Sovereign AI Data Centers: The New Strategic Reserve

As our economic system is increasingly digitalized, the importance of controlling it directly is becoming more and more apparent for many countries. For a while, this has taken the form of pushing for controlling the nation’s most important softwares like payment and banking systems, social media, search engines, etc.

This trend is increasingly important also because, from mostly neutral systems, digital services from major nations like the US can increasingly be used as geopolitical leverage. For example, the SWIFT payment system has been progressively turned into a major tool for sanctioning nations at odds with the USA, such as Russia or Iran.

This question of digital sovereignty is now quickly evolving, as AI becomes the crucial digital infrastructure, overtaking softwares and quickly replacing search engines. And if AI can write progressively written software directly (sometimes called “vibe coding”), having enough AI capability could help a nation develop further its digital sovereignty. So today, the most important question is becoming sovereign AI infrastructure.

In the initial stages of AI development, the key constraint was computing power and the required hardware, with GPUs in short supply and the semiconductor industry struggling to keep up. While these shortages are not entirely resolved, for example, with the supply of memory hardware, this is progressively becoming less of an issue.

However, as larger and larger AI data centers are getting built, a new constraint emerges, one much more difficult to solve quickly: energy supply.

So it is becoming urgent not just for AI data center companies, but most nations to figure out quickly a new source of reliable low-carbon energy.

The Death of the Global Cloud

Global Cloud = Global Risk

Until recently, most of the world has been using American IT infrastructure for its cloud computing needs. With the exception of Chinese firms Alibaba and Huawei, together holding only 6% of the global cloud market, all the global top cloud infrastructure providers in 2025 were American firms.

Source: Statista

This creates a massive vulnerability for most nations, especially as global geopolitical instability ramps up and tensions between the USA and its allies can spike seemingly randomly, as illustrated by the recent spat over the control of Greenland, European nations’ position toward Israel, or the use of European military bases to strike Iran and European warships to escort trade in the Hormuz Strait.

Data centers abroad might also simply not be safe from real-world damages, as illustrated by the series of Oracle, OpenAI, and Oracle data centers going dark in the Middle East after Iranian strikes, either directly on the facilities or on their energy supply.

In this context, sensitive or strategic data should maybe never be hosted abroad, irrespective of a country’s geopolitical position. This is true for national databases like healthcare, banking, or the military. But this is true for the commercial data of large corporations as well.

Sovereign Clouds Rising

As proof of this trend, Canada launched in April 2026 a new “Sovereign AI Compute” initiative. The initiative, part of the overall Canadian Sovereign AI Compute Strategy, aims to create an independent domestic AI compute capacity directly owned by Canadian-only interests.

“These systems will form a core part of Canada’s digital backbone, enabling breakthroughs in areas like health care, energy, advanced manufacturing and scientific discovery. This will strengthen Canada’s global competitiveness, support world-leading research and ensure secure, reliable access to critical digital infrastructure for Canadian innovators.”

The overall Canadian AI strategy is built around 3 pillars:

• Mobilizing private sector investment, so that private funds complement the $700M already invested by the Canadian government.
• Building public supercomputing infrastructure, focused on the Sovereign AI Compute Initiativeand a short-term $200M investment to augment existing public compute infrastructure to address immediate needs.
• The AI Compute Access Fund, a $300M fund providing subsidies to private companies in the form of credit for access to AI computing capacities.

Meanwhile, the EU as a whole has been deploying its “€200B AI Continent Action Plan” since April 2015. It includes building 5 AI gigafactories to train and develop complex AI models, and overall 3x the EU’s data centre capacity in the next 5-7 years

In October 2025, the EU released its “Apply AI Strategy” to boost AI adoption across 10 key industries:

• Healthcare and pharmaceuticals
• Mobility, transport, and automotive robotics
• Manufacturing
• Engineering and construction
• Climate and environment
• Energy
• Agri-food
• Defence
• Security and space
• Electronic communications, cultural, creative, and media sectors.

So while the idea of sovereign cloud and AI compute capacity was mostly the domain of countries like China and Russia a few years ago, 2025 & 2026 are the years when most major economic blocs and developed economies are building sovereign AI capacity separated from the US tech giants and global cloud hyperscalers.

SMRs: The Baseload Solution

AI = Energy

To power all these AI data centers, a steady and reliable source of energy is required. Previously, national power grids were thought to be sufficient for that task.

But a new phenomenon called the Energy-Compute Paradox, or sometimes the AI Energy Paradox, is challenging this assumption. It states that improving the energy efficiency of AI actually leads to a massive increase in total energy consumption rather than a decrease.

At the core of this phenomenon is the AI version of an economic phenomenon called the Jevons Paradox. As the efficiency of AI improves, its cost declines, leading to more demand for AI-driven solutions and overall more energy demand. So the more efficient AI, the more energy is used to power new, growing AI systems.

As a result, AI demand for power is scaling at 10x the rate of national grid upgrades. So a lot more energy supply needs to be added. And even if the extra power is added to the grid, bringing it to the data centers through the grid can be a challenge, as new powerlines and transformers are not built quickly enough.

The Nuclear Option For AI Data Center

Ideally, renewables could be installed in even greater capacity to provide these Tier 4 data centers with carbon-free power.

In practice, the surface required for solar installation, or the location constraints of wind power, means that power grid saturation remains a serious problem. Grid congestion is now the primary bottleneck for data center deployment.

In Texas, CenterPoint Energy reported a 700% increase in large load interconnection requests, growing from 1 GW to 8 GW between late 2023 and late 2024. Utilities like ComEd, PPL, and Oncor are reporting more GWs of data center applications than their historical maximum peak demand.

Camus Energy

In addition, the extremely high requirement of AI data centers for 24/7 uptime and a high-quality power source means that the intermittency of renewables is another problem. The installation of massive battery parks, which are both still expensive and in short supply, is not sufficient to fully solve this issue.

This is why the AI boom is also triggering a renaissance of the nuclear energy sector as well. More specifically, SMRs (Small Modular Reactors), which are quicker to build and can deploy innovative designs.

SMRs are also just in the right power capacity range to supply GW-scale datacenters, with the possibility to fine-tune the exact power capacity by adding or removing modules. This is a major advantage for on-site power generation for industrial purposes, while traditional nuclear power plants are just too big to power anything but supply the grid.

Combined with their low-carbon emissions, these characteristics make SMRs the ideal options for powering AI data centers.

(You can learn more about SMR technology in our associated article “AI’s Energy Crisis Is Fueling a Nuclear SMR Investment Boom”)

The “Compute-GDP” Correlation

The more AI technology progresses, the more it will become the driver for economic growth at the national level. In itself, this is a solid argument for every country to develop its own AI strategy quickly.

The same holds true for domestic AI infrastructures. As billions and even trillions are being poured into data centers and the energy generation to power them, it makes sense that countries will want to keep most of this investment inside their domestic economy.

While the actual result can of course vary, it is in general considered that investment in “sovereign” computing capacity brings a 3:1 economic multiplier; that $1 in investments like data centers and specialized AI hardware can generate $3 in broader economic activity.

One contributing factor is AI adoption in itself, with expected gains in productivity from workers and more efficient systems overall. It also creates many high-qualification and high-income jobs, which can bring extra dynamism to underdeveloped or remote regions.

Another factor is that, contrary to software, AI infrastructure requires a lot of physical inputs and physical labor, both for construction and operation, which flows back into the local and national economy.

Another factor specific to sovereign clouds is that the higher reliability and safety of a national AI infrastructure can speed up the AI adoption rate, incentivize private investment, and reduce risks, all elements leading to increasing competitiveness for the national economy and the companies operating in it.

The Future Of Sovereign AI Infrastructures

So far, the global cloud infrastructure has been dominated by American companies due to a mix of US dominance in software, early mover advantage, and the natural advantages the USA’s massive market and Wall Street’s financial capacities bring to growing companies.

However, as cloud & AI infrastructure keep growing, the critical mass for economies of scale is now reachable by many smaller countries than China or the USA.

In addition, the dawning realization of the importance of digital sovereignty is pushing all major countries to take back control of their IT infrastructure, especially AI capacities.

As such, it is likely that major economic blocs like the EU, but also developed nations like Canada, Japan, South Korea, and others will progressively push for their own sovereign AI infrastructures.

Even relatively less developed countries, like Brazil, for example, are likely to follow suit, as Brazil is already getting fully independent for its payment system with Pix, a very successful national system, so much so that it actually was briefly attacked by President Trump.

“The Office of the U.S. Trade Representative says in a report to the U.S. Congress that PIX has an unfair advantage over U.S.-based companies because it is overseen and operated by the Central Bank of Brazil. The office also says the Brazilian government disadvantages U.S. companies by imposing a network usage fee and refusing to share consumers’ personal data.”

This means that US hyperscalers might at some point see a slight slowdown of their international results, as national options will become increasingly more dangerous competition. However, investors can count on the ever-increasing need for power of AI, be it US-controlled or more decentralized sovereign AI infrastructures.

Investing In Sovereign AI & Energy-Compute Solutions

BWX Technologies, Inc.

Before ambitious startups started working on new designs of SMRs, nuclear reactors used to be giant buildings.

Except for a small niche market that also turned out to be very profitable. Smaller-sized nuclear power for ships like aircraft carriers, submarines, and other military systems.

Source: BWXT

This is the focus of the American company BWXT, which has delivered 400+ reactors for naval nuclear power applications in its entire 60+ year history. It is also active in segments of the nuclear power supply chain, having delivered 315 steam generators for larger nuclear power plants.

A key part of BWXT’s income comes from regular services and maintenance to existing reactors, with a schedule planned until 2053 for various operating nuclear aircraft carriers and submarines.

Source: BWXT

The company is the only one accredited in the USA to produce high-quality uranium at 20%+. This fuel type is needed for so-called micro-reactors, even smaller than SMRs. They can power many applications, like space systems for NASA (essential for the return to the Moon as part of the Artemis missions) and remote military locations.

BWXT is also entering the nuclear medicine field, hoping to capture some of the sector’s $500M annual revenues.

Finally, BWXT is also working on an SMR design, in partnership with GE Vernova (GEV ) and Hitachi (6501.T). GE is one of the main contenders in the nascent global SMR market. Among BWXT and GE Vernova’s successes in the past few years can be mentioned:

• Canada’s Darlington New Nuclear Project,a 1,200 MW project deploying four SMRs by the mid-2030s.
• Already agreed contracts with Estonia for a 600 MW capacityto be built in the early 2030s.
• Partnership in Poland with Orlen Synthos Green Energy (OSGE) for potentially as many as 24 SMRs. Orlen Synthos Green Energy (OSGE) is actively positioning the BWRX-300 as a solution for the data center industry.
  • In mid-2025, OSGE and the Polish Data Center Association established a joint working group to integrate SMRs directly with data centersand also provide district heating for cooling systems.
• A Tennessee Valley Authority-led coalition for accelerating the deployment of the BWRX-300 small modular reactor in the U.S.
  • The US Energy Department is specifically mentioning helping power data centers as a goal of the association.
• Potentially up to 6 SMRs in Bulgaria, 3 SMRs in Norway, and in Sweden and Finland(in partnership with Fortum – HE).

So overall, the role of pioneer of BWXT in miniaturizing nuclear reactors means the company should benefit greatly from the trend of SMRs picking up. It should also benefit from the military build-up of the US Navy in reaction to increasing challenges from Russia and China.

Finally, it is profitable and has predictable revenues, offering a much greater financial safety and cumulative experience than more speculative companies with still only designs to show, and no installed reactors yet.

Latest BWX Technologies (BWXT) News and Developments