As a research-led investor, I decided to do my job and do some research. Here’s what I found. LCOE bad. baseload good. My conclusion: no nuclear probably isn’t the answer. Well at least not the full answer. It’s a partial answer. I’m not here to tell you Amazon, Google, Microsoft and Meta are wrong. But I sort of am?

I’m skeptical on timeline and cost. While we’re talking about scaling to 10GW and then 100GW datacentres by 2030ish, NuScale have been plugging away for 15 years and despite regulatory approval, cancelled it’s first project because of a lack of interest. If it ever gets off the ground it won’t be before 2030. Yikes, right?

Cost is really the ball game. While NuScale's levelized cost of electricity (LCOE) ranges from $89-135/MWh, beating traditional nuclear power's $110-160/MWh, it remains far more expensive than combined cycle gas at $45-70/MWh and solar plus storage at $30-60/MWh. And the truth is nuclear prices keep going up whilst the cost of solar/wind/batteries keep going down. The arrows are all pointing in the wrong direction. I’m not saying a bet on nuclear is a bet against solar/wind+batteries, because we are likely to need all the power generation we can get, but it sure limits the market potential.

Nuclear and SMRs will still a significant role because data centers require "six nines" reliability (99.9999% uptime), and nuclear offers a basically unmatched combination of reliability, location flexibility, and energy security that makes it an ideal baseload power source. Unlike gas plants, nuclear facilities can store years worth of fuel on-site, eliminating supply chain vulnerabilities. While solar and wind will likely provide cheaper supplementary power, nuclear could serve as the backbone of a hybrid power solution that delivers AI from the plug reliably.

This is just SMRs in the context of serving datacentres. Nuclear and SMRs have a larger role in decarbonising the economy, especially replacing gas plants. This decarbonization mission involves providing reliable power for industrial processes, district heating, and general grid stability where the economics and timeline constraints might be less demanding than AI infrastructure needs. In these applications, SMRs can compete more effectively with alternatives because they're solving different problems - grid stability, energy security, and industrial decarbonization - rather than racing to meet the explosive growth of AI power demand. The timeline mismatch that makes SMRs challenging for data centers becomes less problematic when planning long-term grid transitions, and the higher costs might be more acceptable when weighed against the full system value of reliable, carbon-free baseload power.

If you just wanted a primer, you can go now. Take care.

Here’s a summary of SMRs before you:

  1. Pressurized Water Reactors: Uses regular water for cooling - the most proven and widely used design.
  2. Sodium-Cooled Fast Reactors: Uses liquid sodium for cooling, which allows for better fuel efficiency.
  3. Molten Salt Reactors: Uses hot liquid salt that acts as both the fuel and coolant.
  4. High-Temperature Gas-Cooled Reactors: Uses helium gas to cool special fuel pellets at very high temperatures
  5. Fast Reactors (Other Coolants): Experimental reactors using different coolants like lead.
  6. Heat Pipe Reactors: The simplest design, using metal tubes filled with liquid metal to move heat.
  7. Fluoride Salt-Cooled Reactors: Uses regular nuclear fuel but cools it with liquid salt instead of water.

Technical Summary

Let’s start from the beginning to get on the same page. Small Modular Reactors (SMRs) are small. More specifically they are smaller than other nuclear reactors. Traditional nuclear outputs between 1,000-1,600 megawatts electric (MWe) per reactor. SMRs usually go at 50-300 MWe per module. Also they are modular. To massively overgeneralise (why else are you here?), they are designed to be built in factories not on-site. Imagine a production line where standardized nuclear reactor components are assembled in a controlled environment, with consistent quality checks and workers who become increasingly efficient at building the same design over and over. These completed modules are then shipped to the power plant site by truck, rail, or ship - essentially arriving as nuclear "LEGO" pieces ready to be assembled. The Ford Assembly line for nuclear basically.

SMRs are not a single thing. The architecture varies across designs. Light Water Reactor (LWR) based SMRs, currently the most mature, utilize proven PWR technology scaled down and modified for modular construction. These systems typically operate at pressures of 15-16 megapascals (MPa) and temperatures around 315°C. (Espresso machines operate at around 0.9 MPa, so this is like 17 espresso machines combined). Non-LWR designs include molten salt reactors operating at atmospheric pressure with coolant temperatures exceeding 600°C, enabling higher thermal efficiencies. High-temperature gas-cooled reactors, using helium as coolant, can reach temperatures of 750°C, making them suitable for industrial process heat applications. And sodium-fast reactors operate at low pressure with coolant temperatures around 500°C, offering better fuel efficiency through closed fuel cycles.

Now, numbers. The nuclear power market makes up ~10% of global electricity, about $350-400 billion annually and 32% of zero-carbon electricity generation across ~440 operating reactors worldwide. So SMRs have a large market to sell into, but the bull case for SMRs is not to replace existing nuclear, it’s to massively grow that 10% of global electricity. The IAEA projects that nuclear power capacity could double by 2050 in scenarios focused on meeting climate goals, with the majority of that coming from SMRs. But datacentres are a totally new customer outside of those forecasts.

Reactor Types