A Power Market Model with Hypersaclers and Modular Datacenters

The rapid adoption of AI has led the growth of computational demand, with large language models (LLMs) at the forefront since ChatGPT's debut in 2022. Meanwhile, large amounts of renewable energy have been deployed but, ultimately, curtailed due to transmission congestion and inadequate demand. This work develops a power market model that allows hyperscalers to spatially migrate LLM inference workloads to geo-distributed modular datacenters (MDCs), which are co-located with near renewable sources of energy at the edge of the network. We introduce the optimization problems faced by the hyperscaler and MDCs in addition to consumers, producers, and the electric grid operator, where the hyerscaler enters an agreement to lease MDCs while ensuring that the required service level objectives (SLOs) are met. The overall market model is formulated as a complementarity problem, where the proof is provided showing the existence and uniqueness of the solutions. When applying the model to an IEEE RTS-24 bus case study, we show that even with a provision that requires MDCs to disclose the CO$_2$ emissions associated with their energy supply sources, renting less polluting MDCs is unlikely to yield meaningful emission reductions due to so-called contract-reshuffling. The situation can be mitigated when conventional loads are supplied by forward contracts through power purchase agreements. This also leads to a decline in system congestion when the hyperscaler becomes increasingly cost-aware.

Source: arXiv:2602.19310v1 - http://arxiv.org/abs/2602.19310v1 PDF: https://arxiv.org/pdf/2602.19310v1 Original Link: http://arxiv.org/abs/2602.19310v1