# Data center heat dissipation & detection

> Explore data center cooling, heat dissipation and detection in an interactive 3D model, with satellite and local sensing methods and cited evidence.

By [@luiscosio](https://www.luiscos.io/).

Interactive model: https://luiscosio.github.io/sieve/

## How do data centers dissipate heat?

Cooling moves heat from the computers into air, water, the ground or a nearby user. The model shows towers, dry coolers, water outfalls, storage and heat reuse. Storage fills up, and reuse needs demand. [2](https://eta-publications.lbl.gov/sites/default/files/2024-12/lbnl-2024-united-states-data-center-energy-usage-report_1.pdf) [33](https://www.vantaanenergia.fi/en/about-us/projects/varanto-the-cavern-thermal-energy-storage/)

## Can satellites detect data-center heat?

They measure infrared from exposed surfaces: roofs, ground and surface water. They can’t read server power or see a buried or underwater hall. Weather, sunlight, materials and pixel size all change what an image means. [56](https://scienceandglobalsecurity.org/archive/sgs08zhang.pdf) [59](https://www.usgs.gov/landsat-missions/landsat-9) [16](https://arxiv.org/abs/2609.18824)

## Is satellite imagery the only detection method?

No. You can also pick electrical metering, network activity, local sound, vibration and EM sensors, and on-site inspection. Each needs its own access and has its own limits. [13](https://arxiv.org/abs/2607.22619) [54](https://www.nerc.com/globalassets/who-we-are/standing-committees/rstc/whitepaper-characteristics-and-risks-of-emerging-large-loads.pdf) [12](https://arxiv.org/abs/2408.16074)

## What does the evidence status mean?

Demonstrated: seen at real data centers. Analogue: seen in another kind of system. Proposed: a published proposal or simulation. Inference: our reading of cited facts. Gap: the reviewed sources don’t cover the claim. None of these is a probability. [1](https://epoch.ai/data/data-centers-documentation/methodology) [62](https://beyondparallel.csis.org/thermal-imagery-analysis-of-yongbyon/) [16](https://arxiv.org/abs/2609.18824)

## Why are some detection methods gray?

A gray method has nothing to look at in this setup. Dry cooling alone, for example, makes no tower plume. That’s a different thing from weak evidence or bad weather. [2](https://eta-publications.lbl.gov/sites/default/files/2024-12/lbnl-2024-united-states-data-center-energy-usage-report_1.pdf) [23](https://d-nb.info/1241453594/34)

## Is this a validated prediction of detection performance?

No. The scene is a drawing, and heat arrows show direction only. It predicts no temperatures, thresholds or probabilities. For real data, open a field example, starting with Colossus 1 under Land thermal IR. [59](https://www.usgs.gov/landsat-missions/landsat-9) [60](https://edcintl.cr.usgs.gov/downloads/sciweb1/shared/co/nli_pecora/pecora_22/presentations/TS_6-1_TIRS%202_MMontanaro.pdf)

## Evidence labels

This exhibit assigns these categories. Citations support the examples; they aren’t a standard scoring scale.

### D · Demonstrated in data centers

The cited work documents this signal at data centers. That doesn’t mean every selected setup can be detected.

Example: identifying cooling equipment in images of data centers. [1](https://epoch.ai/data/data-centers-documentation/methodology)

### A · Observed in another system

Someone has seen this effect at another kind of site. Using it for this data-center setup is our extrapolation.

Example: satellite images of warm-water outfalls at nuclear power plants. [62](https://beyondparallel.csis.org/thermal-imagery-analysis-of-yongbyon/) [26](https://www.sciencedirect.com/science/article/pii/S0034425723002584)

### P · Proposed or modelled

A published proposal or model supports the idea. The cited work hasn’t shown field detection of this setup.

Example: estimates of the heat and sound signatures of subsea data centers. [16](https://arxiv.org/abs/2609.18824)

### I · Inference from cited sources

This exhibit combines cited mechanisms or site facts in a way the sources haven’t tested.

Example: a substation suggests large electrical capacity. Tying that capacity to a specific workload needs more evidence. [19](https://arxiv.org/abs/2311.02651)

### ? · Gap in reviewed evidence

The reviewed sources have no demonstration, analogue or model for this specific claim. It’s an open question.

Context: the literature notes little evidence on hidden military data-center capacity. That doesn’t measure how detectable it is. [19](https://arxiv.org/abs/2311.02651)

### — · No source in this selection

The selected heat path has nothing for this method to look at. That’s a fact about the model setup. It says nothing about the strength of the research.

Example: dry cooling alone has no evaporative tower plume. [2](https://eta-publications.lbl.gov/sites/default/files/2024-12/lbnl-2024-united-states-data-center-energy-usage-report_1.pdf) [23](https://d-nb.info/1241453594/34)

## Scope note

By scenario assumption, chip acquisition, supply-chain accounting, chip attestation and on-chip workload checks are excluded. Cooperative site inspection is still in scope. [69](https://arxiv.org/abs/2303.11341) [70](https://www.cnas.org/publications/reports/preventing-ai-chip-smuggling-to-china) [71](https://arxiv.org/abs/2609.07637) [13](https://arxiv.org/abs/2607.22619) [21](https://www.rand.org/pubs/working_papers/WRA3056-1.html) [72](https://static1.squarespace.com/static/64edf8e7f2b10d716b5ba0e1/t/6827b67275666f3757f134ea/1747433075281/Location+Verification+two-pager.pdf) [3](https://arxiv.org/abs/2507.15916)

[Complete references](https://luiscosio.github.io/sieve/references.md)