The nuclear industry is splitting into two competing visions for powering artificial intelligence infrastructure. While most companies chase smaller, transportable reactors that can be built in factories and trucked to sites, Russia is pouring 16,000 metric tons of steel and concrete into a permanent installation in Siberia, and US startups are drilling a mile underground to test radical new cooling concepts. These parallel approaches reveal fundamentally different bets on how to solve the energy crisis created by AI's explosive power demands.

Why Is Russia Building a Reactor That Can Never Leave Its Site?

In Seversk, a closed city near Tomsk in western Siberia, Russia's state nuclear company Rosatom is assembling the BREST-OD-300, a 300-megawatt reactor so large it cannot be delivered in one piece. The central cavity shell alone weighs 143 metric tons and stands over 45 feet tall. Instead of being shipped as a complete unit like conventional reactors, the machine arrives in six massive sections, each weighing more than 1,000 metric tons, manufactured at plants in Volgodonsk and near St. Petersburg. Workers then assemble these pieces on site, welding steel cavities and pouring concrete into the gaps as the structure climbs.

The reactor uses molten lead as its coolant instead of water, operating at around 500 degrees Celsius at close to atmospheric pressure. Rosatom calls this design "integral" because the vessel is not a simple metal can but a hybrid metal-concrete structure. The entire installation, including transport packaging, will weigh approximately 16,000 metric tons once complete. To move the peripheral cavity shells through the region, crews had to lift power lines and pull down road signs so the convoy could squeeze through.

The vessel assembly is scheduled for completion by the end of 2026, but the reactor itself will not start up until 2028 or 2029. This represents a significant delay from Rosatom's original 2026 target. The project is part of a larger Pilot Demonstration Energy Complex that combines three functions on one site: a fuel manufacturing plant, the reactor itself, and a reprocessing facility to recycle spent fuel. The fuel plant already began operations at the end of 2024 with four production lines.

What Makes Molten Lead Cooling Different From Water?

The choice of molten lead as a coolant offers safety advantages but introduces engineering challenges. Lead boils at extremely high temperatures, meaning losing the coolant is genuinely difficult to achieve. This eliminates the need for a core-catcher melt trap that conventional water-cooled reactors require. Rosatom emphasizes "natural safety" as a core selling point of the design.

However, hot lead dissolves ordinary steel slowly, eating away at the very components meant to contain it. Engineers had to develop special steels rated to survive up to 600 degrees Celsius in this corrosive environment. Nuclear Engineering International identifies liquid-metal corrosion as the main threat to the entire project. The concept is not entirely new; the Soviet Union built lead-bismuth-cooled reactors in the mid-1970s to power seven submarines, so the technology has decades of operational history. Pure lead, however, is more challenging than the lead-bismuth alloy used in those submarines because pure lead does not become liquid until above 300 degrees Celsius, requiring the plumbing to stay hot enough to prevent the coolant from freezing solid inside the machine.

How Are US Startups Taking a Different Approach?

While Russia builds massive and immobile, American companies are pursuing the opposite strategy. Deep Fission, a California-based advanced nuclear energy company, has delivered a prototype reactor canister for its underground small modular reactor system to a test site in Parsons, Kansas. The company's design places a pressurized water reactor inside a borehole approximately one mile underground, using the surrounding water column to help maintain operating pressure and provide cooling.

The prototype has completed fabrication, hydrostatic testing, and delivery, allowing the company to move into non-nuclear testing phases before any fuel is introduced. The Proof-of-Concept Well program is designed to test installation methods, infrastructure readiness, and operating procedures at nearly full scale. Unlike conventional reactors that require massive containment structures and cooling towers, Deep Fission's approach simplifies construction by leveraging the natural pressure and cooling properties of the surrounding geology.

"The arrival of our prototype reactor canister at the Kansas site is a clear step forward in moving from design to deployed infrastructure," said Mark Pérès, Chief Nuclear Officer of Deep Fission. "Successfully manufacturing, testing, and delivering this hardware demonstrates performance of our design and supply chain capabilities."

Mark Pérès, Chief Nuclear Officer at Deep Fission

Heat generated by Deep Fission's Gravity Nuclear Reactor is transferred through a closed-loop system to a heat exchanger before traveling back to the surface through a secondary loop, where it can be converted into electricity using equipment similar to conventional geothermal power plants. This design approach reduces the need for traditional reactor vessels and containment structures while maintaining the safety and efficiency of proven pressurized water reactor technology.

How Are Energy Storage Systems Bridging Nuclear and AI Infrastructure?

Beyond reactor design, companies are developing advanced energy storage systems specifically engineered to handle the unpredictable power demands of artificial intelligence data centers. Aegis Critical Energy Defence Corp. announced a strategic research initiative with McMaster University to develop the High C-Rate Fast-Transient Energy Storage System (HCFT-ESS), a platform combining premium European automotive battery technology with next-generation battery management systems, advanced thermal management, and artificial intelligence.

Unlike conventional battery energy storage systems designed primarily for long-duration energy shifting, the HCFT-ESS platform is engineered to deliver extremely fast power response, exceptional thermal stability, and intelligent energy management. The company is pursuing a phased commercialization roadmap targeting three rapidly growing sectors.

• AI Data Centers (Target: Within 12 Months): Development of a high-power energy storage system capable of supporting rapidly fluctuating AI computing loads with ultra-fast transient response, improving power quality, resilience, and operational efficiency for next-generation AI and hyperscale data centers.
• Port Electrification and Heavy Industrial Applications (Target: 12 to 24 Months): Expansion of the HCFT-ESS platform into high-power applications including port electrification, ship-to-shore operations, container cranes, regenerative energy recovery systems, and heavy industrial facilities requiring rapid power delivery and energy optimization.
• Advanced Hybrid Energy Systems (Long-term): Expansion of the platform into hybrid energy systems integrating small modular reactors, micro modular reactors, renewable generation, and advanced battery energy storage for remote communities, defense installations, utilities, and mission-critical infrastructure.

The research program, valued at approximately $3.71 million over four years, is being conducted through McMaster University's Centre for Mechatronics and Hybrid Technologies, which is internationally recognized for advanced battery pack development, battery management systems, and thermal modeling. The facility provides access to one of North America's most advanced battery development and testing environments.

"This initiative is about much more than research. It's about building the next generation of intelligent energy infrastructure," said Ramtin Rasoulinezhad, Chief Executive Officer of Aegis Critical Energy Defence Corp. "By combining proven premium European automotive battery technology with proprietary innovation in battery management, thermal management, intelligent controls, and system integration, we are creating a scalable technology platform designed to support multiple commercial products across AI infrastructure, ports, and advanced hybrid energy systems."

Ramtin Rasoulinezhad, Chief Executive Officer at Aegis Critical Energy Defence Corp.

Steps to Understanding Nuclear's Role in AI Infrastructure

• Recognize the Scale Difference: Russia's BREST reactor represents a permanent, large-scale installation approach, while US companies like Deep Fission pursue modular, deployable designs that can be installed at multiple sites without massive construction projects.
• Understand Cooling Trade-offs: Molten lead cooling offers safety advantages but requires specialized materials to prevent corrosion, whereas underground placement leverages natural geology for pressure and cooling, each solving different engineering constraints.
• Consider Energy Storage Integration: Advanced battery systems like Aegis's HCFT-ESS are not replacements for nuclear power but complementary technologies that smooth out the rapid power fluctuations created by AI workloads, making nuclear plants more effective at supporting data centers.
• Track Timeline Expectations: Russia's BREST project has already slipped from a 2026 startup date to 2028 or 2029, illustrating that first-of-a-kind nuclear projects routinely experience delays, while US demonstration projects are still in prototype and testing phases.

The divergence between Russia's massive permanent reactor and America's underground and compact approaches reflects different assumptions about nuclear's future in the AI era. Russia is betting on large, efficient plants that serve regional grids for decades. The US is betting on flexibility, modularity, and the ability to deploy reactors closer to where AI companies actually need power. Both strategies acknowledge the same underlying reality: artificial intelligence's energy demands are forcing a reckoning with nuclear power, and the industry is experimenting with fundamentally different designs to meet that challenge.