The "Power Wall" Facing AI Data Centers

The evolution of AI is generating an unprecedented scale of electricity demand. Training large-scale language models (LLMs) of the GPT-4 class requires tens of thousands of GPUs/TPUs running at full capacity for months, consuming dozens of GWh of electricity per single training cycle. In the inference phase as well, an era has arrived where billions of users utilize AI on a daily basis, and data center power consumption is increasing exponentially.

According to the report "Electricity 2025" published by the IEA (International Energy Agency) in 2025, global data center power consumption is projected to double from approximately 415 TWh in 2024 to 945 TWh by 2027. Looking further ahead, AI data centers alone are forecast to require 176 GW of power capacity by 2035. This is a staggering figure, equivalent to approximately three times the total generating capacity of France's nuclear power plants (roughly 61 GW).

Against this demand, renewable energy sources such as solar and wind power face fundamental constraints. First, the capacity factor of solar power is only about 25% and wind about 35%, making them ill-suited for data centers that require stable operation 24 hours a day, 365 days a year. Second, large-scale renewable energy plants require vast tracts of land, and construction near data centers is often geographically difficult. Third, grid capacity has already become a bottleneck — in the United States, obtaining permits for new large-scale transmission lines takes an average of more than ten years.

Against this backdrop, data center operators have developed an urgent need for "a power source that can be installed on-site, is compact yet high-output, and capable of operating around the clock." The answer that has rapidly emerged is the Small Modular Reactor (SMR).

What is SMR: Differences from Conventional Nuclear Reactors

SMR (Small Modular Reactors) are a modular nuclear power generation system in which reactors with an output of 300 MW or less are mass-produced in factories, then transported by truck or rail to the site for assembly. Unlike conventional large-scale reactors (output of 1,000 MW or more), which are constructed on-site over more than a decade, SMRs aim to reduce construction periods to 3–5 years through factory production, significantly improving cost predictability.

SMRs offer several technical advantages. First, passive safety systems (such as natural circulation cooling) eliminate by design the risk of core meltdown during loss of external power—as occurred in the Fukushima Daiichi accident. Second, their modular nature provides the flexibility to incrementally increase output in response to demand. Third, with capacity factors exceeding 90%, they possess ideal characteristics as a stable baseload power source for the needs of data centers.

From an investor's perspective, the most significant aspect is the potential for SMRs to bring "software-like scalability" to nuclear power. Conventional reactors were custom-built one at a time—analogous to handcrafting a prototype with every iteration in the semiconductor industry. SMRs seek to standardize and mass-produce reactors the way TSMC mass-produces chips, achieving cost reductions through learning curve effects.

Funding and Business Strategies of Major SMR Startups

Currently, the SMR market has multiple competing startups with different technical approaches, each with distinct investor bases and business models.

NuScale Power is the first mover in the SMR industry and the only company to have received design certification from the NRC (U.S. Nuclear Regulatory Commission) for an SMR in 2023. It went public through an IPO and is traded on public markets. However, its first project in Utah, the "Carbon Free Power Project (CFPP)," saw costs balloon threefold from an initial estimate of approximately $3 billion to roughly $9 billion, and the project was cancelled in November 2023. This experience left important lessons for the entire SMR industry. NuScale's failure posed a fundamental question to the market—"Are SMRs truly cost-competitive?"—and subsequent startups have all made cost management and securing customer commitments their top priorities.

Oklo, chaired by Sam Altman, went public in 2024 via SPAC for approximately $306 million. The company is developing a fast neutron reactor called "Aurora," which has a unique technological advantage in its ability to recycle spent nuclear fuel. Its moves in 2025–2026 have been remarkable, including a 12 GW power supply agreement with data center giant Switch and a contract with Meta for up to 1.2 GW. The 12 GW scale is equivalent to twelve standard nuclear power plants, an exceptional scale for an SMR startup order. A hallmark of Oklo's strategy is its power-as-a-service model. Rather than purchasing an SMR, customers buy electricity from SMRs that Oklo builds and operates. This applies the cloud computing business model to energy, dramatically reducing upfront investment risk for customers.

Kairos Power is developing a molten salt-cooled high-temperature reactor and attracted attention with a 500 MW power supply agreement with Google—the first instance of Google making a direct commitment to nuclear power, which rapidly elevated SMR credibility across the technology industry. Kairos plans to complete construction of its demonstration reactor "Hermes" in Tennessee during 2026 and begin initial test operations.

TerraPower, founded by Bill Gates in 2008, has raised more than $650 million in cumulative funding. Notably, Nvidia's investment arm NVentures participated in its funding. The fact that the world's largest GPU/AI chip maker made its first direct investment in an energy company to address the power consumption challenges posed by its chips symbolizes the severity of the power crisis facing the AI industry. TerraPower's "Natrium" reactor features a unique design combining a sodium-cooled fast reactor with a molten salt thermal energy storage system, allowing flexible output adjustment from 345 MW to 500 MW. In March 2026, TerraPower obtained the first-ever construction permit from the NRC for an SMR and broke ground on construction in Kemmerer, Wyoming. This is a landmark event in nuclear regulatory history, signifying that SMRs have transitioned from "designs on paper" to "power plants actually being built." The target for commercial operation is 2030. Meta has also recently announced a power supply agreement with TerraPower, and Meta intends to secure a combined total of 6.6 GW of nuclear power through agreements with Oklo, TerraPower, and Vistra.

X-energy employs high-temperature gas reactor (HTGR) technology and has raised more than $1.4 billion in cumulative funding. Its 5 GW power supply agreement with Amazon is one of the largest Big Tech nuclear contracts by value. Amazon is also pursuing power procurement from Constellation Energy's existing nuclear power plants, making nuclear power a pillar of its data center energy strategy.

Last Energy and Aalo Atomics are emerging players that have each raised approximately $100 million. Last Energy focuses on the European market and is developing factory-manufactured microreactors (20 MW class). Aalo Atomics was founded by former SpaceX and NASA engineers and targets the distributed energy market with smaller, lower-cost SMRs.

Radiant has raised more than $300 million and is developing portable microreactors (1 MW class). Its primary targets are military bases and remote locations rather than data centers, and it has its sights set on contracts with the U.S. Department of Defense.

Big Tech's Nuclear Power Grab — Over $10 Billion in Commitments

From the second half of 2024 through 2026, the scale of nuclear investment by Big Tech companies escalated rapidly. Looking at this development comprehensively, it becomes clear that the AI infrastructure race has structurally shifted from a "competition for computing power" to a "competition for power procurement capability."

Microsoft's moves are the most emblematic. In September 2024, Microsoft signed an agreement with Constellation Energy worth an estimated $16 billion to restart TMI Unit 1, which had been shut down following the Three Mile Island accident in 1979. The unprecedented decision to invest over $10 billion in "reviving" a nuclear power plant speaks volumes about how acute the power demand from AI operations has become. For Microsoft, the growth of its AI business through its partnership with OpenAI cannot be realized without securing power, and nuclear energy is positioned as an "indispensable infrastructure investment."

Google, in addition to its 500 MW contract with Kairos Power, is actively engaging in nuclear policy through its participation in the Nuclear Energy Institute. Google's Vice President of Data Centers has indicated that SMRs are essential for achieving the goal of "24/7 carbon-free energy," and the prevailing view in the industry is that the Kairos contract is only the first of many.

Amazon is advancing a diversified energy strategy, including nuclear power, centered on its 5 GW contract with X-energy. AWS's data center expansion plans require tens of additional gigawatts of power per year, and the company has adopted a portfolio approach that avoids dependence on any single technology or supplier.

Meta is the most aggressively pursuing contracts with multiple SMR startups. The company plans to secure a total of 6.6 GW of nuclear power, including agreements with Oklo (up to 1.2 GW), TerraPower, and Vistra. Zuckerberg has publicly stated that current power infrastructure is insufficient for large-scale model training beyond Llama 4, and his remarks that "the future of AI hinges on securing energy" drew significant attention in the investor community.

The combined nuclear-related commitments of these four Big Tech companies easily exceed $10 billion (approximately ¥1.5 trillion). This scale represents a historic turning point for the nuclear industry, signaling that the traditional nuclear business model — long dependent on government subsidies — is fundamentally shifting toward a demand-driven model powered by private technology companies.

Explosive Expansion of VC Investment — 2025 Marks an All-Time High of $2 Billion

Venture capital investment in nuclear startups reached a record high of approximately $2 billion in 2025, a remarkable growth rate when compared to the $200–300 million per year seen just five years earlier in 2020.

This surge is not being driven solely by traditional "cleantech-focused VCs." Tech VCs, hedge funds, and even AI chip companies have begun deploying capital into nuclear energy.

Breakthrough Energy Ventures (founded by Bill Gates), in addition to its direct investment in TerraPower, has also invested broadly in nuclear supply chain companies, pursuing an investment strategy aimed at the entire SMR ecosystem.

Lowercarbon Capital (founded by Chris Sacca) is one of the earliest climate tech-focused VCs to enter the SMR space and has participated in seed rounds for multiple SMR startups.

NVentures (Nvidia)'s investment in TerraPower stands out as the first known case of an AI hardware company making a direct investment in the energy sector. Nvidia CEO Jensen Huang has stated explicitly that "the next bottleneck for AI is not compute, but power," and NVentures' investment decision reflects this conviction.

Jane Street (the largest quantitative trading firm)'s entry into SMR investment symbolizes the growing interest from the financial industry in nuclear energy. Data center power demand is a problem that directly affects the algorithmic trading infrastructure of financial institutions, and the investment is seen as being driven by energy security considerations.

ARK Invest (Cathie Wood) is playing a role in raising retail investor interest in the nuclear sector through aggressive purchases of Oklo shares on public markets. ARK's "Big Ideas 2026" report positions SMRs as "essential infrastructure" for the AI era and has revised its market size estimate for 2030 upward to three times its previous forecast.

Sam Altman and Bill Gates — AI Pioneers' Bet on Nuclear Power

To understand the core of SMR investment, it is essential to look at the moves of Sam Altman and Bill Gates.

Altman, as CEO of OpenAI, is one of the people in the world who most acutely feels AI's power demands, while also personally investing in both Oklo and Helion Energy (a fusion startup). His role as chairman of Oklo is not merely a portfolio investment, but a signal of his intent to drive an integrated strategy between AI and energy. Altman has stated publicly that "achieving AGI (artificial general intelligence) may require electricity comparable to the entire current US power generation capacity," and this awareness motivates his deep commitment to Oklo.

Gates's investment in TerraPower represents an 18-year long-term commitment since the company's founding in 2008, making it the longest-running energy investment by a tech billionaire. Gates has poured cumulatively billions of dollars into TerraPower, and the construction permit obtained in March 2026 marks the moment when 18 years of patience bore fruit. Gates's approach is not the typical VC pursuit of short-term returns, but a mission-driven investment aimed at "solving the energy problem that humanity needs," with his investments across the entire nuclear ecosystem through Breakthrough Energy Ventures rooted in this same philosophy.

Behind Nvidia's NVentures investing in TerraPower lies a more direct business logic. Big Tech companies, as Nvidia's customers, are struggling to secure power for their data centers, and if the power problem is not resolved, GPU demand growth will also be constrained. For Nvidia, investment in SMRs is a strategic move to expand its own TAM (total addressable market), and as its first foray into the energy sector, the symbolic significance is extraordinarily large.

Changes in the Regulatory Environment — The Path to SMR Realization Opens Up

One of the biggest bottlenecks for SMR commercialization has been the regulatory process, but significant progress was made between 2025 and 2026.

The construction permit that TerraPower obtained from the NRC in March 2026 marks the first-ever construction permit issued by the NRC for an SMR, making it the most significant regulatory milestone to date. Traditionally, the NRC's reactor licensing process often took more than a decade, effectively serving as a barrier to entry for startups. TerraPower's receipt of a construction permit is both a testament to the NRC's efforts to develop a review framework for advanced reactors and an extremely important "precedent" for future SMR companies.

The design certification that NuScale obtained from the NRC in 2023 demonstrates that the SMR design itself meets NRC safety standards, and is a separate process from the construction permit. Although NuScale's CFPP project was cancelled for economic reasons, the design certification itself remains valid and could potentially be utilized in a different project in the future.

The U.S. Congress is also increasingly showing bipartisan support for nuclear energy. The ADVANCE Act (enacted in 2024) is legislation aimed at streamlining the NRC's review process and accelerating the commercialization of advanced reactors, and is expected to significantly reduce regulatory costs for SMR startups.

International Competition — UK, France, and South Korea Advance Their Own SMR Strategies

SMR development is not a phenomenon unique to the United States. The United Kingdom, France, and South Korea are each pursuing their own SMR strategies, intensifying the international competition for technological supremacy.

In the United Kingdom, Rolls-Royce SMR is developing a 470MW-class SMR, backed by £1.8 billion (approximately ¥340 billion) in support from the British government. Rolls-Royce's strategy, in addition to construction within the UK, is oriented toward exports to European countries such as Poland and the Czech Republic, bearing the strong hallmarks of a national strategy to cultivate SMR as an "export industry."

In France, Nuward (formerly EDF SMR), a subsidiary of EDF (Électricité de France), is advancing a distinctly European SMR design. France already generates approximately 70% of its total electricity from nuclear power and harbors national ambitions to lead the world in SMR technology as well. However, Nuward's development schedule lags behind US startups, with commercial operation projected for the mid-2030s.

South Korea enacted a dedicated "SMR Special Act" in 2024, positioning SMR development as a national strategic project. The SMART reactor developed by Korea Hydro & Nuclear Power (KHNP) is advancing through a joint development initiative with Saudi Arabia, targeting expansion into the Middle Eastern market. South Korea has a proven track record in exporting the APR1400 (a large-scale reactor) to the UAE, and places equal emphasis on an export strategy for SMRs.

From an investor's perspective, the greatest competitive advantage of US SMR startups lies in their access to Big Tech as a massive "anchor customer." While SMR development in the UK and France is government-led and centered on public utilities, in the United States private technology companies are driving demand, creating conditions where innovation can be accelerated through market forces.

Criticism and Challenges — The "High-Cost Dead End" Controversy

While optimism about SMRs is dominant, important critical perspectives also exist. An accurate understanding of risk factors is essential for investment decisions, and critical analysis should not be avoided.

The most serious concern is cost. The cost overrun of NuScale's CFPP project (initial estimate ~$3 billion → final estimate ~$9 billion) cast doubt on the premise that SMRs would achieve "cost reductions through factory mass production." A 2025 paper by the University of Pennsylvania's energy policy research group condemned SMRs as a "costly dead end," pointing out that economies of scale may work less favorably than for large reactors. The paper's argument is that reducing reactor output actually raises construction costs per MW, and that cost savings from factory mass production cannot offset this disadvantage.

Anti-nuclear organization Beyond Nuclear has also continuously raised concerns about SMR safety and costs. In particular, it points out that SMRs' "passive safety" relies on unproven new technology, that spent fuel disposal remains unresolved just as with large reactors, and that even "small" reactors still generate radioactive waste whose management costs do not scale proportionally with output.

SMR proponents counter these criticisms as follows. First, NuScale's cost overrun was a problem unique to the first-of-a-kind unit, and significant cost reductions through learning curve effects are expected at the mass production stage. Second, long-term contracts with Big Tech ensure demand certainty and enable factories to maintain high utilization rates, making mass production effects more likely to materialize. Third, they argue that SMRs compete not with large nuclear reactors but with gas turbines and renewables paired with battery storage, and that their comparative advantage as carbon-free 24/7 baseload power is clear.

A sober analysis from an investor's perspective requires facing the fact that SMR economics have not yet been proven. No SMR company other than NuScale has reached commercial operation, and there can always be a significant gap between theoretical cost estimates and actual construction costs. At the same time, it is also true that Big Tech's massive contracts have substantially reduced offtake risk for SMRs, and this "demand guarantee" provides grounds for evaluating SMR economics in a different context from other clean energy technologies.

Commercialization Timeline — From Demonstration Reactors in 2026 to Mass Production in the 2030s

Organizing the upcoming timeline for the SMR industry, commercial deployment is expected to proceed in stages from 2026 through 2030.

2026: Kairos Power completes construction of the Hermes demonstration reactor in Tennessee and begins test operations. Oklo advances construction of its initial demonstration facility. TerraPower breaks ground on full-scale construction in Kemmerer, Wyoming.

2027–2028: Kairos Power and Oklo target the start of initial commercial operations. Dedicated factory construction for mass production of SMRs begins in earnest. Initial power supply to Big Tech data centers commences.

2029–2030: TerraPower's Natrium reactor begins commercial operations. X-energy completes its first unit. Mass production capacity for SMRs is established, with cost reductions demonstrated for second units and beyond.

Early 2030s: Multiple SMR companies achieve commercial operations, and mass production on the order of several dozen units per year begins. International deployment accelerates in earnest.

It should be noted, however, that nuclear projects have historically experienced chronic schedule delays, and the timeline above represents an optimistic scenario. In particular, the NRC review process, supply chain development (global manufacturing capacity for reactor pressure vessels is limited), and securing a skilled workforce could become significant bottlenecks.

Market Size and Investment Opportunities

According to aggregated forecasts from multiple research firms, the global SMR market is projected to grow from approximately $6.3 billion (around ¥950 billion) in 2024 to approximately $13.8 billion (around ¥2.07 trillion) by 2032 (CAGR of approximately 10%). However, given that Big Tech's power demand is expanding at a pace exceeding projections, these market forecasts are likely to be revised upward.

Investor exposure to the SMR market is structured across the following layers. First, direct investment in SMR developers (publicly listed companies such as Oklo and NuScale, or private investment in unlisted companies such as TerraPower and X-energy). Second, uranium mining and nuclear fuel supply chains (such as Cameco and Kazatomprom). Third, nuclear-related components and services companies (such as BWX Technologies and Curtiss-Wright). Since Big Tech's nuclear contracts propagate demand across these entire supply chains, broad investment opportunities exist beyond SMR developers alone.

Impact on the Industry

The rapid expansion of SMR investment is having a compound impact on the AI industry, the energy sector, and the investment community.

First, "power procurement capability" has emerged as a new variable that determines the winners and losers in the AI infrastructure race. While the axes of AI competition have historically been model performance, GPU/TPU procurement volume, and talent acquisition, going forward, "how quickly, how cheaply, and how stably a company can secure large amounts of electricity" will become a determining factor in AI companies' competitiveness. Early commitment to SMRs is a strategic decision that will shape the growth potential of AI businesses five to ten years from now.

Second, the nuclear power industry's business model is undergoing fundamental change. Traditional nuclear power was a "quasi-public utility" heavily dependent on government regulation and subsidies, but Big Tech's demand has driven a massive influx of private capital, accelerating innovation based on market principles. While this shift may bring "the speed and flexibility of a technology startup" to the nuclear industry, it also creates a new challenge: balancing safety assurance with commercial interests.

Third, there are important implications for Japan's energy policy and industrial strategy. Japan has maintained a cautious stance toward nuclear power since the Fukushima Daiichi disaster, but as SMRs rapidly move toward commercialization in the United States, the risk of Japan being left behind in nuclear technology innovation is growing. Japan's accumulated nuclear technology expertise is among the best in the world, and companies such as Mitsubishi Heavy Industries and Hitachi have relevant SMR technologies. However, domestic SMR development projects are significantly behind those in the United States, and this gap cannot be ignored from the perspective of industrial competitiveness.

The definitive answer to whether SMRs can truly solve "the AI power crisis" will not come until the first commercial SMR begins operation around 2030. However, the fact that technology industry leaders—including Bill Gates, Sam Altman, and Nvidia—are all simultaneously betting on nuclear power strongly suggests that SMRs are not merely a buzzword, but a leading candidate for the energy infrastructure of the AI era. What matters for investors is the strategic judgment of how to build exposure to the massive structural trend of power demand from the AI industry, with a clear-eyed understanding of the technological risks and long commercialization timelines associated with SMRs.

References: IEA "Electricity 2025" report; NRC TerraPower construction permit announcement (March 2026); Oklo SEC Filing (SPAC listing, Switch/Meta contracts); Kairos Power–Google 500 MW contract announcement; TerraPower–NVentures investment announcement; X-energy–Amazon 5 GW contract announcement; Microsoft–Constellation Energy TMI restart contract; Meta 6.6 GW nuclear procurement plan announcement; University of Pennsylvania "SMR: A Costly Dead End?" (2025); Beyond Nuclear SMR criticism report; NuScale CFPP cancellation SEC disclosure; ARK Invest "Big Ideas 2026"; Rolls-Royce SMR UK government support announcement; South Korea SMR Special Act (enacted 2024); BloombergNEF SMR market size forecast