Originally published on Utility Dive on November 11, 2024
The thirst for electricity from the growing number of data centers needed for artificial intelligence (AI) and cloud computing is unprecedented. For example, a recent analysis from Bain & Company concluded that by 2028, utilities will need to increase annual electricity generation by 7% - 26% above their 2023 totals. By comparison, the largest ever annual generation increase over a five-year stretch by U.S. utilities was 5%. Bain also found that data centers would account for 44% of all load growth between 2023 and 2028 (compared to 27% for the residential sector and 17% from manufacturing).
Electricity demand from data centers is growing so vigorously that even eye-popping forecasts of its impact on peak energy demand quickly go out of date. The consulting firm Grid Strategies analyzed data submitted to the Federal Energy Regulatory Commission (FERC) by grid planning organizations. In 2022, grid planners forecast peak demand in 2028 would be 835 GW. Just one year later, the forecast rose to 852 GW. In Virginia, which is aggressively courting new data centers, demand is expected to grow from 22 GW in 2024 to 42 GW by 2039 in the portion of the grid operated by Dominion Energy.
Data center owners are scrambling to find places to locate their facilities.
Microsoft and Constellation Energy, for example, recently announced plans to reopen one of the nuclear reactors at Three Mile Island, which closed five years ago, to help power data centers. Constellation said it will invest $1.6 billion to upgrade the facility and will seek approval to operate it until at least 2054.
One of the appeals of nuclear power for data centers is that it provides ample and reliable supplies of carbon-free electricity. This is important to companies such as Microsoft, which has pledged to reduce its Scope 1 and 2 emissions to near zero by 2030. But a significant portion of the electricity to run data centers will have to come from renewables, which makes siting and operations even more challenging.
“Renewables can introduce lots of problems for data centers when it comes to reliability because they’re very intermittent,” said Tarek Ibrahim, head of advanced analytics for Energy Exemplar, a company that provides software to model electric, gas, and water markets to enable better utility and energy system decision making. “Without adequate storage backup systems, it’s going to be very problematic to maintain a constant power supply.”
A host of other factors complicate data center siting even more. These include interconnection timelines, transmission grid capacity, access to water for cooling, a labor pool with sufficient skills as well as wholesale energy market rules and existing and potential state government incentives.
Traditional zonal planning is not well-suited to siting data centers. In zonal planning, assessments of demand, resource availability, grid capacity and other factors are applied to relatively large geographic zones.
By contrast, nodal planning is far more granular and precise. “Nodal modeling is different because it’s more detailed,” Ibrahim said. “Each point of interconnection on the grid is considered a node, and that can be a substation, a point of generation or a load like a data center.”
The detailed, point-of-interconnection view possible with nodal planning provides a depth of technical and market information that is especially helpful for data center siting. For example, nodal planning considers localized transmission grid congestion and voltage stability at specific nodes, which is important in understanding whether the existing grid can reliably provide the supply of electricity data centers need or if it will require upgrades and additional energy storage to avoid disruptions. Individual nodes, even if they are geographically proximate to one another, can vary considerably in grid capacity.
Importantly, nodal planning also provides specific insights about pricing at potential data center locations. “Within regions, nodes will have different prices that vary based on congestion of the local transmission lines and power losses and other factors,” Ibrahim said. “For data centers, it’s very important to know what these locational marginal prices (LMPs) are because you need visibility into how much you would have to pay to get the power needed to feed that data center.”
Energy Exemplar’s PLEXOS® modeling software enables nodal planning that data center developers can use to compare and assess different locations. Because PLEXOS® granularly models how energy flows through specific nodes on the grid, it provides users a view of grid congestion and pricing under a wide variety of scenarios.
Recently, PLEXOS® added a new feature designed specifically to help those seeking to optimally locate a data center. Called available transfer capacity, or ATC, the feature studies the nodes on the grid based on specific commitment and dispatch schedules. “You can feed it multiple schedules that could happen throughout the year, and it studies each one from a generation capacity hosting or load capacity hosting standpoint, meaning how much energy could be injected or withdrawn at a node before there would be base case or contingency constraints binding,” Ibrahim said. “This provides insights about the risk of load shedding or renewable resource curtailment at specific times and under different scenarios.” PLEXOS ATC can also provide the user with a time series graph, showing the variability in hosting capacity over time and providing insight into the system beyond a snapshot in time.
There is little doubt that reliably providing data centers with enough carbon-free electricity is a challenge. But for data center owners, it’s also important to know that sufficient grid capacity does exist today if you know exactly where to site a facility. PLEXOS® can simulate real-world grid constraints at the nodal level to help data center owners find the reliable, cost-effective power supply they need.