Power Under Constraint: Exploding Electricity Demand, Nuclear’s Delivery Problem, and the SMR Bet

Chapter 1. Problem Statement: Meeting Explosive Electricity Demand While Achieving Sustainability 

In recent years, the industrial structure of advanced economies has been reorganizing around electricity. Hyperscale data centers for training and running artificial intelligence (AI) models, advanced semiconductor fabs, electrification of transport through electric vehicles, and the expansion of electricity-based industrial processes are all pushing power demand to unprecedented levels. 

Data centers, in particular, can require hundreds of megawatts (MW) at a single site. To put that scale into context, a common rule of thumb is that 1 MW of continuous power can support on the order of a few hundred to several hundred average U.S. homes, depending on assumptions about average household load. 

The point, however, is not simply that “we need to generate more electricity.” At the same time, international climate governance requires deep cuts in emissions from the power sector. Under the Paris Agreement (adopted in 2015), countries agreed to hold the increase in global average temperature well below 2°C above pre-industrial levels, while pursuing efforts to limit warming to 1.5°C. In practice, this is implemented through national climate action plans known as Nationally Determined Contributions (NDCs), which each country must prepare, communicate, and update over time. 

The broader system is supported by organizations with distinct roles: the UNFCCC Secretariat supports implementation and maintains registries such as the NDC registry; the IPCC assesses the scientific evidence on climate change and mitigation options (policy-relevant but not policy-prescriptive); and the IEA provides energy data, analysis, and scenario pathways that shape how governments and industries think about energy security and decarbonization. 

Crucially, the IEA has warned that AI and the data-center buildout are likely to become major drivers of electricity demand growth. In an April 2025 IEA release, the agency projected that electricity demand from data centers worldwide could more than double by 2030, with AI as the most significant driver of that increase. This “digitalization” dynamic is not just about consumer gadgets; it refers to the wider economic shift toward cloud computing, data-intensive services, and AI-enabled production systems—changes that translate directly into rising demand for round-the-clock electricity in servers, networking, and cooling infrastructure. 

This creates a dual constraint: electricity must expand, but carbon emissions must fall sharply. If the only objective were to meet demand, it might seem easiest to expand fossil-fuel generation. But if emissions are tightly constrained by climate targets, attention necessarily shifts toward low-carbon power sources. 

Among renewables, solar and wind are scaling rapidly, yet they remain fundamentally constrained by time and location: output varies by weather, season, and day-night cycles. For a high-load industrial society—especially one that expects 24/7 reliability for data centers and critical manufacturing—this intermittency raises the question of how to secure a stable supply at the required scale, even as storage and grid technologies improve. In my view, solar and wind will continue to advance and can meaningfully contribute, including at the household level; but making them the single dominant national backbone for firm, always-available power still appears technically and systemically challenging under current conditions.

At the same time, policy design around distributed solar can quickly change the economics of adoption. For example, on August 7, 2025, the California Supreme Court unanimously ordered a lower court to revisit its earlier decision that had upheld California regulators’ roughly 75% cut in compensation rates for rooftop solar customers who sell excess electricity back to utilities (net metering). Importantly, the Supreme Court did not decide the underlying legality of the policy change itself; it required the appellate court to re-examine the issue. This episode shows that even when rooftop solar is technically viable, its adoption and commercial logic can hinge on regulatory frameworks—how the grid values exported electricity, who bears system costs, and how public benefits are weighed. 

Against this background, nuclear power has re-entered the debate as one of the few large-scale sources that can simultaneously provide: (1) high-volume electricity supply, (2) high reliability, and (3) low direct carbon emissions. A key advantage often cited is its high capacity factor—the share of time a plant actually produces electricity relative to its maximum possible output. In the United States, the nuclear fleet has achieved an average annual capacity factor of at least 90% in every year since 2012, reflecting its ability to deliver steady power with relatively few interruptions. For an economy increasingly defined by electricity-hungry, continuous operations, nuclear therefore becomes structurally difficult to ignore when discussing feasible pathways to meet rising demand while maintaining climate commitments. 

Chapter 2. Nuclear Power and SMRs: Technical Foundations, Structural Limits, and Risk Reconfiguration 

Nuclear power plants are fundamentally heat engines. Energy released from nuclear fission is first turned into heat, then used to generate steam, which spins a turbine to produce electricity. The scale of fission energy is often explained by the idea that a very small loss of mass can be converted into a large amount of energy. 

Reactor operation depends on controlling the chain reaction, which can be summarized by one key concept: whether each “generation” of neutrons produces enough neutrons to sustain the next generation. If the reactor produces just enough, it runs steadily. If it produces more than enough, power rises. If it produces less than enough, power falls. In practice, reactor design and operation revolve around keeping this balance stable under real conditions. 

To sustain efficient fission, reactors also manage the energy (speed) of neutrons. In many designs, fast neutrons are slowed down using moderator materials such as hydrogen, deuterium, carbon, or beryllium, because slower neutrons are more effective at sustaining the chain reaction in common fuel cycles. 

Another crucial issue is neutron leakage. In theory, if neutrons never escaped the fuel region, sustaining the chain reaction would be easier. In reality, some neutrons leak out of the core, which reduces the ability to sustain the reaction. This creates a clear design trade-off: core geometry and size affect how much neutron leakage occurs, and leakage directly affects how easily the reactor maintains steady operation. 

From the standpoint of power production, nuclear plants also face the limits of any thermal system. Light-water reactors typically reach about 32–36% efficiency, and while thermodynamics sets an ideal upper bound (the Carnot limit), real plants operate below it because practical systems always have losses. 

These fundamentals apply to both large reactors and SMRs. The more difficult part is not the physics—it is execution. Large conventional nuclear projects have often struggled because nuclear power is capital-cost-dominant. The main cost pressures come from designing, licensing, financing, and building the plant, rather than from buying fuel. That matters because construction delays do not just push schedules; they raise financing costs and can sharply inflate total project cost. 

This is why nuclear economics are commonly discussed using a framework that asks: over the full lifetime of a plant, how much total spending is required (construction, operation, fuel), discounted over time, relative to how much electricity is produced, also discounted over time. In that framework, nuclear tends to be unusually sensitive to discount rates and construction timelines because so much cost is incurred upfront. 

Social and political constraints also matter. Major accidents damaged trust, spent fuel disposal remains contentious, and many reactors worldwide are aging, making lifetime management a persistent issue. As a result, the gap between technical feasibility and real-world delivery can be large. 

SMRs attempt to change this risk profile through standardization and modularization: factory fabrication, on-site assembly, and repeated deployment to capture NOAK learning effects. Their modular scale (often around ~300 MWe) enables staged additions, potentially lowering upfront capital burden and spreading financing risk, and siting on existing nuclear or retired fossil sites may ease integration and acceptance. 

In market terms, net-zero pathways often assume major nuclear expansion by mid-century, with some scenarios assigning a meaningful share to SMRs. SMRs are also positioned for non-grid uses such as industrial heat, hydrogen production, and dedicated data-center power. However, commercialization is conditional on aligned technology, policy, finance, and supply chains—especially HALEU availability and nuclear-grade certification and quality systems (e.g., ISO 19443). 

Engineering trade-offs become sharper as reactors shrink. Smaller cores tend to increase neutron leakage, which may require higher enrichment or optimized reflector design. That can create new trade-offs, including greater reliance on specialized fuel supply and changes in waste characteristics. 

Passive safety concepts often rely on natural circulation: the system depends on density differences in the coolant to drive flow without active pumping. For passive safety to work, the key requirement is that heat removal must stay ahead of decay heat over time, which demands high-confidence thermal–hydraulic and structural validation. 

SMRs also tend to rely more heavily on digital instrumentation and control. That raises concerns about common-cause failure, meaning failures that can affect multiple redundant systems at once if they share the same design or software. Hydrogen monitoring and control remain essential due to explosion risk. 

Overall, SMRs do not eliminate fundamental nuclear challenges: they do not remove high-level waste issues, HALEU and manufacturing infrastructure can be bottlenecks, FOAK projects still face cost-overrun risk, and SMRs are better understood as reallocating and restructuring risk rather than eliminating it. 

Chapter 3. My View 

I support SMRs, but not because of the simplistic claim that “smaller means cheaper.” The real value of SMRs lies in their structural advantages: staged investment, repeatable deployment, and integration with existing infrastructure. In that sense, SMRs are not a technology meant to replace renewables or existing large reactors; they are better understood as a complementary option within a broader power system. 

That said, success is conditional. SMRs will require a stabilized HALEU supply chain, disciplined design freeze and repeat builds, stronger independent verification of digital I&C, and a proactive waste-management strategy established in advance. Complex energy systems do not allow simple answers. The future of SMRs depends less on technological optimism than on the ability to acknowledge complexity and engineer around it with precision. 

At the same time, I do not view SMRs as a perfectly ideal technology. Nuclear power still relies on a multi-step thermal conversion pathway—turning fission energy into heat, then steam, then mechanical motion, and finally electricity. Because energy conversion is never lossless, each step introduces inefficiency. Even in a simplified view with multiple stages, compounding losses can be significant; for example, if each stage operated at 80% efficiency, then starting from 100 units of energy, only about 33 units would reach the final output. 

For this reason, I am also interested in high-temperature fuel cells such as SOFCs (solid oxide fuel cells) and MCFCs (molten carbonate fuel cells). Because they can convert chemical energy to electricity more directly, they may reduce conversion losses and improve overall efficiency. In the next post, I will examine recent research trends in these technologies.