Working Paper · WP-2026

Data Center Design, Resilience, Reliability and Sustainability

Power, Cooling, Redundancy and Cost for 100–300 MW Campuses, with a Virginia Reference Design Applied Nationwide and Globally

Published
Status Working paper
Data cutoff 24 October 2026
Version 1.1
Author Collin B. George, CISSP
License CC BY-NC-ND 4.0
Unclassified // Open Source
Paper record
Type
Working paper with a 200 MW reference design and a cost model, tested in Dominion Energy Virginia’s territory and applied across nine U.S. regions and seventeen international markets.
Governing judgment
Electricity supply, not water, limits new capacity in Virginia and in most other markets examined. Water use is set by the cooling design, and direct-to-chip liquid cooling with closed-loop dry heat rejection takes process water out of the siting question in temperate climates.
Method
A review of grid-operator, regulator and company documents current to October 2026, engineering estimates for a 200 MW reference design, and a cost model.
Keywords
Data centers; electricity interconnection; capacity markets; PJM; ride-through; onsite generation; liquid cooling; water use.
JEL classification
L94, L86, Q41, Q48, Q25, K32, R52.
Companion files
The cost-model workbook, Python script, tariff input file, 41-station design-conditions table, station manifest and hour-count script are listed below.
Revision history
Version 1.0, 12 October 2026. Version 1.1, 24 October 2026. Posted on SSRN 31 October 2026.
Abstract

This paper asks which combination of grid service, onsite generation, storage and cooling gives a 100–300 MW data center campus the best resilience and regulatory fit for energization before 2030. It uses Dominion Energy Virginia’s territory and a Henrico County reference site as the test case, then applies the result across nine U.S. regions and seventeen international markets. The method combines a review of grid-operator, regulator and company documents current to October 2026, engineering estimates for a 200 MW reference design, and a cost model published with the paper.

The governing judgment is that electricity supply, not water, limits new capacity in Virginia and in most other markets examined; water use is set by the cooling design, and direct-to-chip liquid cooling with closed-loop dry heat rejection takes process water out of the siting question in temperate climates.

The configuration designed to meet the proposed ride-through rules, ride out a winter gas event or a grid outage, and qualify as a synthetic minor air source is firm utility service, a grid-forming battery, dual-fuel reciprocating engines limited to outages and emergencies, and warm-water liquid cooling. On a common UPS and ride-through baseline, with the grid bill built from Dominion’s tariff sheets, it models at $34–35/MWh, about $69 million a year, above grid-only service with diesel backup; the difference buys outage endurance and grid-forming transfer, not energy. Zoning, air permits and tariff rules, not equipment, decide what onsite generation may do. Nuclear and geothermal supply are not available to a Virginia campus this decade.

If you read nothing else

Electricity supply, not water, is the binding constraint. The recommended configuration is firm utility service, a grid-forming battery, dual-fuel reciprocating engines limited to outages and emergencies, and warm-water liquid cooling. It models at $34–35/MWh, about $69 million a year, above grid-only service with diesel backup; the difference buys outage endurance and grid-forming transfer, not energy.


Full text

Read the paper

Data Center Design, Resilience, Reliability and Sustainability: Power, Cooling, Redundancy and Cost for 100–300 MW Campuses, with a Virginia Reference Design Applied Nationwide and Globally

The full working paper: the grid constraint, power sources compared, onsite gas and air permits, cooling, water, ride-through and load flexibility, redundancy, controls and security, cost, the recommendation by scale, policy options, limitations, application across the United States and global markets, the options by scale, and the review record. View · Download · SSRN
WP-2026 · 63 pp

Companion files

Reproduce the cost figures

These files reproduce every figure in Section 12 of the paper and the climate tables in Sections 6, 16.1 and 17.3. They are published under the same CC BY-NC-ND 4.0 license as the paper.

Cost model workbook, version 2.1

Inputs, tariff, model and design-conditions sheets. Change any blue input and recalculate.
XLSX

Python model script

Computes the same results as the workbook, using the tariff input file.
Python

Tariff input file

Each Dominion component with its rate, unit, effective date and source sheet.
JSON

Design-conditions table

41 stations: cooling design temperatures and hours above each threshold.
CSV

Station manifest

The provenance record for each station, and the input to the hour-count script.
CSV

Hour-count script

Downloads each station’s climate file and recounts the hourly exceedances. Needs network access.
Python

Companion README

What each file is, how to use it, the status date of the tariff inputs and the model’s version history.
Markdown

Limitations

What this paper does not establish

The conclusions do not rest on any single estimate, but several figures must still be replaced with primary or site-specific data before the paper supports a filing, a contract or an investment decision. These include engine, fuel-cell and small modular reactor costs that are engineering estimates, and fit-out and outage figures taken from secondary summaries. Section 15 lists each item and what would replace it.

The cost model is the author’s. AI assistance was used for source retrieval, citation checking and technical review, and every accepted finding was checked by the author against the cited source. This is not peer review, and the paper has not had practitioner review.

The rules it discusses are in motion. PJM’s ride-through requirements are a proposal, not yet a rule.


Citation

Suggested citation

Suggested citation

George, C. B. (2026). Data Center Design, Resilience, Reliability and Sustainability: Power, Cooling, Redundancy and Cost for 100–300 MW Campuses, with a Virginia Reference Design Applied Nationwide and Globally. Sanctir working paper, version 1.1, October 24, 2026. Sanctir LLC, Redmond, Washington. Licensed under CC BY-NC-ND 4.0. SSRN 7671020.


Author

About the author

Collin B. George, CISSP, is the principal of Sanctir LLC, an independent research and advisory practice working on national security, export controls, sanctions, and defense industrial base risk.

Sanctir is a solo practice. This paper was prepared in the author’s personal capacity, is not official U.S. government analysis, and is not affiliated with any government agency, academic institution, or defense contractor. It uses no classified, controlled, or proprietary information and reflects no agency position.

ORCID 0009-0007-8162-6839 · SSRN author page · Full background · Contact