What Data Center Flexibility Really Costs
New modeling examines the costs of data center flexibility for operators and how optimal mixes of backup generation and storage shift as flexible load increases.
With data centers flooding interconnection queues, we need strategies to enable faster deployment. One solution gaining traction is data center flexibility. By exchanging faster interconnection for the ability to call on data centers for peak shaving, flexibility can allow for greater usage of existing generation and transmission capacity, defer costly infrastructure upgrades and accelerate time-to-power for new facilities.
What is Data Center Flexibility?
Flexibility here refers to data centers strategically reducing their grid-facing load when the grid is stressed, either by limiting facility energy use or tapping behind-the-meter generation/storage, rather than acting like an inflexible, always-on customer. This can happen in a few ways:
- Spatial shifting: Shifting workloads to other sites in regions that have available power supply. Workloads must be able to tolerate latency impacts and capacity must be available within a data center fleet.
- Temporal shifting: Pausing non-critical workloads and resuming them later on.
- On-site resources: Using behind-the-meter generation (like solar, fuel cells, or turbines) or storage to provide for data center load during peak events.
It’s not a perfect solution. Curtailing load in this way comes with costs that must be borne by data center operators, utilities, or ratepayers. Understanding these costs is essential if flexibility is to play a role in bringing data centers online in current grid conditions.
This report, From Firm Load to Flexible Resource: Understanding Data Center Flexibility and Its Costs, provides (1) a comprehensive literature review on data center flexibility, and (2) illustrative modeling of the CAISO balancing region showing how flexibility requirements could enable large-scale data center deployment with existing capacity. This analysis also examines the costs for data center operators and how the cost-optimal mix of behind-the-meter generation and storage shifts as flexible load grows under different regulatory environments.
Here are the main takeaways from this work:
- Data centers are overwhelming the grid. Interconnection queues are clogged with both new generation and new large loads. Data centers are projected to add tens of gigawatts of demand, but grid expansion is years behind due to long lead times for turbines, transformers, and transmission. The result is a bottleneck: load growth is outpacing the grid’s ability to deliver new capacity.
- Flexibility is finite but meaningful. Data center flexibility can be valuable in avoiding unnecessary utility-scale capacity development, but there’s only so much headroom to capture before diminishing returns set in. Flexibility can help bridge today’s grid bottlenecks and get projects online faster, but it isn’t a substitute for long-term transmission and generation buildout. As Figure 1 suggests, each flexible data center load can increase the energy costs across all flexible data centers, especially when the set of usable generation and storage technologies are constrained.
- Costs increase with flexible load deployment. As more flexible data center load comes online, the cost of maintaining the existing peak load increases as data centers have to deploy higher cost generation to serve increasing energy requirements.
- Low data center penetration favors low-cost options. For initial data center load on grids with steep load duration curves, the cost-minimizing technologies to provide needed flexibility are those with the lowest capital cost, even given high variable costs of energy produced. The modeling in this paper captures this dynamic by allowing emergency diesel generators to run for only a small number of hours each year, covering the steepest peaks that would otherwise be prohibitively expensive to meet with other modeled technologies.
- Regulation shapes the toolkit. Permissive regulation lets operators use thermal or fuel cells, which are cost-effective when grid tariffs exceed their fuel-based costs. Where rules limit the available technologies to storage, economics hinge on capacity factor: if assets only run during rare curtailments, costs rise sharply, though storage can still be cost-effective if TOU tariffs have both frequent cycling opportunities and a large enough price delta between on-peak and off-peak periods.
Figure 1. Illustrative cost increases from different levels of data center development.
Where do we go from here?
Utilities and system operators face a new reality: serving firm data centers is likely to drive costs higher. To make data center flexibility a real solution, planning models must reflect both behind-the-meter self-supply and workload flexibility, and tariff designs must evolve accordingly. Early efforts by utilities, hyperscalers, and regulators are promising, and scaling flexibility will be unlocked through greater public data sharing, more granular modeling, and transparent demonstrations that build confidence and enable deployment.
Other Resources
Dive into the complete results by downloading the attached whitepaper, and listen to the AI generated podcast of the report.
