America’s growing AI power consumption: Strained grids, blackouts, and nuclear energy

AI data centers and the surge in power demand

Data centers have always consumed large amounts of power, but AI raises the stakes dramatically. Training algorithms for large language models like ChatGPT or image generators like Midjourney require robust processing using GPUs and specialized hardware. 

These tasks must be offloaded to cloud-based data centers; your laptop or phone can’t handle the scale. James Walker, CEO of Nano Nuclear Energy Inc., highlights the growing appetite for energy in AI projects. “We’ve spoken to several people building data centers at the moment, and some projections require up to 2GW. I’m sure there are conversations out there asking for much more than that.”

Even a medium-sized AI data center can consume as much electricity as a small city. Technology analyst Jack Gold notes that facilities like Elon Musk’s “XAI” in Tennessee can demand power equal to tens of thousands of homes. The power load could be enormous if dozens or hundreds of such centers come online.

Alex de Vries, founder of DigiEconomist, has monitored digital industries for years, from cryptocurrency mining to cloud computing. “Data centers have accounted for at least 1% of global electricity consumption over the past decade. Due to energy-hungry trends like AI, that percentage is set to climb to 3% or 4% worldwide. They already use more power than the entire country of France.”

Straining the grid: Where does all this power come from?

Modern grids are intricate systems of power plants, substations, and transmission lines distributing electricity across vast regions. Each plant has a maximum capacity for delivery. Adding enormous AI data centers can overload local grids, especially in areas grappling with shortages.

Parts of the United States, such as California and Arizona, have struggled with rolling blackouts during peak demand. Injecting fresh gigawatt-scale data centers into these regions could intensify the challenge. Even in states with surplus capacity—like Texas, North Dakota, and Wyoming—officials must weigh building or upgrading lines, constructing plants, and managing potential impacts on residential power.

According to Alex de Vries, an abrupt increase in demand can trigger higher electricity costs across the board: “If the supply can’t keep up with AI’s demand, the only outcome might be higher power prices. Also, some older energy plants—like coal-based facilities—might have to be reactivated, raising environmental concerns.”

Renewable power sources such as solar and wind have grown significantly. However, their intermittency is a problem for data centers that require a continuous, stable supply. Large battery farms or other energy storage solutions can help smooth out fluctuations but add cost and require substantial land.

The nuclear option: A game-changer or too complex?

One potential answer is nuclear power. It is carbon-free at the point of generation, delivering a constant baseload that can meet data center needs. Yet large-scale nuclear plants come with steep costs, lengthy construction times, and extensive regulatory hurdles. For instance, the UK’s Hinkley Point C faces multi-year delays and billions in budget overruns.

Jack Gold sees promise in smaller, modular nuclear reactors (SMRs:) “Several startup companies aim to build smaller nuke plants quickly. The question is regulation—will the government allow it? It can take a decade just to get permission to build a new plant.”

SMR startups like Last Energy and Rolls-Royce propose “bolt-on” solutions—self-contained nuclear units installed on-site to power data centers. These smaller reactors might also appear on cargo ships or remote locations, fueling industrial systems for decades without refueling. James Walker states that some AI stakeholders have concluded nuclear is the only reliable way to hit energy targets over the next 20 years.

However, these projects remain in the developmental stage. Safety approvals for nuclear power can be stringent, and communities must feel comfortable with a reactor in their backyard, even if modular. In the short term, we’re already seeing cloud giants such as Microsoft strike energy deals with existing or soon-to-be-reactivated nuclear facilities (like Three Mile Island, the site of a historical 1979 nuclear incident in the US).

Water consumption and jobs

Beyond electricity, AI data centers also need vast quantities of water to cool their servers. An average facility might consume over 300,000 gallons (1.2 million liters) daily, comparable to the usage of 100,000 households. Alternative cooling solutions, such as underwater data centers or intricate evaporative systems, come with logistical and environmental challenges.

As for job creation, building these centers can spark short-term booms in construction, but the long-term benefits are less certain. Data centers typically require relatively small, specialized teams to keep them running. Alex de Vries highlights the experience of countries that provided power subsidies to attract data centers, only to be disappointed by minimal employment gains.

The “Stargate” initiative and AI’s future

A prospective new administration policy, the “Stargate” initiative envisions the US leading AI development across diverse fields—defense, transportation, and enterprise. Yet the scale of power required to make this a reality is staggering. Even if states like Texas or Wyoming offer abundant land and existing surplus power, hooking up brand-new gigawatt-scale data centers could take years of infrastructure upgrades.

James Walker underscores potential near-term compromises: “Some data centers might rely on natural gas if nuclear remains out of reach due to regulation. But that’s not always possible in remote areas, and building new pipelines is expensive and time-consuming. Utilities may also refuse to sell that much power to a single consumer.”

Will households and smaller businesses end up paying more or getting less power? Rolling blackouts, rationing, or simply skyrocketing costs could be the reality in certain regions if expansions outpace grid improvements.

The road ahead: Balancing power, progress, and the public

AI is a powerful tool driving innovation in health care, finance, autonomous vehicles, and beyond. Its benefits could be transformative—but at the cost of massive, continuous energy consumption. Over the short term, experts like Alex de Vries see possible price spikes, with power grids straining to accommodate AI’s appetite. Over the longer term, advanced solutions like SMRs, large-scale battery storage, or even re-engineered power grids could bring balance, but these involve time, capital, and political will.

The stark reality is that AI will not slow down. With more advanced models, real-time robotics, and new data-driven services on the horizon, demand for computational muscle—and thus electricity—will only intensify. Whether we build more nuclear plants, expand renewable capacity, or invest in grid resilience, today’s decisions will determine if we can safely, affordably, and sustainably harness the power of AI.

Ultimately, everyday people may not have to “sacrifice” their electricity for AI. However, we might see higher bills, occasional service interruptions, or environmental trade-offs if the necessary grid transformations do not keep pace. The future belongs to those who can successfully integrate AI’s promise with intelligent, robust energy policies—ensuring the lights stay on while tomorrow’s technologies flourish.