deployment. "The combination of distributed on-site generation and large-scale real estate platforms is becoming a viable path to bridge the supply-demand gap." ![201fa9eb5562b3d8b45329ee4deeeed.png](https://imageproxy.pbkrs.com/https://img.zhitongcaijing.com/image/20260520/1779249258631896.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg) From a longer-term investment perspective, analysis from the energy research firm AInvest indicates that power infrastructure investors are viewing AI as a lasting demand driver—investable asset classes in power infrastructure are undergoing structural repricing. If the IEA's forecast for renewable energy capacity continues to grow alongside the demand for grid infrastructure upgrades over the next decade, institutional capital inflows into the AI power supply sector may still be in the structural early stages. ## Industrial Ecosystem Reshaping: From "Behind-the-Meter" Power Supply to "Energy as a Service" In the face of grid bottlenecks, the power supply model for data centers is undergoing structural innovation, with emerging models in the power energy industry attracting particular attention. **"Behind-the-Meter Generation": A Radical Solution to Bypass Grid Bottlenecks** Traditional data centers typically obtain power from the grid through utility companies, but in light of the reality that grid connection delays can last for years, an increasing number of operators are turning to behind-the-meter (BTM) generation models—using natural gas or other energy sources to generate power on-site. The appeal of this model is multifaceted: it allows data center operators to bypass the queue for grid access, gaining a first-mover advantage in timing; direct power supply eliminates transmission and distribution costs, significantly enhancing cost predictability; and self-built power sources can ensure the required power quality and supply reliability standards. However, the costs are equally evident—large-scale natural gas generation faces equipment supply shortages, pressure from emission reduction commitments, and community approval resistance. Data shows that the prices of natural gas generator sets have risen by 10% to 15% annually due to the surge in demand from AI data centers, and this upward trend is expected to continue at least until 2028. The delivery cycle for gas turbines has extended to 1 to 3 years, with queue times now extending beyond 2028—these core supply-side bottlenecks are forming new constraints. Industry insiders further point out that even if companies adopt behind-the-meter power supply models, it does not mean that overall energy pressure will decrease; the related burdens are merely shifted from the grid to the natural gas supply chain, and the tension in the energy market may persist **Giga-level "Energy as a Service": A Paradigm Shift from Assets to Services** On May 12, 2026, Hitachi, Ltd. and the American energy infrastructure investment and operation company X LABS announced a strategic partnership to jointly develop giga-level "energy parks" for AI data centers in North America, providing a one-stop solution integrating power generation, energy storage, and energy management in the model of "Energy as a Service" (EaaS). These energy parks are designed as on-site power supply hubs, integrating multiple power generation sources, energy storage facilities, transmission and distribution infrastructure, and energy management systems, capable of providing reliable and controllable large-scale power to data centers while coordinating with the regional power grid. The cooperation plan will provide full lifecycle services covering design, development, operation, and power supply through a project-specific special purpose vehicle (SPV) — X LABS is responsible for SPV operation, project financing, and site development, while Hitachi provides giga-level transmission and distribution technology, grid stability solutions, and BESS battery energy storage systems. The core innovation of this model lies in its shift of power supply from capital expenditure to operational expenditure, allowing data center operators to quickly access stable power without bearing the burden of large upfront capital investments or managing complex energy operations. The two companies plan to complete the construction of the first energy park in the early 2030s. Although this is not a short-term solution in terms of timeline, the maturity of its model may become an important variable in addressing the power bottleneck of AI data centers. **BESS: From Diesel Replacement to AI Load Buffering** In the ecosystem of data center power infrastructure, Battery Energy Storage Systems (BESS) are playing an increasingly important role. Traditional diesel backup power solutions have exposed many limitations in AI load scenarios, while BESS is not only becoming a viable alternative but also taking on an increasingly critical new function — responding to rapid and significant power fluctuations caused by AI workloads. AI chips can experience drastic changes in power load on a microsecond timescale during training and inference, presenting unprecedented challenges to the power quality of the grid. Ram Nagappan, Vice President of AI Infrastructure at Oracle Cloud Infrastructure, succinctly explained: "Traditional grids and power plants cannot cope with the massive power fluctuations generated by AI data centers, which can occur multiple times per second. BESS can act as a buffer to suppress these power fluctuations, which is becoming a new requirement for AI data centers." MarketsandMarkets predicts that the global BESS market size will grow from USD 50.81 billion in 2025 to USD 105.96 billion by the end of 2030; BloombergNEF reports that global newly installed BESS capacity will reach 112 gigawatts in 2025, an increase of 48% compared to 2024. More significantly, BESS is increasingly being deployed in conjunction with fast-response natural gas generators and synchronous condensers to form hybrid architectures — in projects like Caterpillar's with American AIP in West Virginia, and in transactions where Baker Hughes supplies NovaLT gas turbines to Frontier Infrastructure BESS has become a standard component for enhancing the transient response capability of systems. ## Full Industry Chain Benefit Map: AI Power Investment Matrix from Chips to Concrete The power demand of AI data centers is penetrating every corner of the industry chain, forming a multi-layered benefit chain from large independent power producers to traditional utility companies, building materials suppliers, and specialty electronic distributors. **Primary Beneficiaries: Independent Power Producers and Nuclear Newcomers** Talen Energy (TLN.US) is one of the independent power producers most directly linked to the power supply of AI data centers. The company owns and operates approximately 13.1 GW of U.S. power infrastructure, including 2.2 GW of nuclear capacity at the Susquehanna nuclear power plant. In June 2025, Talen signed an $18 billion, 17-year power purchase agreement (PPA) with Amazon AWS to supply up to 1,920 megawatts of carbon-free power, which will complete the grid connection reconfiguration from the back to the front of the meter during the refueling outage at Susquehanna in the spring of 2026. In early February 2026, Talen completed the acquisition of 2.6 GW of natural gas generation assets in the PJM market from Energy Capital Partners, further enhancing its flexibility and bargaining power in data center co-location negotiations. On March 19, 2026, Talen signed a letter of intent with X-energy to collaborate on small modular reactors (SMRs) to assess the feasibility of deploying Xe-100 advanced reactors in Pennsylvania and the PJM market. Oklo (OKLO.US) represents the cutting-edge "nuclear energy direct supply to AI" energy model. The company recently launched a market capitalization-based equity financing plan of up to $1 billion, backed by an underwriting group composed of top Wall Street investment banks such as Goldman Sachs, Bank of America, Citigroup, and JP Morgan. The company's core projects include the Idaho Aurora nuclear plant (scheduled for first commercial operation in 2028), a 1.2 GW power supply agreement with Meta (Ohio campus, with the first phase expected to go live as early as 2030), and AI modeling collaboration with NVIDIA, attempting to build a dual empowerment system of "nuclear energy + AI infrastructure." In the first quarter of 2026, Oklo's net loss narrowed to $33.1 million (loss per share of $0.19), below the market expectation of $0.20. Vistra Energy (VST.US) signed a 20-year power purchase agreement with Meta in January 2026, selling the entire 2,176 megawatt capacity of the Perry and Davis-Besse nuclear plants to Meta to support its AI data center operations. The company's adjusted EBITDA in the first quarter of 2026 reached a historic high of $1.494 billion. Vistra proposed a core strategy of "speed to power," promoting co-location, demand response, and gas bridging solutions for ultra-large cloud customers to address grid connection delays The company's core profit guidance for the full year of 2026 is between $6.8 billion and $7.6 billion, significantly higher than the $5.7 billion to $5.9 billion for 2025. **Secondary Beneficiaries: Traditional Utilities and Power Infrastructure** On May 18, 2026, Nextera Energy acquired Dominion Energy for $67 billion, driven primarily by the demand for over 51 GW of data center contracts held by Dominion Energy—equivalent to the installed capacity of about 50 large nuclear power plants. Dominion Energy's service area covers the "data center corridor" in Northern Virginia, where it is estimated that about 70% of global internet traffic passes through daily. American Electric Power (AEP.US) and Idacorp (IDA.US) are traditional electric companies whose grid operations are also related to this theme. AES Corporation (AES.US) signed a 20-year power supply agreement with Google in February 2026 to power its new data center in Wilbarger County, Texas, using a co-located generation model. As of the announcement date, AES has signed nearly 12 GW of energy agreements with data center customers, of which 9 GW are PPA directly signed with hyperscale cloud service providers. In March 2026, AES announced it would be privatized in a $33 billion deal, expected to be completed by the end of 2026 or early 2027, which will reshape the company's financial structure in expanding clean power to serve hyperscale data centers. Hitachi (HTHIY) has partnered with X LABS to enter the forefront of the AI data center energy service market through its gigawatt-level energy park collaboration. Hitachi will provide systematic solutions such as high-voltage transmission and distribution, grid stability, and power quality through Hitachi Energy, integrating BESS energy storage technology and the HMAX energy AI optimization platform. **Tertiary Beneficiaries: Construction Supply Chain and Specialized Equipment** The spillover effects of AI infrastructure investment have begun to permeate the construction materials and specialized equipment supply chain. Prologis (PLD.US), a traditional logistics real estate giant, is accelerating its layout for data center power access. In the first quarter of 2026, the company launched $2.1 billion in new development projects, with $1.3 billion allocated for two data center projects. Its data center power reserves—locked in or in progress power access capacity—have reached 5.6 GW, becoming a strategic pillar business for the company. Prologis is also collaborating with NVIDIA, EPRI, and InfraPartners to assess deploying 5 to 20 megawatt micro-prefabricated data centers near underutilized utility substations, aiming to have at least five pilot projects enter the development phase by the end of 2026. Insteel Industries (IIIN.US) is a direct beneficiary of the data center construction boom in the steel supply chain. In the first quarter of 2026, the company's revenue grew by 23.3% year-on-year to $159.9 million, and net profit soared by 602.4% to $7.59 million. Management views data centers and IIJA infrastructure projects as the dual engines of strong demand in 2026—"We believe that data center projects will serve as a timely transitional bridge before the recovery of traditional private non-residential projects." The company plans to invest approximately $20 million in 2026 for upgrades to its factories and information systems to support growth. Richard Yu Electronics (RELL.US) is positioned in the upstream segment of this investment cycle for semiconductor equipment. The company is entering a "long-awaited upcycle," driven by high-margin wafer manufacturing equipment business and long-term demand for AI data centers. Management expects this economic cycle to exceed the typical 6 to 12 months equipment investment cycle. ![9941897fc17363a3ef35dce64889c31.png](https://imageproxy.pbkrs.com/https://img.zhitongcaijing.com/image/20260520/1779249328707632.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg) In addition, other companies in this field include Alliant Energy (LNT.US), TransAlta (TAC.US), Capital Power (CPXWF), and Central Puerto SA (CEPU.US). ## Summary While the global capital markets are still celebrating the addition of thousands of GPUs each quarter, the real bottleneck in the AI industry is quietly shifting—from chip wafer factories to transformers and transmission lines in the power grid. NextEra's $67 billion acquisition of Dominion essentially represents the largest vote of confidence in the future judgment of "power rather than chips." This investment theme is worth continuous attention not only because of its scale—AES with a contracted capacity of 12 GW to Prologis with a 5.6 GW power pipeline—but also because it possesses unique anti-cyclical resilience. Regardless of how AI model architectures iterate, how chip processes evolve, or how geopolitical situations fluctuate, once data centers are built, their power demand becomes an almost rigid long-term contract revenue stream. In an industry where technological changes accelerate and competitive landscapes fluctuate dramatically, certainty itself is the most valuable scarce asset. Power—the most unassuming underlying component of AI infrastructure—may indeed be the most certain long-term investment logic for the next decade ### Related Stocks - [D.US](https://longbridge.com/en/quote/D.US.md) - [NEE.US](https://longbridge.com/en/quote/NEE.US.md) - [NEP.US](https://longbridge.com/en/quote/NEP.US.md) - [JXI.US](https://longbridge.com/en/quote/JXI.US.md) - [GRID.US](https://longbridge.com/en/quote/GRID.US.md) - [EMLP.US](https://longbridge.com/en/quote/EMLP.US.md) - [XLU.US](https://longbridge.com/en/quote/XLU.US.md) - [6501.JP](https://longbridge.com/en/quote/6501.JP.md) - [HTHIY.US](https://longbridge.com/en/quote/HTHIY.US.md) - [ORCL.US](https://longbridge.com/en/quote/ORCL.US.md) - [CAT.US](https://longbridge.com/en/quote/CAT.US.md) - [BKR.US](https://longbridge.com/en/quote/BKR.US.md) - [TLN.US](https://longbridge.com/en/quote/TLN.US.md) - [AMZN.US](https://longbridge.com/en/quote/AMZN.US.md) - [XE.US](https://longbridge.com/en/quote/XE.US.md) - [OKLO.US](https://longbridge.com/en/quote/OKLO.US.md) - [GS.US](https://longbridge.com/en/quote/GS.US.md) - [BAC.US](https://longbridge.com/en/quote/BAC.US.md) - [C.US](https://longbridge.com/en/quote/C.US.md) - [JPM.US](https://longbridge.com/en/quote/JPM.US.md) - [META.US](https://longbridge.com/en/quote/META.US.md) - [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md) - [VST.US](https://longbridge.com/en/quote/VST.US.md) - [AEP.US](https://longbridge.com/en/quote/AEP.US.md) - [IDA.US](https://longbridge.com/en/quote/IDA.US.md) - [AES.US](https://longbridge.com/en/quote/AES.US.md) - [GOOGL.US](https://longbridge.com/en/quote/GOOGL.US.md) - [GOOG.US](https://longbridge.com/en/quote/GOOG.US.md) - [PLD.US](https://longbridge.com/en/quote/PLD.US.md) - [IIIN.US](https://longbridge.com/en/quote/IIIN.US.md) - [NEE-N.US](https://longbridge.com/en/quote/NEE-N.US.md) - [NEE-W.US](https://longbridge.com/en/quote/NEE-W.US.md) - [NEE-R.US](https://longbridge.com/en/quote/NEE-R.US.md) - [NEE-T.US](https://longbridge.com/en/quote/NEE-T.US.md) - [NEE-U.US](https://longbridge.com/en/quote/NEE-U.US.md) - [NEE-S.US](https://longbridge.com/en/quote/NEE-S.US.md) - [NEE-V.US](https://longbridge.com/en/quote/NEE-V.US.md) - [ORCL-D.US](https://longbridge.com/en/quote/ORCL-D.US.md) - [W4VR.SG](https://longbridge.com/en/quote/W4VR.SG.md) - [BAC-Q.US](https://longbridge.com/en/quote/BAC-Q.US.md) - [BML-L.US](https://longbridge.com/en/quote/BML-L.US.md) - [BAC-L.US](https://longbridge.com/en/quote/BAC-L.US.md) - [BAC-K.US](https://longbridge.com/en/quote/BAC-K.US.md) - [BML-H.US](https://longbridge.com/en/quote/BML-H.US.md) - [BAC-N.US](https://longbridge.com/en/quote/BAC-N.US.md) - [BAC-E.US](https://longbridge.com/en/quote/BAC-E.US.md) - [MER-K.US](https://longbridge.com/en/quote/MER-K.US.md) - [BAC-O.US](https://longbridge.com/en/quote/BAC-O.US.md) - [BML-G.US](https://longbridge.com/en/quote/BML-G.US.md) - [BAC-M.US](https://longbridge.com/en/quote/BAC-M.US.md) - [BAC-B.US](https://longbridge.com/en/quote/BAC-B.US.md) - [BML-J.US](https://longbridge.com/en/quote/BML-J.US.md) - [BAC-S.US](https://longbridge.com/en/quote/BAC-S.US.md) - [BAC-P.US](https://longbridge.com/en/quote/BAC-P.US.md) - [8648.JP](https://longbridge.com/en/quote/8648.JP.md) - [C-R.US](https://longbridge.com/en/quote/C-R.US.md) - [JPM-M.US](https://longbridge.com/en/quote/JPM-M.US.md) - [JPM-C.US](https://longbridge.com/en/quote/JPM-C.US.md) - [JPM-D.US](https://longbridge.com/en/quote/JPM-D.US.md) - [JPM-L.US](https://longbridge.com/en/quote/JPM-L.US.md) - [8634.JP](https://longbridge.com/en/quote/8634.JP.md) - [JPM-K.US](https://longbridge.com/en/quote/JPM-K.US.md) - [JPM-J.US](https://longbridge.com/en/quote/JPM-J.US.md) - [NVD.DE](https://longbridge.com/en/quote/NVD.DE.md) ## Related News & Research - [NextEra Energy's $67 Billion Dominion Acquisition Will Make It the Dominant Power Player in the AI Era and a Must-Own Energy Stock](https://longbridge.com/en/news/286896538.md) - [BREAKINGVIEWS-Electricity mega-deal strains balance of AI power](https://longbridge.com/en/news/286801856.md) - [Combined NextEra-Dominion Would Have 130-GW Large-Load Pipeline](https://longbridge.com/en/news/286830722.md) - [AI Forced Two Utilities to Merge… Here’s How to Profit](https://longbridge.com/en/news/287107470.md) - [What NextEra and Dominion's giant utility merger means for your electric bill](https://longbridge.com/en/news/286827978.md)