Dive Brief:
- California-based startup GridCARE released an analysis of large load flexibility earlier this month that it says should ease concerns that large-scale data centers threaten grid reliability while increasing electricity costs for other customers.
- The company, which emerged from the Stanford Sustainability Accelerator program, is partnering with utilities to deploy artificial intelligence in load management. It modeled a 1-GW data center deployment in a mid-sized utility territory. It found that if the data center incorporates “just modest amounts of flexibility” into its operations by reducing its load or running onsite power resources during periods of high demand, it could reduce costs by 5% across all customer classes or unlock over $1.35 billion in capital for the utility.
- “We hear lots of questions and concerns about how data centers are creating a burden on the grid,” GridCARE CEO Amit Narayan said in an interview. “What we wanted to show is [that] if you … use data centers to drive more utilization of [grid] assets, you can actually reduce rates or create headroom to accelerate investment.”
Dive Insight:
The grid technology company, which helps utilities free up load interconnection capacity, assumed a 1-GW data center would generate about $142 million in incremental annual earnings for a midsize utility. If the utility put that cash toward rate relief, it could decrease rates by 5% across all customer classes, or save about $103 per year for the average residential customer, GridCARE said.
“Maybe this is a shift of narrative…if a data center comes to your territory and it’s strategically interconnected, as [a] customer, you’d be excited because it’s lowering your rates,” Matt Witkin, GridCARE director of strategy and operations, said in an interview.
Alternatively, assuming a typical capital charge rate of 10.5%, the cash could fund about $1.35 billion in capital investments without raising customer rates, GridCARE said in the report. Those investments could extend beyond the upgrades needed to support data center load to encompass reliability, resilience and renewable interconnection projects, it said.
A blended scenario where the utility uses some capital for rate relief and some for infrastructure is likelier than either of those binary outcomes, the report added.
The study from GridCARE echoes the conclusions of a separate analysis by Camus Energy, encoord and Princeton University’s ZERO Lab, which said a flexible 500-MW data center could connect to the grid three to five years sooner than an inflexible one, and nearly eliminate incremental power supply costs.
The Camus-encoord-Princeton analysis, which Google funded, addressed the load interconnection question from the data centers’ perspective. It modeled “six real candidate sites within one PJM utility’s territory” capable of supporting a 500-MW data center.
The analysis assumed the data center would agree to interconnect flexibly, with part of its load supported by a firm service agreement from the utility and the remainder — modeled at 20% — with conditional or interruptible service. When local grid capacity is constrained, the utility could interrupt service on the conditional portion, requiring the data center to reduce its load or make up the shortfall with resources like onsite generators or virtual capacity procurements — which Camus and its partners call a “bring-your-own-capacity” model.
That arrangement would allow the data center to connect to PJM’s grid in two years, Camus and its partners said — three to five years faster than an inflexible data center without its own capacity resources. By avoiding the need for new generation or energy storage and internalizing its own capacity costs, the flexible data center would contribute $733 million per gigawatt toward the $764 million per gigawatt cost of its incremental load, the analysis found.
For all that, Camus and its partners found the flexible data center would be able to use grid power more than 99% of the time. To offset curtailments due to transmission or generation constraints, the data center would need to dispatch onsite batteries, generators or load flexibility between 40 to 70 hours per year, they said.
