Have you heard? Demand for new data centers is skyrocketing

Data center demand for electricity is projected to triple by 2028 to more than 12% of overall U.S. power demand. This growth is straining the grid, delaying the build out of new manufacturing facilities, and passing new costs onto customers already struggling with rising electricity rates.

One of the big projected cost drivers, however, is based on the false assumption that data centers will always need a constant, unwavering 24/7 power supply. If we keep building out the grid under this extreme assumption, we will end up with massive amounts of expensive infrastructure sitting idle most of the time.

Thankfully, there is a better, more affordable approach.

Flexibility as a scalable solution

Data centers that operate flexibly — temporarily reducing their power use during peak demand — could be a game-changer, even if they only do this a handful of times per year. Instead of building out gigawatts of new, likely fossil-based generation to serve those few high-demand hours, what if we encouraged data centers to take their foot off the gas when the grid is most stressed?

A now widely cited study from Duke University suggests that if data centers did exactly that — limiting their grid-facing power use during just 0.25% to 1% of annual hours (i.e. 20 to 90 hours a year), the grid could accommodate an additional 76 to 126 GW of load. That's equivalent to effectively doubling today's commercial nuclear fleet, which provides nearly 19% of the nation's electricity production.

This could alleviate pressure on generation and transmission buildouts, improve grid reliability and affordability, and help data centers get connected to the grid faster by reducing the infrastructure upgrades utilities need to make.

The technology toolkit for data center flexibility

A variety of technology approaches to implement flexible loads are gaining momentum.

Emerald AI, which recently announced a $24.5 million seed round, aims to adjust data center energy consumption by orchestrating AI workloads in real-time. They are part of EPRI's DCFlex Initiative, which seeks to demonstrate how data centers can adapt their consumption to better align with grid needs and renewable energy availability.

Companies like FlexGen are using batteries and software to bring data centers online faster by enabling peak load shaving without compromising on critical technical requirements like power quality.

Others like Verrus are developing data centers that are purpose-built to be flexible grid assets — enabling up to 100% load curtailment within one minute of receiving a signal from the utility.

Google recently announced agreements with Indiana Michigan Power and Tennessee Valley Authority to participate in demand response programs.

But the range of technology solutions is both an opportunity and a challenge: it creates multiple pathways to bring data centers online faster, but each utility might evaluate solutions differently, which can lead to market confusion and additional delays.

Flexibility is neither a priority nor a requirement...yet

When a data center asks for a massive amount of power that might stress the grid, utilities often require expensive infrastructure upgrades before approving the connection, which can take years. If data centers were flexible during peak times, however, utilities could approve connections faster with fewer costly upgrades.

We're starting to see policies take shape to enable greater flexibility, such as Senate Bill 6 in Texas, which allows utilities to disconnect loads greater than 75 MW during emergency situations and creates a voluntary program to competitively procure demand reductions in anticipation of grid stress.

Grid operators are also exploring new approaches. The Southwest Power Pool’s High Impact Large Load Interconnection program connects data centers in as little as 90 days if they agree to cut power when the grid is stressed, while PJM recently launched the Critical Issue Fast Path stakeholder process to develop new market rules for flexible large loads.

Nationally, FERC is examining whether data centers can build dedicated power plants on their property — a decision that could affect similar projects across the country.

These policies represent positive progress, but it will take time to see how the market reacts and what participation in these programs actually looks like.

Making flexibility work

Unlocking data center flexibility at scale will require policies that make flexible operations both attractive and standardized. The solution set might include:

• Financial rewards for reducing peak demand: Just as airlines offer cheaper red-eye flights to spread out airport traffic, utilities could offer significantly lower electricity rates during off-peak hours. Data centers that shift their workloads away from hot summer afternoons, such as when air conditioners strain the grid, would see substantial savings on their power bills.

• Standard rules for flexible operations: Clear guidelines that work the same way across different utilities and states, so data center developers know what's expected and how flexibility will be measured.

• Clear communication during peak times: Agreed-upon ways for utilities and data centers to coordinate when the grid is stressed, including when to reduce power, how to communicate and who's accountable for what.

The stakes are clear. Regions that successfully integrate flexible data centers into their grid operations will maintain reliability while supporting AI's growth. Those that don't face an uncomfortable trade-off: turn away data centers and their economic benefits, or risk grid instability from inflexible demand. The technical solutions exist today. The question is whether we'll implement them before the grid reaches its breaking point.