CoChat

Grid topology and the spatial cost of AI data-centre siting

Open access
Deng, Xiuyue, Poletti, Stephen, Hazledine, Timothy · 2026 · Research Square

The DOI matches its Crossref record; 8 of the 8 references it cites were found in a registry or an index; none of the 32 sources its claims cite is checked for support yet.

Abstract

The rapid growth of AI data centers is straining power grids. Yet whether siting costs are driven by short-run electricity prices or network topology remains unclear. Here, we address this by comparing two distinct 2030 power systems, New Zealand's radial grid and Ireland's meshed ring. Testing identical loads across hundreds of scenarios varied by gas and carbon prices, demand, weather and transmission capacity, we find network topology dominates siting costs. New Zealand splits into two cost zones across the Cook Strait link, where the 4.18% gap matches the one-way converter loss exactly and vanishes when that loss is removed. In contrast, Ireland stays a single cost group under routine operation with variations below 0.1%. Our results demonstrate that price signals effectively guide data center siting primarily where the network is bottlenecked. In highly meshed grids, physical grid reinforcement becomes essential.

Overview

CoChat Summary

TL;DR

The paper compares identical firm AI-like load injections across two calibrated 2030 power systems and finds that siting cost is governed by network topology, not short-run price parameters. New Zealand’s radial/islanded grid splits into two cost classes separated by the Cook Strait HVDC converter loss, while Ireland’s meshed ring remains a single class under routine operation.

Key findings

  • In New Zealand, injecting the same 100 MW load at each candidate node produced exactly two equivalence classes aligned with the islands: South Island nodes at 71.96 EUR/MWh and North Island nodes at 74.96 EUR/MWh, a 4.18% gap that rose modestly to 4.95% at 500 MW.
  • The New Zealand gap matched the Cook Strait HVDC one-way converter loss exactly: with η = 0.96, the paper reports (1 − η)/η = 4.17%, and the two classes merged when the link was made lossless, leaving only 0.013% spread.
  • In Ireland, the same injection produced a single class: the generation-cost spread across five county nodes was 0.016% at 100 MW, with the raw spread including uniform scarcity rent only 0.007%.
  • Under perturbations, Ireland stayed single-class across the main sensitivity suite; the widest reported generation-cost spread was 0.127% under a halved gas price, and a frozen-transmission test only broke the single class at higher loads, yielding 3.54% at 300 MW and 76.01% at 500 MW.

Methods

  • The study uses a controlled cross-country experiment in PyPSA-Earth, solving both systems as linear direct-current optimal power flow models over 2,920 three-hour time steps for the full year.
  • The main treatment is a firm, inelastic, weather-independent point load injected at every candidate node: 8 nodes in New Zealand and 5 county nodes in Ireland, at 100, 300, and 500 MW.
  • Comparison is based on the change in annual system cost per injected MW, with Ireland’s metric decomposed into a uniform scarcity component and a generation component; equivalence classes are defined by a spread below 1%.

Limitations

  • The paper states that the model is a linearized DC optimal power flow and therefore excludes N-1 contingencies, voltage/reactive-power limits, and full AC security; it also treats the Irish ring as a lossless transport network without reactance.
  • It omits storage and demand response, and the authors note these could introduce inter-temporal coupling not captured by the static treatment.
  • Evident from the text: the results are based on 2030 scenarios, not realized systems, and the Irish baseline is intentionally islanded by removing cross-border interconnectors, so the conclusions apply to the modeled topology rather than the live market.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.21203/rs.3.rs-10300906/v1

Cites

8 of 8 references verified0 citation errors · 0 not checkable

Statistics

Recomputing p-values needs the paper’s full text.

Verification

Claims the paper makes

0 Valid · 3 Not valid · 22 Not checked · 130 Own results and interpretations

51 more claims

Create a free account to see all 155 claims and to verify the 32 references still open.

See all claims — free account
  • It may state two separate claims as one. 88
  • Probably not a substantial claim. 22

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 8 references

Research works printed but not linked: 3

  1. 9Wan H, Fang L GridOperationalBenefitAnalysisofDataCenterSpatialFlexibility:CongestionRelief,RenewableEnergyCurtailmentReduction,andCostSaving.arXiv preprint(2025).
  2. 17Chen D, Zhou Z, Cai Y, Qin J & Katchova, A. Concentrated siting of AI data centers drives regional power–system stress under rising global compute demand. Preprint at arXiv (2026).
  3. 20CRU Review of Large Energy Users Connection Policy. (Commission for Regulation of Utilities, Dublin, Ireland, 2025).

Metadata

Authors
Deng, X., Poletti, S., & Hazledine, T.
Affiliations
  • University of Auckland (NZ)
  • The University of Auckland
Venue
Research Square
Published
2026-07-10
OpenAlex
W7167893792
Cited by
0
References
8
Type
posted content
Language
en
Open access
Locations
Research Square (acceptedVersion, cc-by)
Topics
  • Physical Sciences › Engineering › Electrical and Electronic Engineering › Electric Power System Optimization
  • Physical Sciences › Engineering › Electrical and Electronic Engineering › Integrated Energy Systems Optimization
  • Physical Sciences › Engineering › Electrical and Electronic Engineering › Smart Grid Energy Management
Keywords
AI data centers, power grid topology, radial grid, electricity prices, transmission capacity, grid reinforcement, converter losses

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Grid topology and the spatial cost of AI data-centre siting. https://app.cochat.ai/papers/10.21203/rs.3.rs-10300906/v1
The paper
Deng, X., Poletti, S., & Hazledine, T. (2026). Grid topology and the spatial cost of AI data-centre siting. Research Square. https://doi.org/10.21203/rs.3.rs-10300906/v1

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.