It's no mystery that data centers require access to copious, reliable power. Yet surprisingly, data center operators frequently underestimate the amount of time it takes to get grid power in place, sometimes by years.

The result: The savviest operators are now making power access a primary determinant of where to locate their next data centers. And in areas where grid power is not quickly available, data center developers are looking elsewhere, turning to on-site power alternatives.

In our latest research, Bloom Energy's Mid-Year Data Center Power Report, 84 percent of data center decision-makers reported that power access is a top-three consideration in data center site selection.

Decision makers named it a top factor far more than they cited any other criteria, including proximity to customers, proximity to fiber-optic cables, favorable local regulations, or the availability of labor. 

This shift is already reshaping the energy landscape

In Virginia, 25 percent of electricity already goes to data centers. By 2028, data centers may account for as much as 12 percent of nationwide electricity demand, a massive increase over the 4 percent that they represented in 2023.

When we look at the distribution of data centers in the US, we see that some states, such as Virginia, Texas, and California, have far more data centers than others. Virginia, for instance, with 578 data centers, leads the US with 25 percent of the country’s data center load. This is undoubtedly a reflection of the state’s ability to deliver power to data centers quickly.

For developers, power availability is no longer a consideration; it is a strategic constraint and a business-critical lever. Time-to-power translates directly into revenue and market growth.

Time-to-power gaps

What data center developers are likely to discover is that it takes far longer to get power delivered to a new site than they are currently planning. In many key markets, the report shows that there's a one-to-two-year gap between when developers expect power and when the grid will actually be available.

Data center operators generally expect power to be available in 2027, two years from now. But in Atlanta, the Bay Area, Dallas-Forth Worth, and Phoenix, it may take until 2028 to deliver the power to a new data center. In Chicago or Ohio, it may take until 2028 or 2029. In Austin/San Antonio, power for a new data center might not be available until 2029 or 2030.

AI-driven demand

There's no question what's driving this increasing demand for compute power: Artificial intelligence. Now a dominant technology in virtually every industry, AI brings with it a demand for large, more advanced data centers filled with power-intensive chips, such as NVIDIA's Blackwell GPUs.

Rack after rack of GPU-filled AI servers add up. The data center leaders we surveyed expect the median data center power capacity to reach 250-300MW by 2030, and 350-400MW by 2035. Indeed, some data center campuses will require more than a gigawatt of power, enough to supply over 800,000 average American homes.

On top of that, hyperscalers are already making plans for next-generation architectures that will enable these massive compute farms with higher power densities.

To make such densities possible, they are planning to distribute electricity throughout the data center using direct current (DC) instead of alternating current (AC).

About half of the executives at hyperscalers and colocation facility developers we surveyed say their organizations are considering a transition to DC power distribution for at least some of their locations within four years.

Turning to on-site generation

To meet the need for hundreds of megawatts of quickly available, reliable power, and to efficiently support the next generation of DC data centers, many developers are turning to on-site generation for primary power.

The report shows that 38 percent of data centers will likely have at least some on-site power generation by 2030, rising to nearly 50 percent by 2035. Even more striking, 27 percent will be relying entirely on on-site generation for primary power by 2030. A year ago, that figure was just 1 percent.

Think about that: Last year, just one data center in 100 expected to rely entirely on power generated on-site; five years from now, we could see that become more than one in four.

Taking control of the timeline

Simply put, the lengthy timelines associated with the grid can't keep pace with the fast-moving demands of the industry, so the industry is taking control with on-site generation.

Operators are looking beyond basic generation and transmission to solutions that offer fast deployment, flexible load following, and low local emissions, all while meeting the industry’s uncompromising reliability standards and economic requirements.

When you control your power, you control your timeline. What separates viable projects from stalled ones is the rapid availability of reliable energy. And for a growing number of data center developers, that means on-site power.

Want to learn how data centers are talking control of their power? Download the Bloom Energy Mid-Year Data Center Power report here.