Right, task number one, we need more energy. We can talk about efficiency all we want but cooling and optimization can only get you so far. We’ve got 250 MW facilities, we need 4x that. The best way would be to generate more of it right there next to the servers. This is on-site power generation. This gets around the problem of needing to upgrade the grid, which we will get on to, you can build reliable and carbon-friendly capacity and if you sell it right, you can contribute to national energy independence without having to rely on Russia and others. If globalization taught us anything it is that globally interconnected supply chains are a bad thing and we are all the poorer for it. Anyway, how to make loads of electrons? Three words: Small. Modular. Reactors. Also fuel cell systems, and concentrated solar power with thermal storage.
Small Modular Reactors (SMRs) represent a significant innovation in nuclear power technology, offering a scalable and potentially more flexible approach to carbon-free electricity generation. These compact nuclear reactors, typically capable of producing 50-300 MW per module, are designed to address many of the challenges associated with traditional large-scale nuclear plants. Companies like NuScale and TerraPower are at the forefront of SMR development, each with unique designs aimed at enhancing safety, efficiency, and deployability. NuScale's design, for instance, features a fully integrated reactor vessel that houses the nuclear core, steam generators, and pressurizer, all within a single unit. TerraPower's Natrium reactor, on the other hand, combines a sodium-cooled fast reactor with a molten salt energy storage system, offering both baseload power and the ability to adjust output to complement renewable energy sources. Google recently signed a partnership with Kairos Power to bring 500 MW of nuclear power with the first SMR online by 2030, followed by additional reactor deployments through 2035.
The modular nature of SMRs offers several key advantages. Firstly, it allows for scalability, enabling power output to be tailored to specific needs and potentially expanded over time. This feature is particularly attractive for data centers, which often experience growing power demands as they scale up operations. Secondly, the smaller size and standardized design of SMRs could significantly reduce construction times and costs compared to conventional nuclear plants. The contrast with traditional nuclear projects is stark. The Olkiluoto 3 reactor in Finland, a conventional large-scale plant, took approximately 17 years from the start of construction to commercial operation. Similarly, the Vogtle 3 and 4 reactors in the U.S. have faced over a decade of construction, plagued by delays and cost overruns. These examples highlight the challenges associated with traditional nuclear projects and underscore the potential benefits of the SMR approach.
Proponents of SMRs aim to dramatically reduce the timeline for nuclear power plant deployment. The goal is to bring the entire process - from planning and approvals to construction and testing - down to 6-10 years. This ambitious target reflects a recognition of the urgent need for clean, reliable baseload power in the face of climate change and growing energy demands, particularly from sectors like AI and data centers. The regulatory landscape for SMRs is evolving. While traditional nuclear plants often face 3-5 years of regulatory approvals, there's growing pressure to streamline this process for SMRs, given their standardized designs and enhanced safety features. Some argue that if the development of AI data centers were to be viewed as a matter of national or geopolitical importance, similar to the urgency seen with Operation Warp Speed during the COVID-19 pandemic, regulatory and construction timelines could be significantly compressed.
However, it's important to note that nuclear energy remains a contentious political topic. Germany, for instance, has moved away from nuclear power, influenced by a combination of historical, political, and cultural factors. France on the other hand, has embraced nuclear energy, with nuclear power accounting for about 70% of its electricity production as of 2022. France's commitment to nuclear energy is evident in its plans to increase nuclear's share in its energy mix to 50% by 2035, following a period of planned reduction.
The success of SMRs in powering AI data centers and other high-demand applications will depend on several factors: continued technological development, regulatory adaptation, public acceptance, and the ability to deliver projects on time and on budget. If these challenges can be overcome, SMRs could play a crucial role in providing the reliable, carbon-free baseload power needed to support the growing energy demands of AI and other advanced technologies.
| Reactor Type | Cost ($/kWe) | Efficiency | Safety Features | Fuel Cycle | Waste | Deployment Timeline | Pros | Cons |
|---|---|---|---|---|---|---|---|---|
| Pressurized Water Reactors (PWR) | 3,000-5,000 | 30-35% | Passive safety systems, containment structures | Enriched uranium (3-5%) | Similar to conventional reactors | Near-term (5-10 years) | Proven technology, regulatory familiarity | Higher pressure operation, complex systems |
| Sodium-Cooled Fast Reactors | 4,000-6,000 | 40-45% | Passive safety, low-pressure operation | Can use spent fuel, depleted uranium | Reduced long-lived waste | Mid-term (10-15 years) | Efficient fuel use, waste reduction | Sodium reactivity, less regulatory experience |
| Molten Salt Reactors | 3,500-5,500 | 45-50% | Inherent safety, low-pressure operation | Various fuel options, including thorium | Reduced waste volume | Mid to long-term (15-20 years) | High efficiency, flexible fuel options | Less mature technology, corrosion challenges |
| High-Temperature Gas-Cooled Reactors | 4,000-6,000 | 45-50% | Inherent safety, robust fuel design | TRISO fuel (enriched uranium) | Similar volume to PWRs, but more stable | Mid-term (10-15 years) | High temperature applications, robust fuel | Higher enrichment needs, less regulatory experience |
| Fast Reactors (Other Coolants) | 4,500-6,500 | 40-45% | Passive safety features | Can use various fuel types | Potential for waste reduction | Long-term (15-20+ years) | Efficient fuel use, waste reduction | Less mature technology, unique material challenges |
| Heat Pipe Reactors | 6,000-10,000 | 30-35% | Passive safety, simple design | HALEU (High-Assay Low-Enriched Uranium) | Minimal waste due to small size | Near to mid-term (5-15 years) | Highly portable, minimal operator needs | Higher costs, limited output |
Pressurized Water Reactor (PWR) designs, particularly those in the 300-500 MWe range, are emerging as leading candidates for near to mid-term deployment in large AI data centers. Designs like the NuScale Power Module (250 MWe per module) or the Rolls-Royce SMR (470 MWe) might be favored for near-term deployment due to their advanced stage of development, regulatory familiarity, and scalability to meet high power demands. These PWR-based SMRs offer a good balance between near-term availability and the capacity to support energy-intensive AI operations. While advanced designs like molten salt reactors or high-temperature gas-cooled reactors could potentially offer higher efficiency and additional benefits in the long term, their longer development timelines make them less suitable for addressing the urgent power needs of rapidly expanding AI operations. Sodium-cooled fast reactors, such as TerraPower's Natrium (345 MWe), present an interesting mid-term option, potentially offering operational flexibility with integrated energy storage, which could be valuable for managing variable AI workloads.
Pressurized Water Reactors (PWR):
Sodium-Cooled Fast Reactors: