In Brief:
- AI companies are growing at a rapid pace — and continued growth depends on increasing energy supply.
- Data centers are sprouting up all over the country, sometimes facing community backlash thanks to their water usage and the strain they threaten to put on the energy grid.
- A report from an energy think tank makes the case for a broader perspective on how to meet data center energy needs.
Hundreds of billions of dollars are pouring into artificial intelligence, driving the U.S. economy. The profits that technology leadership investors seek depend on meeting the electricity needs of data centers, warehouse-sized facilities filled with servers that power the digital world, including AI.
By the end of 2025, U.S. data centers are projected to need 22 percent more grid power than they did the year before and almost three times more soon after. S&P Global puts the 2030 need at over 134 GW (gigawatts). One GW is enough to meet the average annual energy needs of about 700,000 homes.
Eric Gimon, a senior fellow at the nonpartisan think tank Energy Innovation, hopes to provide legislators and state officials working to attract data centers with a more nuanced picture of what they can do to fill gaps between energy demand and supply.
“It’s the normal human thing when we’re dealing with complexity to cling to the simplest ideas,” he says. “People have devolved to the thinking that the way to deal with a big new load is a big new generator.”
Through a new report, Gimon hopes to move beyond what he calls “firm fixation.” He makes the case that overemphasis on “firm” power — electricity from sources that are always “on,” such as nuclear reactors or gas-fired power plants — leads to missed opportunities.
It can cause utilities and developers to overlook solutions like energy efficiency, or batteries connected to solar and wind generators, that cost less, can be implemented more quickly and reduce risk from overinvestment in local fixed capacity vulnerable to shifts in demand, he argues.
Negotiating Power
It’s well known that data centers can have serious environmental impacts and require massive amounts of water to cool the servers. Communities have increasingly pushed back on their construction, with complaints about noise and the possibility of increased energy costs. States and localities have the best chance of making deals to address these issues when a data center is first built, Gimon says — when it’s getting permits, authorization to connect to the grid, and when the utility is deciding how much to charge.
This is the time to ensure data center developers make fair commitments to help cover the infrastructure costs that come with their energy needs, and to push them toward solutions that make the most of existing resources and support the energy grid, Gimon argues. These could include energy efficiency requirements, using data center waste heat to generate electricity or flexibility regarding peak use times.
Gimon also points to a report from Rewiring America arguing that household energy efficiency and rooftop solar could potentially free up enough existing grid energy to meet the needs of the biggest data centers. Governments may be able to secure agreements from companies to fund efficiency projects for the rest of the grid to help meet their needs.
They may be better able to get these kinds of commitments from large technology companies like Amazon Web Services or OpenAI, which are building huge data centers to deliver their services. These companies may have a vested interest in gaining community buy-in and have a greater ability to make long-term energy use projections, Gimon argues.
It may be harder to secure these kinds of promises from “co-location” facilities — large warehouses owned by developers and rented out to multiple separate entities. These developers may be hesitant to make any deals that might affect their ability to draw tenants or the long-term value of their properties, Gimon says.
(Anthony Souffle/TNS)
Drawing From Existing Resources
It’s common for conversations about data centers and energy use to focus on giant power sources like gas-fired plants or nuclear energy facilities. And certainly they will have to make use of those existing energy sources.
But Gimon argues that states and localities shouldn’t be afraid of a varied approach to meeting energy demands, leaning more heavily on renewables like wind and solar. The usual criticism of these sources is that they don’t provide a constant enough supply to power giant data centers.
That's not the whole picture, Gimon says. Energy demand from AI data centers may be enormous, but they don’t operate at peak demand at all times. “Depending on what they are doing, they could be running their chips 70 percent of the time, 50 percent of the time or 40 percent of the time,” says Gimon. “They also turn off and on a lot.”
They also tend to be built in phases. “That’s part of what makes it difficult to serve them with one big power plant,” says Gimon. “You can be undersized or oversized depending on where you are in phase in or phase out.”
Using batteries can assuage concerns about the intermittent nature of renewables. According to the U.S. Energy Information Agency, wind and solar account for more than two-thirds of utility-scale electric-generating capacity additions planned for 2025.
If governments can secure agreements with data centers to avoid peak energy use during periods of high demand for other grid users (like the afternoon of a summer day, for example), it can create more flexibility in choosing energy sources that can meet need.
“By exploring the nuanced solutions, policymakers can avoid overcommitting to outdated firm resources and instead adopt strategies that embrace modularity, flexibility, and clean energy,” Gimon writes. This can require a deeper dive in the complexity of energy systems by those in government who have responsibility for regulation or legislation.
The Tip of the Spear
Over the last 20 years, energy use decreased from 60 percent of what was available to users to 40 percent, Gimon says. The emergence of an energy-intensive industry, and cross-sector conviction that the nation’s fate depends on growing it as fast as possible, is a head-spinning reversal of this trend.
Data centers are just the tip of the spear, Gimon says. The dawning era of building electrification and the growth of electric vehicles, including heavy duty trucks, mean load growth will continue even if the data center craze crashes and burns.
“We’re in a new era of demand growth that we haven’t seen in a long time in the U.S.,” Gimon says. “It’s going to require an all-of-government approach to manage that, to manage new facilities, manage new generation and find ways to do a better job of using the infrastructure that we have.”
