Nuclear Renaissance: Can Business Model Innovation Lower Cost? (Part II)

A key bottleneck of the nuclear industry is its unit economics, as noted in our previous scene-setter. The cost of nuclear simply hasn’t fallen much, despite deep and distributed knowledge globally of designing and building reactors for seven decades.

Without overcoming the cost challenge, the nuclear renaissance will fizzle out before it even gains much steam. Bending the cost curve of nuclear over the medium term, however, is less about technology and is almost entirely about business model innovation that’s focused on achieving the two Ss: standardization and scale.

In other words, nuclear plants need to be built differently and many more reactors need to be built. There are essentially two ways to do that: build much larger conventional nuclear plants with up to half a dozen reactors a la the South Korean model or build many small modular reactors (SMRs) that can provide distributed baseload power. In Part II of this nuclear renaissance series, we focus on SMRs to examine its potential in lowering the cost of nuclear.

Getting To a Dozen SMRs

It is notoriously difficult to determine cost curves of an embryonic industry, especially one where few real-life projects are in play at the moment. That said, the promise of SMRs is to halve the cost/kW of electricity generated compared to conventional nuclear reactors, which translates into roughly a fall from $5 billion to $2.5 billion for a 1 GW plant.

In other words, SMRs are pushing nuclear to become a volume business. That scale isn’t an outlandish number either—various estimates suggest that SMRs will reach cost parity with conventional nuclear reactors and achieve commercial viability after about a dozen reactors are built, according to US Department of Energy projections. After that inflection point, the cost curve is expected to continue falling if the manufacturing pace is sustained at five to ten reactors per year.

The problem, of course, is the current lack of volume. Only two projects, one Russian and one Chinese, are operational. Russia’s floating Akademik Lomonosov plant is composed of two tiny 35 MW SMRs that began commercial operation in May 2020, and Russia is already starting on its first land-based SMR. Meanwhile, China’s Shidao Bay SMR project (a 150 MW high-temperature gas-cooled reactor) went online in 2023 while its ACP 100 Linglong One is expected to come online by 2026.

The sparse data on these so-called “first-of-a-kind (FOAK)” projects aren’t promising so far. The Russian and Chinese projects took twice as long to build as the average conventional nuclear plant that are much larger (see Figure 1). But at least they’re built, whereas the fate of SMR projects in the United States has been worse. For instance, US startup NuScale canceled its once-promising SMR project near Idaho Falls due to cost overruns. What was supposed to be $4.2 billion ended up being $9.3 billion, or more than $20 million/MW, about 2-4 times the average cost of conventional nuclear reactors.

Figure 1. SMRs Today Take Longer to Build Than Conventional Nuclear Reactors
Source: International Atomic Energy Agency (IAEA); author’s calculations.

Of course, FOAK projects are inherently risky and cost prohibitive. Being a first mover in a capital expenditure-driven industry like nuclear is often about paying for the steep learning curve. So while SMRs carry with them similar types of challenges associated with capex-intensive energy segments—from lack of funding and astronomical costs to delays and labor force deficiency—there are silver linings on the horizon that could make SMRs realize its intended potential.

Demand and Fordism

For SMRs to move from the FOAK proof-of-concept stage to scale, demand holds the key. Like for most long-term energy projects, investors and funders like to see stable and guaranteed markets for that energy source. That much-needed demand for SMRs could well come from the explosion in artificial intelligence (AI) growth.

It’s well-known by now that sustaining AI is as much a software story as it is an energy one. The electricity that data centers need to power and train AI could double to 35 GW by 2030 in the United States alone. Globally, data centers already consumed an estimated 460,000 GWh of electricity in 2022, which is also projected to more than double in two years. This can be seen in the capex of top “hyperscalers” or large-scale data centers for cloud services and AI—they have grown at an average CAGR of 30% over the past five years (see Figure 2).

Figure 2. Hyperscalers’ CAPEX Have Substantially Gone Up
Source: Bank of America Global Research; Newmark Market Report.

This matters for SMRs because data centers need 24/7 electricity, and for many big tech companies, they would prefer to reduce their carbon footprint by having clean energy rather than coal to power their data centers. SMRs can meet both of those criteria much better than solar and wind. If current data center energy demand projections hold up, it would require more than 1,000 100 MW SMRs to power them. In other words, data center demand alone would easily allow SMRs to get to scale.

But that is not all. SMRs can also be part of the coal phase-out solution in various countries. Some 2,000 GW of installed coal capacity globally need to be replaced in the coming decades, and China alone would need between 3,500 to 10,000 units of SMRs to break up with coal. For instance, phasing out old coal plants in rural areas provide promising opportunities to tap the unique advantages of SMRs, as the case of an SMR project in Wyoming is demonstrating.

It would appear that, “if you build (SMRs), they will come,” as demand does not seem to be a huge obstacle. What is more difficult to achieve is to standardize and scale the manufacturing process. In this sense, the SMR industry needs more Fordism.

To wit, there are currently over 80 different SMR designs under various stages of development across 19 countries (see Figure 3). Assuming all such projects pan out, that amounts to over 14 GW of SMRs over the next decade.

Figure 3. Most Planned SMR Projects Are Set To Be Completed by the 2030s
Source: World Nuclear Association; company websites and filings; and various media.

While that capacity sounds impressive, it’s not clear whether many of the proposed projects are built with a single design or assembled in a factory, certainly not the two FOAK SMRs that are currently operational. In fact, the proposed SMRs include seven different reactor designs, though pressurized water reactor appears to be the winning design so far (see Figure 4).

Figure 4. Currently Planned SMRs Rely on Various Designs
Source: World Nuclear Association; company websites and filings; various media; and author’s calculations.

Such heterogeneity in SMR projects is the exact opposite of how the emergent industry intends to drive down costs. Like Ford’s Model T of its time, SMRs need to be built in modular fashion under a giant factory roof with uniform components and standardized processes. The efficiencies from that assembly line approach were responsible for lowering the Model T’s cost by 60% in a decade. Of course, SMRs aren’t cars, but to lower cost and reach scale requires the same Fordism concept of simplifying design and standardizing production.

Herein lies the chicken and egg problem. Building SMRs alone are already capex-heavy, it is difficult for startups to also invest in large factories without some assurances of future demand. In this sense, SMRs are closer to aircraft, in that new aircraft developments are expensive because they require new processes and materials, with each aircraft being at least a 20-year asset. This is why aircraft manufacturers often sign deals on new models that are still in development to ensure demand.

That demand for SMRs does seem to be there, so will SMR factories now follow? Given the number of projects in the pipeline, the “factory-ization” of SMRs isn’t as much of a pipe dream as it was a few years ago. The next 5-7 years could well see this new business model of building nuclear reactors take root. And should it be realized, it would transform swathes of the suburban American landscape, dotted with data centers flanked by miniature nuclear plants.

Amy Ouyang is a research associate at MacroPolo. You can find her work on the global energy transition and its intersection with the economy, technology and industrial policy here.