Strategic Throttling in Large Cloud Computing Platforms

Large cloud providers compete on price and service quality while serving customers who differ in willingness to pay and delay sensitivity. We model this competition for cloud platforms where computing time is nonstorable and service quality depends on processing speed and interruption risk. When the customer base is large enough, aggregate uncertainty in capacity utilization vanishes in steady state, so that static pricing and precise segmentation by quality become viable. The provider leverages this predictability by partitioning capacity into service tranches that throttle lower-tier users, inducing self-selection into contracts that differ in speed and interruption probability. A pay-as-bid auction implements the provider-optimal allocation while preserving incentive compatibility. Under duopolistic competition, efficient quality provision emerges when switching costs are low or vertical heterogeneity is large—driven either by competitive pressure or by a platform’s incentive to expand coverage. Our results reveal a key trade-off: scale reduces demand uncertainty and improves operational efficiency, but the resulting market power sustains service throttling. Lowering switching costs or promoting interoperability improves welfare more effectively than structural interventions that reduce scale.

History: Olivia Liu Sheng, Senior Editor; Sameer Mehta, Associate Editor.

Supplemental Material: The online appendix is available at https://doi.org/10.1287/isre.2024.1124.