US Power Grid Collapse Warning: AI Data Center Moratorium and the SMR Investment Trajectory [EN]
"Electrons are the new oil. But the U.S. power grid is not ready. From 2027, AI will hit the wall of 'power shortages,' and no technological innovation can solve this bottleneck. The biggest winner at this point will be the one who owns the 'pipes' called power."
— System View Macroeconomic Framework
Prologue: A Market Observer's Perspective
This report analyzes the most severe 'hidden crisis' of the tech industry in 2026. It is the paradox that at the exact moment AI technological competition is reaching its peak, the most fundamental resource to power it—electricity—is running short. In 2024, data center power demand was merely a theoretical projection. However, the situation changed rapidly from late 2025. In Q4 2025, the new pipeline of data center projects fell to half of the previous quarter, and the U.S. East Coast grid operator (PJM) began warning of a 6 GW reliability shortfall in 2027. A more decisive signal is the 'narrowing of options' for tech companies. Microsoft ($15.2B) and Meta ($10B) are no longer relying on existing grids, instead shifting capital to the Middle East and energy-rich regions. Investors are still focusing on the 'chip competition' among tech companies, but the real constraint lies in how much 'electricity to run the chips' can be secured. This difference is the core factor determining investment success or failure.
EXECUTIVE SUMMARY
The new constraint in the 2026 energy market is the Electron Gap (Power Shortage). According to the IEA baseline, data center power demand is projected to increase from 415 TWh in 2024 to 945 TWh in 2030. This translates to an annual growth rate of 15%, exceeding the global average power growth (3.5%) by more than four times. Gartner warns that power shortages will constrain over 40% of new AI data center expansions by 2027.
The industry's response has already diverged into two paths. The first is grid modernization investments by legacy power companies like NextEra Energy and Dominion Energy, which is the only realistic solution to meet short-term (2026-2027) demand. The second is next-generation nuclear technologies like NuScale's SMR (Small Modular Reactor) and Helion's nuclear fusion, scheduled for gradual commercialization post-2027. These two paths will operate complementarily, not competitively.
The restructuring of investment portfolios has already begun. During the 2000 dot-com bubble, router and fiber-optic companies skyrocketed due to increased internet traffic, but the telecommunications companies that 'owned and operated' the infrastructure maintained a long-term advantage. Even in the current AI era, no matter how fierce the competition for 'tech specs' is, there is a harsh reality: 'it is useless if you cannot supply electricity.' Therefore, the most stable profit opportunities in 2026-2028 lie in the infrastructure expansion of existing power companies, while post-2028, next-generation technologies like SMRs and nuclear fusion will provide additional growth engines.
However, this base scenario is not absolute. First, efficiency innovations in AI models could constrain the growth of power demand. Second, aggressive government energy policies (SMR subsidies, streamlined regulations) could drastically accelerate commercialization. Third, the diversification of the global power supply chain (Kazakhstan, Iceland) could ease the pressure on the U.S. grid. If these alternative scenarios materialize, the growth rates of power companies could be lower than current projections.
01. Macroeconomy: Power Shortages Redefine Tech Competition
└ Extreme Disconnect Between IEA Projections and Actual Grid Capacity (Hard Data)
The IEA's Base Case projection is shocking. Data center power demand jumping from 415 TWh in 2024 to 945 TWh in 2030 means more than a simple 'increase'; it means that a single data center sector will monopolize over 15% of global power growth. A more decisive figure is that a single AI task consumes up to 1,000 times the power of a standard web search. Due to this, the impact of a single data center on an entire regional power grid has increased exponentially.
The actual grid is not prepared to handle this. The U.S. East Coast grid (PJM) has warned of a 6 GW reliability shortfall in the 2027 phase, which implies 'planned blackouts' in major economic zones like New York and Pennsylvania. The sharp drop in the Q4 2025 data center project pipeline to 25 GW (50% of the previous quarter) is a signal that tech companies have accepted this constraint as reality.
Simultaneously, during 2025, community backlash delayed or blocked 16 data center projects ($64B). Data center moratoriums in Maine, increased environmental lawsuits in Texas, and resident pushbacks in Virginia are materializing as severe 'regulatory barriers.' Now, data center construction is redefined not as a purely technical or economic issue, but as a political and social one.
02. [Risk Transfer Timeline] The 4-Phase Trajectory of Power Shortages Transferring to Asset Markets
└ Decreased Tech CapEx → Revaluation of Power Companies' Profitability (4-Phase Breakdown)
The process by which power shortages transfer to the asset market is broken down as follows:
[Macro Trajectory] 4-Phase Impact of Power Shortages on Portfolios
| Phase | Market Signal (Trigger) | Asset Impact |
|---|---|---|
| Phase 1 (Late 2025) | Data center project pipeline halves | Reduced growth expectations for tech stocks |
| Phase 2 (Mid 2026) | Major tech firms announce delays in new data center construction | Revaluation of U.S. power companies as 'beneficiaries' |
| Phase 3 (Late 2026) | Emergency CapEx increases by existing power companies | Upward price momentum for NextEra, Dominion |
| Phase 4 (Post-2027) | Initial SMR orders / Nuclear fusion development progress | Capital shifts to next-generation energy stocks |
* Analysis Summary: Power shortages become the Initial Condition determining the 'success or failure of tech competition,' dramatically altering the direction of asset allocation.
The most crucial signal is the 'capital flight' of tech companies. As Microsoft invests $15.2B in the UAE and Meta $10B in Louisiana, they are effectively declaring that they will no longer rely on the U.S. public grid. Furthermore, signing a nuclear fusion power purchase agreement with Helion Energy is a tacit admission that 'existing energy systems have reached their limits.'
03. System Architecture: The 'Power Topology' Determining the Outcome of Tech Competition
└ Grid Capacity Becomes the New 'Monopoly'
On the surface, AI competition is being waged over 'chip performance' (GPUs, TPUs). However, underneath lies a more fundamental battle over who will monopolize the 'pipeline known as the power grid.'
Currently, the U.S. power grid is separated into three independent grids (Eastern/Western/Texas), and the capacity of each is physically limited. What is more sophisticated is that data centers do not merely consume 'power.' They require massive amounts of water for cooling, and battery storage systems are essential to prevent transmission line damage. This means the 'total infrastructure footprint' of a data center is much larger than expected.
Consequently, future data center location choices will be dictated not by 'tech' companies, but by 'energy' companies. NextEra's strategic partnership with Google, and Dominion reserving grid capacity in specific areas upon Amazon's request, are signals that this shift has already begun.
04. Reorganization of the Capital Ecosystem: Transaction Structure Between the 'Owner of Power' and the 'Pioneer of Tech'
└ Surging Tech CapEx Guarantees 'Certain Demand' for Power Companies
AI CapEx by tech companies is at an all-time high. Five major companies—Alphabet, Amazon, Meta, Microsoft, and Oracle—plan to invest $969B in CapEx this year, with $662B allocated to data center infrastructure. This is not a simple 'investment'; it implies fixed cash flows locked in by 'power purchase agreements' for the next 10 years.
Who is the biggest beneficiary in this setup? The answer is clear. The existing power companies. Because:
- ① Immediate Supply Capacity: Companies like NextEra and Dominion possess infrastructure they have operated for decades.
- ② Preemptive Regulatory Clearance: Legacy power companies have already secured community approvals and government permits.
- ③ Certain Demand: Long-term contracts (20+ years) are signed until the tech companies' data center investments 'roll over.'
- ④ Dividends + Growth: Stable cash flows strengthen their status as dividend-paying companies while simultaneously providing 5-7% EPS growth.
What about the tech companies on the other side? They now face a 'conditional entry barrier called power.' No matter how excellent a chip they make, it cannot be sold without the 'electricity' to run it. Because of this, companies like Microsoft and Meta can no longer operate under the assumption that 'we can build data centers anywhere.'
05. Historical Comparative Analysis: Lessons from the 2000 'Dot-Com Bubble'
└ Comparing Long-Term Returns of 'Tech Manufacturers' vs. 'Infrastructure Owners' (Decalcomania)
[System View Data] Comparison of Asset Performance Post-Bubble Burst
* Interpretation: During the 2000 bubble, Cisco collapsed -90% after a 3x spike. Conversely, infrastructure owner AT&T fell by -40~60%, but due to dividends and share buybacks, telecom companies overwhelmingly beat tech companies in long-term cumulative returns.
The lesson from the 2000 internet bubble is clear. 'Tech manufacturers' (routers, fiber optics) are extremely sensitive to bubbles, but 'infrastructure owners' (telecoms) generate more stable returns in the long run.
Applying this to the current situation:
- Cisco in 2000: Current Nvidia, AMD (Chip manufacturing)
- AT&T/Verizon in 2000: Current NextEra, Dominion (Power companies)
However, there is one difference. Unlike the dot-com companies of 2000, current tech companies (FAANG) are already generating massive cash flows and profits. Therefore, a 'complete collapse of tech companies' is unlikely to occur. But it is almost certain that power companies will hold the upper hand in 'relative profit-generation capabilities.'
06. System Fracture Defense Logic: Macroeconomic Refutations of Counter-Scenarios
└ Q1. "Won't the U.S. government unleash unlimited SMR subsidies for rapid commercialization?"
[Defense Logic]: That is possible. Both the Biden/Trump administrations prioritize strengthening the semiconductor and energy industries, and actually invested $575M in NuScale. However, government subsidies alone cannot eliminate 'technology risk.' NuScale is still in the 'Pilot' phase, and actual commercialization by 2028-2030 is the best-case scenario. Even if the government shortens the construction timeline by a year, it cannot resolve PJM's 2027 power shortage. Therefore, the short-term demand for 2026-2027 must inevitably be handled by legacy power companies.
└ Q2. "Won't the emergence of efficient AI models constrain power demand?"
[Defense Logic]: This is also possible. If OpenAI presents evidence that 'small models are sufficient,' data center power demand forecasts might prove excessive. However, this is merely an 'adjustment in the speed of demand,' not an 'overall demand reduction.' For example, even if the IEA's 945 TWh projection is lowered to 800 TWh, it is still nearly double the 2024 level. Moreover, as AI integration into personal devices (Local AI) progresses, the need for more powerful central data centers may paradoxically increase. Thus, it is unlikely that 'efficiency innovations' will completely solve the power shortage problem.
└ Q3. "If data centers are globally decentralized, won't the burden on the U.S. grid decrease?"
[Defense Logic]: True. Data centers will decentralize to low-cost power regions like Kazakhstan, Iceland, and Ireland. Indeed, some companies are already moving in that direction. However, this strategy has limits. First, those countries have constraints too (Ireland already declared a data center moratorium in 2024). Second, the issue of U.S. 'AI sovereignty' arises. The government will strongly support domestic data center expansions, which restricts 'global decentralization.' Third, due to latency issues, domestic U.S. data centers remain essential. Therefore, 'global decentralization' can 'mitigate' the burden on the U.S. grid but cannot solve it perfectly.
Macro Scenario: Probabilistic Future Trajectories
└ Scenario A (Base Case): 2-3 Years of Growth Led by Existing Power Companies (65% Probability)
Premise: SMR commercialization proceeds as scheduled in 2028-2030, and legacy power companies confirm their capacity to handle 2026-2027 short-term demand.
Scenario: EPS growth rates of NextEra, Dominion, and Duke Energy maintain a 5-7% annual level from 2026-2028. Dividends steadily increase (NextEra +10% in 2026, +6% post-2027), and they simultaneously fulfill dual roles as 'growth stocks' and 'dividend stocks'. If signals of an impending PJM power shortage emerge in mid-2027, emergency CapEx announcements will follow, further driving stock prices upward. In 2028, initial SMR orders begin, generating portfolio restructuring signals.
└ Scenario B (Prolonged Crisis): Prolonged Power Crisis due to SMR Delays + Tighter Regulations (20% Probability)
Premise: NuScale commercialization is delayed from 2028 to 2029-2030 due to technological/regulatory reasons, accompanied by tighter U.S. regulations (expanded data center moratoriums).
Scenario: Actual power shortages begin constraining tech companies' AI CapEx from 2027. This shakes the assumption of 'certain demand' for existing power companies. Conversely, the need for emergency infrastructure expansion is maximized, causing CapEx to surge. However, community backlash and environmental regulations extend construction timelines. Consequently, the 'growth premium' of power company stocks shrinks, leaving only dividend yields. Even if SMRs are commercialized post-2028, tech CapEx has already contracted, delaying demand recovery.
└ Scenario C (Tail Risk): Early Resolution of Crisis via AI Efficiency + Govt SMR Acceleration (15% Probability)
Premise: Companies like OpenAI release extremely efficient AI models (achieving high performance with small parameters), and the U.S. government drastically accelerates SMR construction.
Scenario: The growth rate of data center power demand falls below IEA projections due to improved AI model efficiency. Concurrently, government-led acceleration of SMR construction pushes commercial operations to start a year earlier than 2028. In this case, growth expectations for existing power companies shrink, but the upward momentum of SMR-related companies (NuScale, fusion investment funds) is maximized. Paradoxically, the 'early resolution of the energy problem' could further accelerate tech companies' AI CapEx. Here, the portfolio transition from 'legacy power companies → next-gen nuclear' proceeds rapidly.
Implications from an Investor's Perspective (Exit & Entry)
└ Entry Strategy: Position Composition by Probability
① [High Certainty] NextEra Energy (NEE) — Core Position (50%)
Target Price: Current $91-94 level → Target $95-105 by late 2026
Entry Rationale: A strategic nuclear energy partnership with Google clarifies future monetization paths, and a 10% dividend growth rate (2026) is secured as management guidance. Its technological leadership is the most advanced among U.S. power companies, providing a balanced position that combines a current 2.8% dividend yield with growth.
② [Medium Certainty] Dominion Energy (D), Duke Energy (DUK) — Satellite Position (30%)
Target Price: D $60 → $71, DUK $120 → $140
Entry Rationale: Weaker technological leadership than NextEra, but huge investment opportunities exist depending on the severity of regional power shortages. Dominion can seize emergency CapEx opportunities due to the deepening U.S. East Coast power shortages and offers a high dividend yield (approx. 4.5%). The current PER is undervalued compared to NextEra, indicating upside potential.
③ [High Risk] NuScale Power (SMR) — Hedge Position (10%)
Target Price: Current $11-12 → $60-80 upon commercial contract order
Entry Rationale: Dramatic upside potential upon successful commercialization, but remains in the 'promise' stage until 2027. NRC certification completion (May 2025) boosted technological credibility, but actual data center orders are absent. Restrict to small bets (less than 10% of the total portfolio) until the first commercial contract is signed between 2028-2030, structuring entry to immediately expand weighting upon the first order announcement.
└ Exit Strategy: Liquidation Conditions by Signal
Warning Signal 1: Official 'Data Center Construction Halt' Announcements by Tech Companies
If this is observed, growth expectations for existing power companies will fundamentally shrink. Immediately reduce growth-targeted positions by over 50%, maintaining only positions aimed at dividends.
Warning Signal 2: Regulatory Blockage of 20+ Additional Data Center Projects
If a massive amount of additional projects are blocked by regulations as in 2025, political risk surges. In this case, reduce portfolio weight to 30% levels.
Opportunity Window: SMR Commercialization Data Center Contract Order
The moment NuScale secures an actual data center contract, use this as a signal to immediately expand SMR-related company weight to 20-30%. This marks the point where the role of legacy power companies transitions from 'meeting short-term demand' to 'portfolio diversification.'
[Action Plan] Dynamic Management of Power Infrastructure & Next-Gen Nuclear Portfolios
• Upon SMR contract order announcement → Immediately expand SMR weight to 20-30%
Conclusion
The fundamental transition of the 2026 energy market has already begun. Competition in AI technology remains fierce, but victory is no longer possible through tech specs alone. This is because an AI system that does not receive sufficient power, no matter how excellent, cannot operate. This is 'the most critical constraint of 2026.'
The investment opportunities created by this reality are extremely clear. In the short term (2026-2027), legacy power companies (especially NextEra) will experience structural growth driven by surging data center demand. If PJM power shortages materialize in mid-2027, emergency CapEx will explode, driving stock prices further upward. In the medium term (2027-2030), SMRs will enter their initial commercialization phase, providing additional growth engines. Long-term, if nuclear fusion achieves commercialization, the game of the energy market will change entirely.
The wisest investment strategy is to position across all three paths while dynamically adjusting weights based on timing and probabilities. At this moment, as Sam Altman's declaration that "electrons are the new oil" translates into capital market profit opportunities, the owners of legacy power grids and the pioneers of next-generation energy technologies will reap the greatest rewards.
※ Disclaimer
This report does not solicit the purchase or sale of specific assets and is a macroeconomic dynamic analysis article based on public data. Analysis results can vary significantly depending on the data center industry, the investment direction of tech companies, regulatory policies, and the development progress of SMR and nuclear fusion technologies; all investment decisions and responsibilities lie with the investor. In particular, SMR and nuclear fusion technologies are still pre-commercialization and hold a strong speculative nature due to high technological and regulatory uncertainties.
Sources and References
[¹] Industry Journal — Surging Power Demand for AI Data Centers... Power Shortage Crisis by 2027 — 2024-11-12 — https://industryjournal.co.kr/news/239080
[²] Maeil Business Newspaper — “2027 is the Danger Zone” Power Consumption Surging Alongside AI — 2024-11-14 — https://www.mk.co.kr/news/it/11168979
[³] Chosun Ilbo — “40% of AI Data Centers Will Experience Power Shortages by 2027” — 2024-11-13 — https://www.chosun.com/economy/tech_it/2024/11/13/454FFGXYLNGQZOZ6PXVSEBCMZQ/
[⁴] The Guru — Warning Lights Flash for Securing 106GW of New Power Amid US 'Supply Chain Bottlenecks' — 2026-04-21 — https://www.theguru.co.kr/news/article.html?no=100857
[⁵] Korea Economic Daily — Surging Power Equipment Investment, Grid Restructuring & Renewable Expansion Boost the Stock Market... — 2025-10-23 — https://www.hankyung.com/article/202510249808a
[⁶] NH-Amundi Asset Management — Korea Power Equipment, A Supercycle Proven by Earnings! K-Power Infrastructure... — 2025-??-?? — https://www.nh-amundi.com/investment/pick/QahrX2iy1q1whpZR
[⁷] ZDNet Korea — “Exploding Power Demand from AI & Energy Transition... The Golden Time for Infrastructure Investment” — 2025-04-28 — https://zdnet.co.kr/view/?no=20250429132817
[⁸] Data365 — [Industry Analysis] Overview of the US Power Grid Industry in '25 and Korean Companies... — 2025-??-?? — https://data365.co.kr/?page=56&mid=data&document_srl=86547
[⁹] Introl — PJM Grid Crisis: 6GW Shortfall Constraints 40% of AI Data Centers by 2027... — 2026-02-05 — https://introl.com/ko/blog/pjm-grid-crisis-6gw-shortfall-data-center-power-2027
[¹⁰] Royfactory — AI Infrastructure and Energy Constraints: Reality and Outlook of the 2025 Bottlenecks — 2025-11-23 — https://royzero.tistory.com/entry/ai-infra-energy-bottleneck-2025
댓글
댓글 쓰기