Despite data centers nestling into their role as the backbone of modern digital infrastructure, the rapid proliferation of AI is pushing facilities to their limits. Around the world, unprecedented constraints in electrical grid capacity are now actively shaping the trajectory of the industry.

Leaders warn that data centers could soon account for as much as 14 percent of total electricity capacity, driven largely by the explosive energy demands of AI. In this DCD>Talks episode, Dave Bell, VP of data center and microgrid development for VoltaGrid, provides an example that demonstrates the stark scale of this shift:

“Consider the four and a half gigawatt Stargate facility – when you put that into context, it is equivalent to the power consumption of an entire city. In Calgary, Alberta, where I’m from, Stargate’s load would be four times that of the entire City of Calgary.”

A comparison like this underscores the critical reality that the race to build the digital future now hinges not only on compute power, but on the world’s ability to supply the energy required to sustain it.

Energy market pressures

Pressures on the energy market are intensifying as coal-fired plants – traditionally considered the foundation of power generation for decades – retire in large numbers, leaving the grid to navigate a growing set of obstacles. According to Bell, these include the following:

Electrification

This process is accelerating far beyond electric vehicles, reaching into larger-scale production industries that are shifting from traditional coal- or gas-fired systems to electric alternatives to curb greenhouse-gas emissions.

Onshoring manufacturing

The push to onshore manufacturing is driving the return of large industrial facilities, which are demanding substantial new energy supplies while also competing with sectors like mining and digital infrastructure.

This trend also highlights that as more industries draw sustained power, understanding and planning for expendable energy becomes essential to ensure the grid can reliably meet continuous, real-world demand.

“When we talk about the capacity market, utilities tend to focus on total capacity in gigawatts or terawatts. But we need to shift the conversation toward terawatt-hours and gigawatt-hours – the actual expendable energy that determines what the system can truly deliver,” says Bell.

Retirement

Much of the traditional power generation now retiring has provided predictable energy, but is being replaced largely by intermittent resources like wind and solar. Consequently, natural gas is positioned to shoulder the bulk of demand over the next decade, especially as the data center industry prepares for substantial new growth. Bell adds:

“With reliability standards up to five nines, operators simply can’t depend on intermittent power sources and still meet those requirements. Renewables like wind and solar play an important role, but they can’t provide the consistent, firm energy needed at that level. That’s why I see natural gas carrying the load for at least the next five, seven, even ten years. When you consider the 80 gigawatts of growth expected in the data center market, natural gas is the only resource capable of delivering that kind of dependable energy.”

AI power load variability

Compounding these challenges is the volatility of AI power loads. While compute spikes are familiar, the grid has never seen the kinds of rapid, large-scale fluctuations expected with multi-gigawatt AI applications.

Waste not, want not

The wise old saying that reminds us that waste today leads to scarcity tomorrow, ‘waste not, want not,’ is especially relevant as AI data centers drive growing power demands. In this way, ensuring energy sustainability means carefully monitoring the grid’s heat rate to avoid overconsumption.

“The electric grid uses roughly as many British thermal units (BTUs) per kilowatt-hour as it generates; across the US, that figure averages about 8.1 million BTUs per megawatt-hour, excluding line losses,” Bell notes, adding:

“When we take a look at these power plants of the future, how do we make sure that fuel consumption matches the electricity we generate? One simple and effective way is to take transmission line losses out of the equation – locating energy supply closer to users can significantly improve the heat rate across the system.”

Beyond that, Bell exclaims that keeping energy production below 8.1MMBtu per megawatt-hour supports sustainability, while proximity to the energy supply helps lower costs. Since economic output is closely linked to energy consumption, minimizing energy costs also boosts productivity, making efficiency a key driver of both sustainability and economic growth.

How will the energy landscape unfold?

Bell anticipates a gradual evolution rather than a single, dramatic breakthrough. The introduction of nuclear power is expected to continue, alongside a shifting conversation around its role, but the more immediate transformation will come from bringing energy generation closer to demand. He concludes:

“For example, the attainment zones in the US – when we first started this practice, we could only reduce our emissions enough to get around 400MW in an attainment zone. We’ve doubled that capacity and are now able to get almost double that power in the same areas because we found new ways to clean our emissions through our selective catalytic reduction systems.”

Looking ahead, artificial intelligence may offer the next leap forward, helping to optimize energy use and address some of the sector’s most pressing challenges.

Watch the full DCD>Talks episode with Dave Bell of VoltaGrid here.