AI’s growing appetite for electricity is forcing a nuclear reckoning, but while fission offers a potential path forward, fusion remains a promise for another decade.

The race to build increasingly powerful data centers has sparked growing concerns about whether America’s electrical grid can accommodate the demands of artificial intelligence. Critics warn that the expansion of AI infrastructure could overwhelm an already strained power system, prompting calls for politicians to impose moratoriums on new data center construction.

But restricting development is hardly a solution to a problem that demands innovation. Rather than slowing the expansion of data centers needed to support AI and its emerging applications, we should explore alternatives to the traditional model of connecting every new facility directly to the electrical grid.

For decades, many utility companies have been slow to modernize aging infrastructure, improve resilience, or adequately prepare for increasing electricity demand. Accustomed to a regulated industry with relatively predictable growth, too many utilities continue to operate with a 20th-century mindset.

That approach is increasingly unsustainable. Electricity demand is not merely growing; expectations are changing. Businesses require greater reliability, stronger redundancy, and infrastructure capable of supporting technologies that operate continuously and cannot tolerate prolonged interruptions.

Modernizing the nation’s electrical infrastructure is essential. But so is recognizing that the grid of the future may need to function differently from the centralized system built over the previous century.

Two nuclear technologies are attracting particular attention: fission and fusion. Although both produce energy through nuclear reactions, they occupy fundamentally different stages of technological and commercial development.

Nuclear fission, the process used in conventional nuclear power plants, is a proven technology. Its smaller-scale evolution, the small modular reactor (SMR), offers the prospect of providing dedicated electricity to energy-intensive facilities, including next-generation AI data centers.

Nuclear fusion, by contrast, remains a technology with considerable promise but no commercially operating power plants supplying electricity to the grid.

For anyone planning a mission-critical facility today, that distinction matters enormously.

A large data center cannot be designed around the assumption that a promising technology will become commercially available within three, five, or even ten years. The same principle applies to stock exchanges, emergency communications centers, and other facilities where reliability is an operational necessity.

The challenge is selecting technologies that can realistically be delivered, licensed, and operated within a project’s development schedule. SMRs are an increasingly credible option for future projects, although many designs remain in regulatory review or early deployment rather than being commercially available off the shelf.

This is not merely a theoretical concern. More than three decades ago, when we were developing Chicago’s 911 Emergency Call Center, several promising technologies were supposedly just around the corner. Some appeared capable of substantially improving the facility’s performance.

But we could not design an emergency communications center around equipment that might become available years later. Instead, we selected the most advanced proven technologies available while incorporating a modular design that would allow newer systems to be introduced once they demonstrated their reliability.

That philosophy remains just as relevant today.

A mission-critical facility must be designed around what can realistically be installed, tested, and maintained when it becomes operational. Future technologies should be accommodated through adaptable design, not treated as guaranteed solutions before demonstrating commercial viability.

Several prominent SMR and microreactor designs are undergoing regulatory reviews, pursuing licenses, or participating in pre-application discussions with the U.S. Nuclear Regulatory Commission (NRC).

Recent regulatory reforms could accelerate their development. The NRC’s new 10 CFR Part 53 framework is intended to provide a more flexible licensing pathway for advanced nuclear technologies than the traditional Part 50 and Part 52 approaches.

Rather than relying primarily on prescriptive requirements developed around conventional reactors, Part 53 emphasizes safety performance and risk-informed evaluations. This allows developers greater flexibility in demonstrating how different reactor technologies can satisfy regulatory standards.

The distinction matters for emerging designs such as helium-cooled microreactors and sodium-cooled systems, whose engineering differs substantially from traditional light-water reactors.

Nevertheless, regulatory reform does not automatically translate into immediate commercial availability. Manufacturers must still secure financing, complete licensing requirements, establish supply chains, and demonstrate that their reactors can operate safely and economically.

Several companies illustrate both the promise and the challenges.

NuScale Power is among the most advanced U.S. developers of light-water SMR technology. Its original 50-megawatt-electric design received NRC certification in 2023, while an uprated 77-megawatt-electric version received standard design approval in May 2025.

The newer design allows multiple modules to be combined into larger generating facilities, potentially making the technology suitable for customers requiring substantial and continuous electricity supplies. However, regulatory design approval does not mean a completed commercial facility is ready for immediate deployment.

Another prominent developer, X-energy, is pursuing its Xe-100 reactor, an 80-megawatt-electric, high-temperature gas-cooled design. Working with Dow, the company is advancing a proposed nuclear facility in Seadrift, Texas.

In May 2026, the NRC completed an environmental assessment of the project, representing an important regulatory milestone. The Xe-100 could eventually provide both electricity and industrial heat, expanding its usefulness beyond conventional power generation.

Oklo is pursuing another approach with its Aurora Powerhouse, a liquid-metal-cooled fast reactor designed for smaller-scale nuclear generation. Following earlier licensing setbacks, the company has continued working through NRC pre-application activities.

Together, these projects suggest that advanced fission could eventually provide data centers with dedicated electricity generation, reducing dependence on conventional utility infrastructure while strengthening overall resilience.

Fusion presents a different challenge.

Unlike fission, which generates energy by splitting heavy atomic nuclei, fusion seeks to release energy by combining lighter nuclei. It is the process that powers the sun and has long been regarded as a potential source of abundant electricity.

The scientific promise is considerable. But the distance between experimental success and a dependable commercial power plant remains substantial.

There are currently no commercial fusion reactors supplying electricity to the grid. By comparison, conventional fission has decades of operating experience, while smaller reactor projects are already operating in countries such as China and Russia.

Before fusion can be considered a serious alternative for powering large data centers, developers must demonstrate considerably more than successful laboratory experiments. They must prove their systems can deliver electricity consistently, withstand demanding operating conditions, and meet the reliability requirements of critical infrastructure.

Then there is the question of cost.

Without a commercially operating fusion plant, it remains difficult to determine whether the technology can generate electricity at a competitive price. Estimates of future construction, maintenance, and operating expenses remain speculative.

One company attempting to overcome these obstacles is Munich-based Proxima Fusion, a spin-off from the Max Planck Institute for Plasma Physics.

Proxima is developing a fusion reactor based on the stellarator, a device that uses complex magnetic fields to confine extremely hot plasma. Its objective is to translate decades of scientific research into a commercially viable power system.

The company is pursuing an ambitious development schedule, including a demonstration reactor near Munich and a subsequent commercial project intended to generate electricity during the 2030s.

Those ambitions deserve attention, but they remain projections rather than demonstrated commercial capabilities.

For operators making infrastructure decisions today, fusion cannot yet be treated as a dependable alternative to existing electricity supplies or emerging fission technologies.

That does not diminish its longer-term potential. Fusion could eventually become an important part of the global energy mix, particularly if developers demonstrate reliable operation and competitive generating costs.

The rise of artificial intelligence and robotics will place enormous demands on electricity infrastructure throughout the 21st century. Meeting those demands will require more than constructing additional power plants or expanding existing transmission networks.

It will require a more adaptable approach incorporating distributed generation, redundant systems, advanced nuclear technologies, and continued modernization of the electrical grid.

Fission-based SMRs may become an important part of that strategy as commercial deployment advances. Fusion could eventually offer another option, provided its developers overcome the remaining engineering and economic challenges.

The objective should not be to replace the electrical grid but to strengthen it through greater reliability, resilience, and redundancy.

For the AI revolution to reach its potential, the infrastructure powering it must be just as innovative as the technologies it supports.