P044 Remote sensing of facilities

Can remote sensing (optical, thermal, radar, seismic) estimate a data centre's power use or find undeclared sites, within what error and at what false-positive rate?

Remote sensing (EO, thermal, SAR, seismic) of data centres.

Terms: EO: electro-optical (visible-light imagery). SAR: synthetic-aperture radar (all-weather radar imaging). MWIR: mid-wave infrared (a thermal band). MASINT: measurement-and-signature intelligence (detection from physical signatures).

Status

Analyzed A desk analysis of the visual signature by facility size: cooling-tower arrays (economic above 5–10 MW) on a 2,000 m² footprint are anomalous against EIA building statistics; at 1 MW air-cooled chillers and at 100 kW a container look like ordinary commercial buildings. 'Find undeclared sites' clause only, with no false-positive rate and no search without prior candidates; the 'estimate power use within what error' clause has no held work; thermal, radar and seismic clauses are treated qualitatively (seismic sensing 'a confirmation tool, not a search tool'; vibration dampening could defeat it). The classifier result the authors cite (Clymer 2025: all facilities above 10 MW found, all four below missed) is not held and is not evidence here.

"Satellite surveillance requires no cooperation from the target state, making it deployable unilaterally. It is most useful as a first filter for facilities at 10,000 H100-equivalents and above, with corroboration from other measures."
Koopmanschap & Barten (2026), How to Catch a GPU: A Taxonomy of Verification and Enforcement Mechanisms for International AI Agreements (arXiv 2607.22619v1), App. B, Satellite surveillance (2026 (arXiv 2607.22619v1; stamp reads 16 Jun 2026)), L826–828

Posed by

Also raised in: Harack et al. (2025), Verification for International AI Governance (Oxford AIGI); Baker, Kulp, Marks, Brundage & Heim (2025), Verifying International Agreements on AI: Six Layers of Verification for Rules on Large-Scale AI Development and Deployment (RAND WR-A4077-1); Scher & Thiergart (2024), Mechanisms to Verify International Agreements About AI Development (MIRI); Aarne & Petrie (2025), International Security Applications of Flexible Hardware-Enabled Guarantees (flexHEG Part 3); Amodo (2026), AI 2040 Plan A — How You Can Help.

Sub-questions

IDSub-questionWhat the sources report
P044.1What is the error (MW or GPU count) of satellite/EO and permit-based capacity estimates of known sites against ground truth, and what share of global capacity do trackers cover?a source suggests an answer (not checked)
P044.2Can thermal-IR (e.g. 3.5 m MWIR) or SAR imagery estimate an AI data centre's operating load to a stated precision, or tell idle from loaded?open
P044.3Can cooling-plant heat-rejection measurements bound a site's IT/unmetered electrical load to a stated precision (MW)?open
P044.4Has automated EO detection found previously unknown data centres, with detection and false-positive rates?partly answered
P044.5What is the smallest data-centre load detectable from commercial thermal satellite imagery at a stated detection rate (a target to be agreed)?open
P044.6On an expanded labelled set, what is a satellite classifier's true-positive rate for AI facilities of 1-10 MW?partly answered
P044.7Has any workload information been recovered from EM or acoustic emissions at a data-centre perimeter (MW-scale facility, not a single device)?open
P044.8At what standoff distance do seismic sensors detect an N-MW facility's chiller/fan signature, with and without vibration isolation?open

Sub-questions not yet stated precisely enough to check

IDSub-question
P044.9Can remote sensing distinguish AI from non-AI data centres or other industry (cooling differentials are easy to hide)?
P044.10Can geophysical MASINT tell data-centre construction from other underground construction?

Possibly relevant work found by automated search

Found by AI agents searching for this problem's sub-questions, mostly by keyword and citation in the OpenAlex scholarly database and on the web; a second AI pass reviewed each one, and we spot-checked some by hand. They seemed reasonably related, as far as one can tell from outside each paper's field. Some come from neighbouring fields, such as arms control and financial audit, and bear on the problem by analogy. Not reviewed against the status above; a paper listed here does not change that status. 928 of 1052 quotes checked word for word against the paper's text.

For P044.1: What is the error (MW or GPU count) of satellite/EO and permit-based capacity estimates of known sites against ground truth, and what share of global capacity do trackers cover?

For P044.2: Can thermal-IR (e.g. 3.5 m MWIR) or SAR imagery estimate an AI data centre's operating load to a stated precision, or tell idle from loaded?

For P044.3: Can cooling-plant heat-rejection measurements bound a site's IT/unmetered electrical load to a stated precision (MW)?

For P044.4: Has automated EO detection found previously unknown data centres, with detection and false-positive rates?

For P044.7: Has any workload information been recovered from EM or acoustic emissions at a data-centre perimeter (MW-scale facility, not a single device)?

For P044.8: At what standoff distance do seismic sensors detect an N-MW facility's chiller/fan signature, with and without vibration isolation?

* Full text not reviewed; more likely to contain errors.

Data