Collaborative optimization of electricity, computation, and thermal resources in data centers: Progress, challenges, and future directions

Abstract

Data centers increasingly rely on coordinated operation across power, computation, and thermal management resources, but existing studies differ substantially in the resources coordinated, decision methods adopted, and evidence supporting their conclusions. This literature review synthesizes 311 unique studies spanning power–computation, power–thermal, computation–thermal, and power–computation–thermal coordination. The literature remains concentrated in pairwise coordination, particularly power–computation, while broader integration across all three domains has received growing but comparatively less attention. Across the four coupling families, scheduling and resource allocation remain prominent, whereas reinforcement learning and other advanced decision methods are being applied to increasingly diverse coordination problems. More importantly, broader resource integration is not consistently accompanied by stronger treatment of uncertainty, validation, or applicability. Differences in system boundaries, workloads, climates, infrastructure, baselines, and evaluation settings also limit the comparability and transferability of reported benefits. These findings suggest that future progress depends not only on expanding coordination scope, but also on establishing more reliable evidence for when and where coordinated flexibility is effective. Key priorities include quantifying usable flexibility under uncertainty, assessing the incremental value of broader coordination, strengthening validation, clarifying applicability domains, and improving comparability across studies.