Locational Marginal Emissions for Carbon-Aware Data Center Operations in Large-Scale Power Grids
arXiv preprint (math.OC), MIT Operations Research Center · Luc Cote, Andy Sun
The authors compute hourly locational marginal emissions (LMEs) over one year on a 1493-bus model of the US Western Interconnection (WECC) and look at them from a data-centre operator's point of view.
CC BY 4.0 — share with attribution
On this page
Overview
Carbon accounting methods for electricity consumption face challenges regarding physical deliverability, double counting, additionality and impact magnitude; LMEs show potential to address many of these key issues, but their use in large-scale power grids remains understudied. LME characteristics create three distinct regions: the hydropower-dominated Pacific Northwest/Montana, with low and stable LMEs; the coal-heavy Intermountain West, containing often high LMEs; and the Sunbelt, where mixed generation leads to variable LMEs correlated with solar output.
Regional average LMEs: Pacific Northwest and Montana about 90 kgCO2/MWh; Intermountain West (Idaho, Wyoming, Colorado, Utah) about 710 kgCO2/MWh. Across WECC, the average LME-based emissions intensity attributed to load was about 672 kgCO2/MWh, while the grid-average emission factor was about 202 kgCO2/MWh; the load carbon account (419.1 MMT) was almost 3 times the grid's scope 1 emissions (125.9 MMT). Emission rates assumed in the simulated grid: coal 800, gas 450 kgCO2/MWh, wind/solar/nuclear/hydro 0.
Carbon-aware shifting of up to 20% of data-centre load within a day (load bounded between 80% and 120% of baseload) realised 86-92% of the LME-predicted emission reduction; for example, 8 × 100 MW sites across WECC were expected to save 4,168 t CO2. Geographic diversity matters more than size: the 8-site WECC portfolio realised 333% of the reduction of the 3-site California portfolio with only 267% more controllable load, and doubling site size from 100 to 200 MW raised realised reductions by only 171-179%.
Shifting load on locational marginal prices (LMPs) instead of LMEs cut system cost by about 5% but increased system emissions by 78.5 t CO2, because load moved toward cheaper, coal-margin Mountain West sites. Siting a constant 200 MW load at 1000 random bus/hour samples gave a realised emission increase of 1093.6 t CO2, 114% of the LME-expected 959.1 t; siting a 200 MW generator gave -845.0 t, 92.6% of the expected 912.8 t. Urban data-centre buses mostly did not show the midday solar dip in LMEs (Sacramento was the exception) because they sit downstream of congestion that curtails renewables. Limits: simulations restricted to small areas such as California gave much less accurate LME signals; the model is for the US Western Interconnection and is a preprint.
Key points
- Regional average LMEs: Pacific Northwest and Montana about 90 kgCO2/MWh; Intermountain West about 710 kgCO2/MWh; the Sunbelt is variable and correlated with solar output.
- Across WECC, the average LME-based emissions intensity attributed to load was about 672 kgCO2/MWh against a grid-average emission factor of about 202 kgCO2/MWh; the load carbon account (419.1 MMT) was almost 3 times the grid's scope 1 emissions (125.9 MMT).
- Emission rates assumed in the simulated grid: coal 800 kgCO2/MWh, gas 450 kgCO2/MWh, wind/solar/nuclear/hydro 0.
- Carbon-aware shifting of up to 20% of data-centre load within a day (load between 80% and 120% of baseload) realised 86-92% of the LME-predicted emission reduction.
- Geographic diversity matters more than size: the 8-site WECC portfolio realised 333% of the reduction of the 3-site California portfolio with 267% more load; doubling site size raised realised reductions by only 171-179%.
- Shifting load on LMPs cut system cost by about 5% but increased system emissions by 78.5 t CO2 (load moved to coal-margin Mountain West).
- Siting a constant 200 MW load at 1000 random bus/hour samples gave an emission increase of 1093.6 t CO2 (114% of the LME-expected 959.1 t); a 200 MW generator gave -845.0 t (92.6% of the expected 912.8 t).
- Urban data-centre buses mostly did not show the midday solar dip in LMEs (Sacramento the exception); simulations restricted to small areas such as California gave much less accurate LMEs.
Using it in design
- In data-centre carbon accounting, consider hourly locational marginal emissions (LME) rather than an annual-average grid factor; the LME-based intensity (about 672 kgCO2/MWh) was about 3 times the average factor (202).
- In siting, assess the region by LME pattern (Pacific Northwest ~90, Intermountain West ~710 kgCO2/MWh); geographic diversity is worth more than single-site size.
- Drive load shifting by emission signal (LME) rather than price signal; shifting on LMP cut cost 5% but raised emissions.
Monthly newsletter · free
Keep your reading list current
Once a month: new library sources, changes to Turkish regulations and upcoming data centre events.
SoSo CAD
Compare the cost of design decisions
Evaluate equipment costs, energy expenses and total cost of ownership (TCO) using your inputs and assumptions in SoSo CAD.
Request a demo
Who should read it
Planners and researchers doing carbon-aware data-centre siting, load shifting and carbon accounting.
Document contents
- Summary
- Introduction
- Methods (1493-bus WECC model, hourly LME computation, LME-guided load shifting, data-centre and renewable siting)
- Results (three LME regions, LME versus grid-average factor, load shifting, geographic diversity, LMP versus LME, siting)
- Discussion and limitations
- Conclusion
Related standards
- GHG Protocol Scope 2 Guidance
Read the document
locational-marginal-emissions-dc-2025.pdf
CC BY 4.0 — share with attribution
The document is provided unaltered because its licence permits redistribution.
Publication details
- Authors
- Luc Cote; Andy Sun
- Publisher
- arXiv preprint (math.OC), MIT Operations Research Center
- Year
- 2025
- Document number
- arXiv:2512.18819
- Document language
- English
- Source type
- Scientific
- Access
- Open licence — downloadable
- Licence
- CC BY 4.0 — share with attribution
- Licence evidence
- arxiv.org
Original licence note
arXiv OAI-PMH record for 2512.18819 gives license http://creativecommons.org/licenses/by/4.0/.
Citation
Luc Cote; Andy Sun (2025). Locational Marginal Emissions for Carbon-Aware Data Center Operations in Large-Scale Power Grids. arXiv preprint (math.OC), MIT Operations Research Center. arXiv:2512.18819. https://doi.org/10.48550/arXiv.2512.18819
Events on this subject
Browse the events calendarAll content and rights belong to the respective authors and publishers. SoSo CAD provides only the citation, its own summary and a link to the source, and is not responsible for the sources' content.
Related sources
All sources on this topic
Proceedings of the Twelfth ACM International Conference on Future Energy Systems (ACM) 2021
A Guide to Reducing Carbon Emissions through Data Center Geographical Load Shifting
For geographic load shifting, consider locational marginal carbon emissions (lambda-CO2) as the signal instead of LMP or regional average carbon intensity; it gave the largest reduction (-1.44%).
- Lambda-CO2 shifting saved about 0.14 t CO2 per MW shifted and still reduced generation cost by 0.27%; LMP shifting cut cost by 1.07% but saved less carbon.
Huawei Digital Power 2022
Next-Generation Data Center Facility White Paper
Support CUE, WUE and a heat-reuse fraction next to PUE in reports.
- Above about 20 kW per rack liquid cooling becomes primary, with air as auxiliary.
arXiv preprint (University of Alberta) 2026
To Defer or To Shift? The Role of AI Data Center Flexibility on Grid Interconnection
In large-load interconnection negotiation, present AI load flexibility (a deferral window of up to about 3 hours) as a lever for interconnection capacity and timing; returns diminish beyond 3 hours.
- In the IEEE 14-bus study, data centres on buses connected to congested lines benefit most from geographic load shifting.
arXiv preprint (cs.AI), Harvard T.H. Chan School of Public Health 2026
Assessing the Carbon Emissions and Energy Consumption of U.S. Hyperscale Data Centers
In fleet or facility energy estimation, use capacity × 8,760 × facility-load coefficient (0.48-0.70, central 0.58) and report the range (67.7-98.6 TWh) explicitly.
- State-level intensities (central scenario): Virginia 535, Oregon 440, Iowa 645, Nebraska 740, Wyoming 1035 gCO2/kWh; Virginia had 142 HDCs, 21.0 TWh, 11.5 Mt.
arXiv preprint (cs.SE), Google 2024
Carbon accounting in the Cloud: a methodology for allocating emissions across data center users
In allocating carbon to data-centre users, give idle power to whoever holds the resource allocation and share dynamic power by actual usage; because an idle server draws about 45% of peak power, the idle share is large.
- Resources with equivalent busy power used to weight idle-power allocation: 1 Intel Skylake N1 CPU core = 20 GiB RAM = 1 TiB SSD = 6 TiB HDD.
Frontiers in Environmental Science (Frontiers Media SA) 2026
Net sustainability assessment of AI-driven data centers in the GCC: an MCDA approach framework
In AI data-centre investment assessment in hot, arid regions, weigh carbon, water, social and governance criteria together with explicit weights (0.40/0.25/0.20/0.15); indirect water (desalination energy) was about 0.68 times direct cooling water.
- In the extended boundary, embodied carbon adds 15-25% to the operational footprint.