Data center demand response is often discussed in the context of large facilities, but smaller sites can also contribute flexible load while managing electricity costs. For operators of small and medium data centers, the challenge is to identify which loads can be adjusted, when changes are practical, and how to protect operational requirements such as service quality and equipment limits.
A framework developed for small and medium data centers combines energy-efficiency measures, load flexibility and demand response. It considers both the facility operator’s perspective and the distribution system operator’s need to coordinate multiple sites. The central lesson for automation and operations teams is that demand response works best when it is treated as a managed operating strategy—not as an isolated instruction to cut power.
Why focus on small and medium data centers?
Data centers rely on energy-intensive information and communication technology (ICT), alongside supporting infrastructure such as cooling and uninterruptible power supplies (UPS). Although individual smaller facilities may have less capacity than large campuses, the combined electricity use of small and medium data centers is significant. The thesis notes that they account for more than half of data center electricity consumption and that smaller facilities may have greater energy-efficiency challenges than larger ones.
Yet energy research and improvement programs have often focused on large facilities. Smaller operators may therefore have fewer established processes for monitoring energy use, evaluating efficiency opportunities or assessing how much load can be shifted. A structured approach can help make those decisions more systematic.
Build efficiency into load management
Demand response is not a substitute for energy efficiency. Efficiency measures reduce the energy needed to provide a service; demand response adjusts consumption in response to electricity prices or a program request. Combining the two can give operators a clearer picture of their baseline load and the flexibility available without compromising essential operations.
The proposed framework uses two broad stages:
- Assess efficiency and flexibility: Examine energy use and determine which loads could be adjusted, over what period, and under what operating constraints.
- Apply a demand response approach: Use dynamic electricity tariffs or an incentive-based contract to guide load changes during specified periods.
This sequence matters. Without a reasonable understanding of normal consumption and operational constraints, it is difficult to distinguish genuine flexibility from reductions that could disrupt workloads or simply shift consumption to another period.
Which data center loads can be managed?
The optimization model considers three areas: ICT workload, cooling and UPS. These loads are connected. Changes in computing activity affect server power and heat output, which in turn influences cooling demand. UPS operation also affects the facility’s power profile and can be considered within the management strategy.
ICT workload
Workload management is not simply a matter of turning servers off. The model represents computational capacity allocated to meet a demand response event, allowing ICT load to be considered alongside the other facility loads. In practice, operators need to account for their service requirements and the consequences of changing workload allocation.
Cooling and UPS
Cooling is a substantial part of the data center’s supporting infrastructure, and its demand is linked to the heat produced by ICT equipment. The model includes cooling load and UPS behavior so that these components are considered together rather than optimized in isolation. The source does not imply that every cooling or UPS load can be freely curtailed; available flexibility depends on the modeled constraints and operating conditions.
Optimization at the facility and grid levels
At the data center level, a mixed-integer linear programming (MILP) approach selects how flexible loads are managed. Under a price-based strategy, the objective is to reduce energy costs in response to dynamic tariffs. Under an incentive-based arrangement, the model seeks to meet the requested load reduction while accounting for the contractual approach.
The second layer addresses coordination across multiple facilities. A random-rotation and fairness algorithm selects which participating data centers are called for each demand response event. By aggregating their flexible loads, the distribution system operator can treat the group contractually as a larger resource, while rotating selection to distribute participation more equitably.
For integrators and energy managers, this distinction is useful: a facility-level controller decides how a site responds, while the aggregation logic determines which sites are asked to participate. Both layers need clear constraints and a defined basis for measuring response.
What the modeled results show
In the thesis simulations, incentive-based demand response produced reported savings of 1.33% for small data centers and 5.15% for medium data centers. With dynamic tariffs, the reported savings were 0.21% and 0.68%, respectively. These are results from the scenarios studied, not guaranteed savings for other facilities; outcomes depend on the modeled loads, tariffs, incentives and operating assumptions.
From the distribution system operator perspective, simulated reduction potential varied with the scenario. For a single-day analysis, the reported range was 18.02% to 91.16%; across 22 business days, it ranged from 13.12% to 71.38%. The analysis also found that incentive-oriented scenarios encouraged participation, while unchanged or penalty-based cases could produce less favorable conditions. These figures describe the modeled scenarios and should not be treated as a forecast for a particular grid or data center portfolio.
Practical implications for operators
Before joining a demand response program, operators can use the framework’s logic to structure an internal assessment. Establish a view of ICT, cooling and UPS consumption; identify operational limits; and determine which tariff or contract structure is relevant. Then evaluate how a requested reduction would be delivered and whether rebound consumption could affect costs in the following period.
Demand response can create a route for smaller facilities to manage energy costs and contribute to grid flexibility, but participation depends on making flexibility measurable and operationally credible. The practical takeaway is to pair efficiency assessment with coordinated load management, and to evaluate incentives, constraints and recovery effects before committing to a response strategy.