Data centre campus construction tracking for investment intelligence
Very-high-resolution optical and thermal infrared imagery can track hyperscale data centre construction from groundbreak to energisation, giving investors and energy planners an independent view of capacity timelines. Internal IT fit-out remains invisible from orbit.
Sensors
- Maxar WorldView-3: 30 cm panchromatic, 1.24 m multispectral at nadir; revisit roughly 1 to 4.5 days depending on latitude and tasking priority. The resolution is sufficient to distinguish cooling tower units, transformer bays and rooftop mechanical plant from surrounding hardstanding.
- Airbus Pléiades Neo: 30 cm native resolution across four multispectral bands; same-day stereo acquisition available, enabling rudimentary building height estimation from shadow length or stereo parallax. Revisit 1 to 2 days at mid-latitudes.
- Planet SkySat: 50 cm resolution with tasking on demand; revisit can be daily with appropriate tasking. Best used for change detection between epochs rather than fine structural interpretation. Useful for high-cadence construction-pace monitoring.
- NASA ECOSTRESS: Thermal infrared instrument on the International Space Station; spatial resolution approximately 38 m × 69 m, irregular revisit of roughly 3 to 5 days depending on ISS ground track. Detects surface temperature anomalies consistent with operational cooling rejection heat. Cannot resolve individual cooling towers but can flag thermally active building footprints against shell-and-core neighbours.
- Landsat 8/9 TIRS: Thermal infrared at 100 m resolution (resampled to 30 m in products), 16-day revisit. Coarser than ECOSTRESS but a consistent 40-year archive enables baseline thermal characterisation of a site before construction begins. Useful for establishing pre-development ambient temperature.
What the construction sequence actually reveals
Hyperscale data centre campuses follow a recognisable physical sequence. Ground clearance and foundation work appear first as bare soil or compacted hardstanding. Steel frame erection follows, producing shadows that allow approximate building height estimation from high-resolution stereo or even single-pass shadow geometry. Roof installation, external cladding and then the arrival of cooling infrastructure mark the transition from structural shell to mechanical-electrical completion.
Each of those stages is observable from orbit at 30 cm to 50 cm resolution. What is not observable is anything inside the building envelope: server rack installation, power distribution units, cabling, or the energisation of IT load. This is the honest limit of the method, and it matters. A building that looks complete from 500 km may be a shell awaiting fit-out for six months or more. Investors who conflate structural completion with operational capacity will systematically overestimate near-term supply.
Cooling towers and substations as capacity proxies
Two external features are particularly informative. Cooling towers, whether mechanical-draft or dry-cooler arrays, are large, visually distinctive and appear only when a facility is approaching operational readiness. Their installation typically precedes energisation by weeks to a few months. At WorldView-3 or Pléiades Neo resolution, individual cooling units can be counted, and their footprint area correlates loosely with the heat rejection capacity the operator has committed to.
Substations are the second indicator. A hyperscale campus drawing 100 MW or more requires dedicated high-voltage infrastructure. Transformer bays, switchgear buildings and incoming transmission lines are all visible at 30 cm resolution and their construction progression, from civil works to equipment installation to the arrival of large power transformers (which are distinctive rectangular objects), provides a parallel timeline to the building work. Substation completion is a necessary but not sufficient condition for the campus to be live.
Neither proxy is a direct measurement of IT capacity. A developer may install cooling and power infrastructure ahead of demand to attract tenants, or may phase energisation across buildings that appear structurally identical. Treat these signals as probabilistic indicators, not confirmed capacity figures.
Thermal infrared: distinguishing live buildings from shells
An operational data centre rejects heat continuously. Even with the most efficient cooling systems, a 100 MW facility dissipates roughly 100 MW of waste heat to the environment. That thermal signature is detectable from orbit, though the spatial resolution of available thermal sensors limits interpretation.
ECOSTRESS, at approximately 38 m × 69 m pixels, can identify buildings within a campus that are thermally elevated relative to their neighbours. A shell-and-core building with no IT load will be close to ambient temperature; an operational hall will show a persistent warm anomaly, particularly in night-time acquisitions when solar heating of rooftops is absent. The method works best when comparing multiple buildings on the same campus, using cooler neighbours as a local control. It does not work well for single-building facilities or when cloud cover interrupts the acquisition sequence.
Landsat TIRS provides a longer historical baseline but at 100 m resolution cannot separate individual buildings on a dense campus. It is more useful for detecting the aggregate thermal footprint of a campus becoming active over a 12-to-24-month period.
What the archive tells you that current imagery cannot
Commercial very-high-resolution archives now extend back to 1999 for some providers. For data centre campuses in established markets, this means a site's full construction history is often recoverable, including groundbreak date, construction pace, and the sequence in which buildings came online. That archive intelligence is useful for calibrating how long a developer's typical construction cycle runs, which in turn improves forecasting for sites currently under construction.
Archive analysis also reveals planning-to-completion slippage. A campus that broke ground three years ago but shows no cooling infrastructure yet is telling you something that a planning-permission database alone cannot. Combining satellite observation dates with publicly filed planning documents, grid connection applications and corporate announcements produces a richer picture than any single source.
Honest limits: what orbit cannot see
Internal fit-out is invisible. Server density, power utilisation effectiveness, actual IT load drawn, and the identity of tenants occupying individual halls are not observable from any currently operational civil satellite. A building that has passed every observable milestone, cooling towers installed, substation energised, roof complete, may still be months from generating revenue if fit-out has not begun.
Cloud cover interrupts optical acquisition. In persistently overcast markets such as northern Europe or parts of Southeast Asia, gaps of two to four weeks between usable optical acquisitions are common in winter months. Synthetic aperture radar can detect structural changes through cloud but does not provide the visual interpretability needed to count cooling units or identify transformer equipment. Thermal infrared is similarly cloud-limited.
Very-high-resolution commercial tasking is not free, and systematic monitoring of dozens of campuses across multiple geographies requires a deliberate data budget. Prioritising sites by investment materiality is a practical necessity rather than a methodological choice.
Turning observations into investment-grade intelligence
The analytic workflow combines change detection across a time series of high-resolution images, manual or machine-assisted feature extraction for cooling and substation infrastructure, and thermal anomaly scoring from ECOSTRESS or Landsat TIRS. The output is a per-campus construction stage classification updated on a defined cadence, typically monthly for routine monitoring and within 48 hours when a specific event, such as the arrival of large power transformers, is flagged.
Satellize structures this kind of monitoring as a recurring analytics layer delivered as GIS-compatible files or structured data feeds, which clients can ingest directly into their own investment models. The methodology is consistent with the approach used in the Tonga crop-estimation programme: define observable physical proxies for the quantity of interest, be explicit about what those proxies do and do not measure, and update on a schedule that matches the decision cadence of the client.
For energy planners, the same dataset answers a different question: when will new grid load materialise, and at which substations? For competitive intelligence teams, the question is which developer is building fastest and in which markets. The underlying observations are identical; the analytical frame changes.
Typical figures
| Best optical spatial resolution | 30 cm (WorldView-3 panchromatic, Pléiades Neo) |
| Optical revisit (tasked) | 1 to 4.5 days depending on sensor and latitude |
| Thermal infrared resolution | 38 m × 69 m (ECOSTRESS); 100 m (Landsat TIRS, resampled to 30 m) |
| Thermal revisit | 3 to 5 days (ECOSTRESS, ISS-dependent); 16 days (Landsat 8/9) |
| Minimum detectable structure | Individual cooling tower units and transformer bays visible at 30 cm resolution; rooftop mechanical plant distinguishable at 50 cm |
| Archive depth | Commercial VHR from approximately 1999 (DigitalGlobe/Maxar); Landsat thermal from 1982 |
| Delivery latency (tasked acquisition to product) | 24 to 72 hours for routine; same-day feasible for urgent tasking on some platforms |
| Delivery formats | GeoTIFF imagery, GeoJSON change polygons, CSV stage-classification tables, PDF milestone reports |
| Cloud limitation | Optical and thermal both cloud-blocked; persistent overcast can cause 2 to 4 week gaps in northern European and tropical markets |
Analytics Satellize can run
| Construction stage classification | Manual and semi-automated change detection across VHR image time series; stage taxonomy based on observable physical milestones (groundwork, frame, roof, cooling, substation) | Monthly GeoJSON layer with per-building stage label and confidence rating; PDF summary report |
| Cooling infrastructure inventory | Object-based image analysis at 30 cm resolution to count and classify cooling tower units and rooftop mechanical plant | Structured CSV with unit counts, estimated footprint area and installation date range per campus |
| Substation construction progression | Sequential VHR image interpretation tracking civil works, equipment arrival and transformer installation | Timeline chart and GIS polygon layer; alert on transformer delivery event |
| Thermal activity score | ECOSTRESS and Landsat TIRS anomaly detection comparing building surface temperatures to local ambient and to shell-and-core neighbours on the same campus | Per-building thermal status flag (ambient / elevated / strongly elevated) updated each available acquisition; time-series chart |
| Construction pace index | Footprint area growth rate calculated from polygon digitisation across monthly VHR acquisitions; compared against developer historical benchmarks from archive | Pace index score and projected completion date range; updated monthly |
| Multi-campus supply pipeline dashboard | Aggregation of per-campus stage classifications across a defined market geography; weighted by estimated building footprint area | Interactive GIS dashboard or structured data feed for ingestion into investor supply models |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.