The Environmental Potential of Hyper-Scale Data Centers: Using Locational Marginal CO$_2$ Emissions to Guide Geographical Load Shifting

TL;DR

This paper presents a model for geographically shifting data center loads based on local CO$_2$ emissions data, enabling significant reductions in carbon footprint without system operator collaboration.

Contribution

It introduces a bottom-up load shifting approach using locational marginal CO$_2$ emissions, which does not require direct system operator cooperation.

Findings

Load shifting can significantly reduce CO$_2$ emissions.

Modest load shifts yield substantial environmental benefits.

The approach outperforms centralized market bidding strategies.

Abstract

Increasing demand for computing has lead to the development of large-scale, highly optimized data centers, which represent large loads in the electric power network. Many major computing and internet companies operate multiple data centers spread geographically across the world. Thus, these companies have a unique ability to shift computing load, and thus electric load, geographically. This paper provides a "bottom-up" load shifting model which uses data centers' geographic load flexibility to lower CO$_2$ emissions. This model utilizes information about the locational marginal CO$_2$ footprint of the electricity at individual nodes, but does not require direct collaboration with the system operator. We demonstrate how to calculate marginal carbon emissions, and assess the efficacy of our approach compared to a setting where the data centers bid their flexibility into a centralized market. We find that data center load shifting can achieve substantial reductions in CO$_2$ emissions even with modest load shifting.