The Data Centre Cooling Studio: One Instruction, a Whole Plant Designer

By DigiEntropy Engineering · 2026-07-25 · 14 min read

One paragraph of instruction became a cross-brand data-centre chiller plant designer: 8,760-hour runs on real compressor maps, 70 hubs, ten tabs, and five rounds of review by twelve AI personas. An interactive tour.

About this post — Authored by an AI assistant using DigiEntropy's polynomial corpus, the universal compressor predictor, and the project's chart-generation tooling. Charts are produced by Python scripts that read the same database the live site queries; tables and formulas are pulled from the same engineering modules. Findings reflect the project's current dataset and methodology — send corrections or deeper questions to admin@digientropy.com.

DigiEntropy's Data Centre page designs the cooling plant for a data centre, hour by hour, from real compressor data. You give it a site, the IT load and the rack platform. It picks the chillers from the whole cross-brand catalogue, runs all 8,760 hours of a typical year through each candidate's real performance map and operating envelope, and then tells you what to buy, what it costs, how it behaves when a chiller trips, and which rules it passes in that country.

It started as one paragraph of instruction, typed late at night. This post shows what the page does, using the real output of one run, and how it was built: AI agents researched it, AI agents built it, and then twelve AI reviewers, each playing a different person from the data-centre industry, took it apart five times.

The post is interactive. Click the tiles, switch the tabs, drag the sliders, hover the charts. Every number comes from one run of the Data Centre page for a 20 MW hall of NVIDIA GB300 racks in Singapore, saved from the same API responses that drew the screenshots.

8,760 - hours simulated - per candidate, per run

Every finalist compressor runs all 8,760 hours of the site's typical weather year on its own capacity-control map, with its real envelope checked every hour. A ten-year check (87,600 hours) runs on top for the chiller-less question.

70 - data-centre hubs - ten years of hourly weather each

Each hub carries ten years (2015-2024) of hourly dry-bulb and wet-bulb from the ERA5 reanalysis (ECMWF / Copernicus, via Open-Meteo), a sourced large-user electricity tariff and a grid carbon factor. 65 of them also have a published water and wastewater tariff.

962 - compressor ratings scanned - for the Singapore brief

Screw and centrifugal ratings at 50 Hz, on the refrigerants Singapore's policy allows (R1234ze(E), R290, R1233zd(E)), from 5 brands, through the same Selection engine the rest of the platform uses.

1.089 - pPUE of the cooling - Singapore, 20 MW GB300

Cooling adds 8.9 % to the IT energy. With fixed and proportional electrical losses, the site PUE estimate is 1.148.

14 - research reports - behind every assumption

…

More from the DigiEntropy engineering blog