Analysis of the Surge in Data Center Power Demand: Why Big Tech Capital is Preempting SMRs (Small Modular Reactors) [EN]

"Nuclear power plants are the only energy source that can provide reliable, carbon-free power 24 hours a day, every day, in any weather."

— Joe Dominguez, Constellation Energy CEO (September 2024)

 

[Prologue: The Market Observer's Perspective]

For the past decade, the dominant narrative of the global energy industry was simple. Solar panel prices dropped by over 85%, wind farms expanded along coastlines, and companies declaring '100% renewable energy' filled the first pages of ESG reports. However, starting in 2024, cracks began to appear in this narrative. The epicenter of the rupture is the AI data center.

What this observer focuses on is not merely the increase in power demand, but the structural shift. In the past, data centers were passive consumers of the power grid. Now, hyperscalers like Microsoft, Google, and Amazon are directly signing contracts to restart nuclear power plants, investing hundreds of millions of dollars into SMR startups, and hiring nuclear engineers en masse. They are seeking to become infrastructure owners, not just customers of the energy market. The moment power was elevated to a core factor of production for AI competitiveness, the energy governance structure was inevitably realigned. The question this report tracks is singular: Is Big Tech capital's preemption of nuclear infrastructure a passing trend, or a structural power shift in the energy system?

EXECUTIVE SUMMARY

According to the IEA (International Energy Agency) 2025 report, global data center power consumption is projected to more than double from approximately 415 TWh in 2024 to 945 TWh by 2030, with a compound annual growth rate exceeding four times that of overall power demand growth.^¹ The core driver of this demand surge is the zero-intermittency requirement of AI workloads, a condition that existing renewable energy sources structurally struggle to meet. In response, Amazon, Google, and Microsoft are injecting massive capital directly into nuclear partnerships.^² Big Tech's entry into nuclear power is not a simple power procurement strategy. It is a vertical integration movement within the industrial structure, directly linking AI computing competitiveness with energy infrastructure ownership.

01. Anatomy of the Demand Shock: What the Numbers Say

The Physical Scale of Power Consumption Altered by AI

As of 2024, global data center power consumption stands at roughly 415 TWh, accounting for 1.5% of global electricity consumption. This figure has grown at an annual average of 12% over the past five years, but the AI boom is accelerating this trajectory.^¹ The IEA's Base Case scenario projects that data center power consumption will reach 945 TWh by 2030, a scale equivalent to the current annual electricity consumption of the entire nation of Japan.^³

What matters is not just the scale, but the density and concentration. Unlike EV charging or air conditioning, which represent millions of distributed demand points, data centers exhibit a centralized demand structure where a single hyperscale campus ceaselessly consumes as much power as a mid-sized city 24/7. According to the IEA, while traditional data centers demand 10 to 25 MW, AI-specialized hyperscale centers often exceed 100 MW. This figure corresponds to the annual consumption of 100,000 typical households.^⁴

The Qualitative Shift in Power Density Created by AI Accelerators

AI-specialized GPU-accelerated servers consume several to dozens of times more power per unit compared to traditional CPU servers. The IEA's 2025 Energy and AI report estimates that a single AI language model query consumes about 2 watt-hours (Wh), large inference models like DeepSeek-R1 consume more than double that, and short video generation requires approximately 25 times more.^⁵ Even if the physical footprint of the data center remains the same, the power density of the computational resources housed within is structurally increasing. This qualitative shift, combined with quantitative growth, is placing an unexpected burden on the power grid.

02. The Structural Limitations of Renewable Energy: Why Nuclear?

The Physical Constraints of Intermittency

The technical standard of the data center industry is the so-called 'Five Nines,' representing 99.999% annual availability. Translated into annual downtime, this is a mere 5.26 minutes. Solar and wind cannot guarantee this level of uptime without sufficiently scaled battery storage systems. Because a power interruption means not just an inconvenience, but a catastrophic financial loss. Theoretically, large-scale Battery Energy Storage Systems (BESS) could bridge this gap, but for gigawatt-scale hyperscale demand, it remains a technically and economically unproven solution.

The IEA's 2025 Energy and AI report forecasts that while renewable energy will remain the fastest-growing supply source for incremental data center power demand, natural gas and coal will cover a significant portion of the additional demand by 2030. It is analyzed that only post-2030 will SMRs be fully deployed as stable, low-carbon baseload power sources.^⁶

The Strategic Revaluation of Baseload Power

The capacity factor of nuclear power is typically over 90%, placing it in a completely different category from solar (around 25%) and wind (around 35%). Nuclear power plants operate reliably 24 hours a day, 365 days a year, regardless of the weather. According to the IEA, in 2024, nuclear power accounted for less than 8% of total U.S. generating capacity yet provided 18.5% of the electricity supply.^⁷ This is precisely where it aligns with the power profile demanded by AI data centers. AWS CEO Matt Garman explicitly articulated this context: "We're going to need gigawatts of power in the coming years, and wind and solar projects alone won't be able to meet the demand."^⁸

03. Big Tech Capital's Entry into Nuclear: Anatomy of the Deals

The Structure and Scale of Vanguard Deals

2024 marks the inaugural year when Big Tech's foray into nuclear transitioned from expressions of intent to signed contracts. The structures of these core transactions are summarized based on data.

Microsoft × Constellation (Crane Clean Energy Center)

On September 20, 2024, Constellation Energy signed a 20-year Power Purchase Agreement (PPA) with Microsoft, announcing the Crane Clean Energy Center project to restart Three Mile Island (TMI) Unit 1 (835 MW), which was shut down in 2019 due to economic reasons.^⁹ The total restart investment is approximately $1.6 billion, and in 2025, the Trump administration supported the project with a $1 billion federal loan. The target restart date, originally 2028, was brought forward by a year to 2027.^¹⁰

Google × Kairos Power

On October 14, 2024, Google signed an agreement with Kairos Power, marking the first SMR Power Purchase Agreement in the company's history. It secures 500 MW through 6 to 7 fluoride salt-cooled SMRs, with the target operational date for the first unit set for 2030.^¹¹ Subsequently, in August 2025, it completed an additional contract by signing the first Gen IV reactor PPA with the Tennessee Valley Authority (TVA) to receive 50 MW from the Hermes 2 plant in Oak Ridge.^¹²

Amazon × X-energy / Energy Northwest

On October 16, 2024, Amazon Web Services led a roughly $500 million investment in X-energy, signing an agreement with the Energy Northwest consortium in Washington state to develop four SMRs (Phase 1 at 320 MW, up to 960 MW). Amazon and X-energy aim to bring more than 5 GW of SMR capacity into commercial operation in the U.S. by 2039.^¹³

04. Macroeconomic Ripple Effects: Reorganization of the Energy Governance Structure

The Historical Anomaly of Capital Allocation

Global data center investment has nearly doubled since 2022, reaching $500 billion in 2024. This investment boom is sparking concerns about a surge in electricity demand.^⁶ The capital expenditure competition among the four major hyperscalers proves this. Amazon executed $85.8 billion in annual CapEx in 2024, among which nuclear-related investments are expanding to billions of dollars in the form of direct investments, PPAs, and loan guarantees.

Vertical Integration of Energy Infrastructure Ownership

The crux of this trend is that Big Tech is shifting from being an energy 'buyer' to an 'owner.' Amazon's equity acquisition in X-energy and board participation (securing 2 seats) is a move that goes beyond mere PPA contracts and enters the ownership structure of production technology.^¹³ Daniel Gross, Director at the Amazon Climate Fund, directly explained the essence of this strategy: "PPAs alone cannot power advanced nuclear. We must directly bridge the capital gap required for development, licensing, and construction."^¹⁴

Changes in Traditional Utilities and Energy Pricing Structures

Big Tech's direct investment in nuclear is acting to fill the SMR development capital gap that traditional utility companies have avoided due to risk. However, this process also entails structural impacts on energy prices. Goldman Sachs projects that data centers will account for 8% of total U.S. power demand by 2030, more than doubling from the current 3%.^¹⁵ The structure wherein demand surges drive up grid capacity market prices, with a portion of those costs being passed onto general electricity rates, is already being observed in certain regions of the U.S.

05. Historical Analogy Comparison: Precedents of Standard Wars and Infrastructure Monopolies

Structural Similarities to the 1890s War of the Currents

The 'War of the Currents' in the 1890s, where Edison's direct current (DC) system clashed with Westinghouse-Tesla's alternating current (AC) system, is a case where capital power and preemptive infrastructure standardization determined the industry structure, rather than pure technical superiority. Just as Westinghouse established the AC standard by winning the Niagara Falls hydroelectric contract then, today's Big Tech companies are becoming the agents effectively selecting the technical standard by funneling massive orders to specific firms before SMR design standards are even finalized. Google's advance order of 7 units from Kairos Power is a prime example.^¹⁶

The Precedent of Cloud Infrastructure Preemption and Structural Lock-in

In the early 2000s, while standardizing its internal IT infrastructure, Amazon created the cloud infrastructure known as AWS, and this preemptive move became the foundation for the current structural oligopoly of the global cloud market. There is a possibility that the preemptive acquisition of SMR nuclear infrastructure will similarly reinforce dominance over AI computational infrastructure. The gap in the operating cost structure between hyperscalers who control their own power supply and small-to-medium AI companies that do not will act as a variable reshaping the competitive landscape of the entire AI industry in the long run.

06. Variables and Limitations: Applying the Brakes on Structural Optimism

Cost and Timeline Risks of SMRs

The investment rationale for SMRs carries a fundamental vulnerability in that it is being constructed while the technology remains commercially unproven. Currently, there are no SMRs in commercial operation in the U.S. or Europe. Kairos Power's Hermes 1 is a non-power demonstration reactor, and X-energy's Xe-100 is undergoing the licensing process. If construction costs continue to be revised upwards, as seen in the precedent of NuScale, the initial investment rationale could falter. The 462 MW SMR project by NuScale and UAMPS was ultimately canceled in late 2023 after projected construction costs skyrocketed dramatically since 2021.^¹⁷

Regulatory, Fuel Supply Chain, and Geopolitical Variables

In July 2024, the U.S. Congress enacted the ADVANCE Act, mandating the NRC to streamline the licensing process for small reactors, signaling an ongoing improvement in the regulatory environment. However, the supply chain for High-Assay Low-Enriched Uranium (HALEU), used as SMR fuel, currently has a high dependence on specific countries. In January 2026, the U.S. Department of Energy finalized the first HALEU contract for Kairos Power, but the diversification of the HALEU supply chain remains an ongoing challenge.^¹⁸

Macro Scenario: Probabilistic Future Trajectories

The scenarios below are constructed based on the three sensitivity cases (Base Case, Lift-Off Case, High Efficiency Case) from the IEA's 2025 Energy and AI report. Quantitative probabilities for each scenario are omitted due to the absence of explicit figures in currently available market consensus data.

Scenario A (Base Case): SMRs as a 2030s Solution, Natural Gas Bridges the Short Term

In the IEA Base Case, while renewable energy handles nearly half of the additional demand up to 2030, a structure is formed where natural gas and coal cover a significant portion of the remainder.^⁶ Big Tech's nuclear investments will yield their first commercial units in the early 2030s, but until then, natural gas generation will fill the power void, causing carbon emissions to increase in the short term. The 2030 carbon-neutral goals of hyperscalers are highly likely to be materially recalibrated. Microsoft already admitting a 23.4% year-over-year increase in carbon emissions for fiscal year 2024 pre-emptively illustrates this.^¹⁹

Scenario B (Structural Shift Case): Accelerated SMR Technology Standardization and Cost Reduction

If the Google-Kairos-TVA Hermes 2 project (targeting 50 MW by 2030) succeeds as scheduled, it will become an historical milestone generating the first commercial data for Gen IV SMRs.^¹² The IEA's Lift-Off Case forecasts that under such an accelerated scenario, data center power demand in 2035 will exceed 1,700 TWh, 45% higher than the base scenario.^¹ If the initial commercial operation data lowers investment uncertainty, a scenario of rapid SMR deployment acceleration opens up from the mid-2030s onward.

Scenario C (Tail Risk Case): SMR Cycle Derailment due to Cost Spikes and Regulatory Delays

If SMR construction costs continue to be revised upwards like the NuScale precedent, HALEU supply chain bottlenecks delay production schedules by years, and changing political environments halt federal support, a significant portion of Big Tech's nuclear bets will remain in long-term limbo without concrete results until the early 2030s. In this case, as mentioned in the IEA's High Efficiency Case, the market could be restructured towards software and hardware efficiency improvements curbing power demand growth; however, power shortages would become entrenched as the actual bottleneck for AI infrastructure expansion, exacerbating geographical imbalances in computational access.^¹

Conclusion

The surge in AI power demand is not merely an increase in energy consumption. This is an event where the strategic status of 'electricity' as a factor of production has been fundamentally elevated. The moment the intermittency limits of renewable energy collided with the baseload requirements of hyperscale data centers, the market sought a new solution, and nuclear power was summoned anew as that solution. In this process, the phenomenon of Big Tech capital directly entering the ownership structure of nuclear infrastructure is breaking down the boundaries between the energy industry and the tech industry.

What the observer must focus on is not the timeline of when SMRs will be switched on. It is the structure. Big Tech becoming the agent effectively determining technical standards by placing massive advance orders for specific SMR designs, the possibility of energy infrastructure being vertically integrated by a few companies like cloud computing, and how the roles of traditional utility companies and national regulatory authorities are redefined in the process—this is the core lens for understanding the next decade. Energy has already become the most definitive bottleneck in the AI race. Whoever controls that bottleneck first will preempt the governance structure of the next cycle.

---

Sources and References

¹ IEA, Energy and AI — Energy Demand from AI (2025) — https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai

² Utility Dive, Amazon announces small modular reactor deals (2024.10.16) — https://www.utilitydive.com/news/amazon-small-modular-reactor-deals-nuclear-dominion-x-energy-energy-northwest/730022/

³ IEA, News Release: AI is set to drive surging electricity demand from data centres (2025) — https://www.iea.org/news/ai-is-set-to-drive-surging-electricity-demand-from-data-centres...

⁴ IEA, Artificial Intelligence Topic Page (2025) — https://www.iea.org/topics/artificial-intelligence

⁵ IEA, Energy and AI — Full Report PDF (2025) — https://iea.blob.core.windows.net/assets/dd7c2387-2f60-4b60-8c5f-6563b6aa1e4c/EnergyandAI.pdf

⁶ IEA, Energy and AI — Executive Summary (2025) — https://www.iea.org/reports/energy-and-ai/executive-summary

⁷ Carbon Credits, Google, Kairos and TVA Unlock Advanced Nuclear Energy (2025.08.20) — https://carboncredits.com/google-kairos-and-tva-unlock-advanced-nuclear-energy-for-americas-ai-data-centers/

⁸ CNBC, Amazon goes nuclear, to invest more than $500 million (2024.10.16) — https://www.cnbc.com/2024/10/16/amazon-goes-nuclear-investing-more-than-500-million-to-develop-small-module-reactors.html

⁹ Constellation Energy, Crane Clean Energy Center Launch Announcement (2024.09.20) — https://www.constellationenergy.com/news/2024/Constellation-to-Launch-Crane-Clean-Energy-Center...

¹⁰ NucNet, Constellation Secures $1 Billion Federal Loan For Three Mile Island Restart (2025) — https://www.nucnet.org/news/constellation-secures-usd1-billion-federal-loann-for-three-mile-island-restart-11-3-2025

¹¹ Google Blog, Google signs advanced nuclear clean energy agreement with Kairos Power (2024.10.14) — https://blog.google/outreach-initiatives/sustainability/google-kairos-power-nuclear-energy-agreement/

¹² World Nuclear News, Google, Kairos Power, TVA announce collaboration (2025.08.18) — https://www.world-nuclear-news.org/articles/google-kairos-power-tva-announce-collaboration

¹³ X-energy, Amazon Invests in X-energy to Support Advanced SMRs (2024.10.16) — https://x-energy.com/media/news-releases/amazon-invests-in-x-energy...

¹⁴ Latitude Media, Inside Amazon's nuclear power investment strategy (2025.04.02) — https://www.latitudemedia.com/news/inside-amazons-nuclear-investment-strategy/

¹⁵ CNBC, Constellation Energy to restart Three Mile Island, sell the power to Microsoft (2024.09.20) — https://www.cnbc.com/2024/09/20/constellation-energy-to-restart-three-mile-island-and-sell-the-power-to-microsoft.html

¹⁶ World Nuclear News, Google and Kairos Power team up for SMR deployments (2024.10.15) — https://www.world-nuclear-news.org/articles/google-and-kairos-power-team-up-for-smr-deployments-in-us-first

¹⁷ The Register, Amazon drops $500M on X-energy's SMRs (2024.10.16) — https://www.theregister.com/2024/10/16/amazon_nuclear_smr/

¹⁸ Wikipedia, Kairos Power (Updated March 2026) — https://en.wikipedia.org/wiki/Kairos_Power

¹⁹ Data Centre Magazine, Microsoft & Constellation's Bid to Restart Three Mile Island (2025.06.30) — https://datacentremagazine.com/critical-environments/microsoft-constellation-restarting-a-nuclear-reactor

※ Disclaimer
This report does not recommend the purchase or sale of any specific assets, nor does it support or criticize any specific regime, government, or politician. It is an article of macroeconomic system analysis based on publicly disclosed data and historical indicators. It is impossible to predict all market variables, and the responsibility for all judgments and subsequent consequences rests entirely with the reader. While the author (Neutral Observer) makes every effort to ensure the reliability of the analysis, the flawless accuracy of the provided information is not guaranteed.