Method
How much has the world tested this number?
A capacity figure for a data hall is only as good as what stands behind it. Every number the engine returns carries an evidence class and a provenance digest, and a computed value is never stronger than the weakest input it came from.
| Class | Described as | What has been done to the number |
|---|---|---|
| E0 | assumed | No data: engineering judgement or a library default. |
| E1 | modelled | Computed from a documented model on assumed inputs. |
| E2 | simulated | Numerical simulation (hydraulic, thermal network, time series). |
| E3 | estimated | Model plus review by a qualified engineer against practice. |
| E4 | validated against test data | Compared against lab, rig or vendor test data. |
| E5 | supplied by the client (not measured or surveyed by us) | The customer's own measured trend data, as supplied. |
| E6 | field-validated | Our model reconciled against measured site performance. |
| E7 | observed in operation | An operating outcome observed after implementation. |
The floor a claim must clear
A claim of each kind may not reach a customer below a minimum class. The engine enforces this; it is not a guideline.
- Electrical capacity: at least E3
- Cooling performance: at least E2
- Capital cost: at least E1
- Schedule: at least E1
Calibration status
Model accuracy against instrumented sites has not been established: 0 reconciled observations across 0 sites, covering 0 of 20 model outputs. Every figure in this report is modelled. The uncertainty ranges shown come from input uncertainty propagated through the model, not from the difference between this engine's predictions and site data — no such difference has been recorded yet.
- State
- uncalibrated
- Reconciled observations
- 0
- Sites
- 0
- Model outputs calibrated
- 0 of 20
This page reads the same file the deliverables do (/reference/calibration.json). When the number of reconciled observations changes, this page changes with it.
Four kinds of number, and what each lets you conclude.
- AssumedE0
A library default or a judgement. It is a placeholder, listed in the data request, and the result says it was not supplied.
- E0 assumed
TODAYPresent in every result where an input was not supplied.
- ModelledE1 E2 E3
Computed from a documented model, a simulation, or a model reviewed by an engineer. A calculation on inputs, not an observation of the hall.
- E1 modelled
- E2 simulated
- E3 estimated
TODAYThe class of the engine's output. The claim floors set which class a claim needs before it reaches a customer.
- Measured or testedE4 E5
Compared with test data (E4) or the client's own measured data as supplied (E5). The client's data is carried as supplied; it was not surveyed by us.
- E4 validated against test data
- E5 supplied by the client (not measured or surveyed by us)
TODAYE5 only where a client supplies measured data. No vendor-test comparison is claimed.
- Field-validated or observedE6 E7
The model reconciled against measured site performance (E6), or an outcome seen in operation after the work (E7).
- E6 field-validated
- E7 observed in operation
TODAYNone. The calibration ledger is empty and no outcome has been observed.
The weakest link sets the label. A result built from nine modelled inputs and one assumed one is reported as assumed. Nothing is shown as measured, field-validated or observed until it is, and the boundary between the bands is the boundary between a calculation and a fact.
Market figures used on this site, and how each was checked
These describe the market, not Time to Power’s results. Each record gives the publisher, date, period, geography and definition, and says plainly whether the publisher’s own page was read or only press coverage that quotes it. A figure we could not trace is listed as withdrawn and is not shown anywhere else. Checked on 3 November 2026; market figures go stale, so treat the date as part of the figure.
Confirmed in press coverage; the original was not opened · survey
More than 80% of surveyed operators say their highest rack density is below 30 kW.
- Publisher:
- Uptime Institute, Global Data Center Survey 2025, 2025-07
- Period:
- Survey fielded April to May 2025
- Geography:
- Global
- Definition:
- Share of respondents (owners and operators, more than 800) reporting that the highest rack power density they deploy is below 30 kW.
Limits of the check: Self-reported by respondents. The survey report itself was not opened; the figure is read from press coverage that quotes it as 'more than 80 percent'. The earlier wording on this site, a precise 82%, is not used.
Confirmed in press coverage; the original was not opened · fact
European AI-focused colocation signings reached 420 MW in the first half of 2026, up from 89 MW in the first half of 2025.
- Publisher:
- CBRE, press release, 2026-08-17
- Period:
- H1 2026 against H1 2025
- Geography:
- Europe
- Definition:
- Signings of colocation capacity for AI-focused customers, as tracked by CBRE.
Limits of the check: The CBRE page refused an automated fetch, so the figure is confirmed from three independent reports that quote it. It measures signings (contracts), not capacity energised.
Read at the publisher · fact
Colocation vacancy across the FLAP-D markets was 6.4% in the second quarter of 2026, down from a 2021 peak of 16.9%.
- Publisher:
- JLL, EMEA Data Centre Report, mid-year 2026, 2026-08-03
- Period:
- Q2 2026
- Geography:
- Frankfurt, London, Amsterdam, Paris, Dublin
- Definition:
- Colocation vacancy as measured by JLL. The page read does not define the denominator.
Limits of the check: Other brokers publish different vacancy figures for the same markets and period (a lower French edition of the same report, a higher CBRE Q1 figure), because definitions differ. Treat it as one broker's measure.
Confirmed in press coverage; the original was not opened · forecast
Mainova, Frankfurt's utility, has said new high-power grid connections in the city will probably not be available again until about the mid-2030s.
- Publisher:
- Mainova AG, as reported in the German trade press, 2026
- Period:
- Forward-looking statement
- Geography:
- Frankfurt and the Rhine-Main region
- Definition:
- Availability of new high-power (data-centre scale) grid connections.
Limits of the check: A utility's own forecast, read from press reports and not from a Mainova document. One report gives 2035. The earlier wording on this site, 'before 2030', contradicted it and has been withdrawn.
Withdrawn: no source found · estimate
Wording previously used: Transformer lead times are reported at roughly 130 to 160 weeks or more, depending on the type of transformer.
- Publisher:
- Mixed trade and analyst reporting (Wood Mackenzie and others), 2026
- Period:
- 2024 to 2026
- Geography:
- Mainly the United States
- Definition:
- Order-to-delivery time; the figure differs between standard large power, generator step-up and substation transformers.
Limits of the check: The earlier wording, '160+ weeks, up from 120 weeks in 2024, Data Center Knowledge 2026', could not be traced: no such article was found, and reported 2024 figures are 140 to 143 weeks, not 120. Not shown on any page until a primary source and an equipment class are read. The engine's own library range is a separate, labelled assumption.
Withdrawn: no source found · survey
Wording previously used: Fewer than 4% of facilities can host racks of 100 kW or more.
- Publisher:
- Not found
Limits of the check: No source states this. Nearby published figures differ (about 1% of operators run racks above 100 kW in one Uptime reading; 9% report racks above 50 kW in the 2025 survey). Withdrawn from the site.
Withdrawn: no source found · estimate
Wording previously used: Amsterdam's grid queue runs about ten years.
- Publisher:
- Not found
- Geography:
- Amsterdam
Limits of the check: No source found for this wording. Withdrawn from the site; a grid operator's own congestion reporting would be the place to source a replacement.
Questions
- Is the engine validated against real sites?
- Not yet. The calibration record is "uncalibrated": Model accuracy against instrumented sites has not been established: 0 reconciled observations across 0 sites, covering 0 of 20 model outputs. Every figure in this report is modelled. The uncertainty ranges shown come from input uncertainty propagated through the model, not from the difference between this engine's predictions and site data — no such difference has been recorded yet.
- What does an evidence class mean?
- It is a claim about how much the world has tested a number, from E0 (an assumption or library default) to E7 (an outcome observed in operation). Classes are ordered, and arithmetic takes the minimum: a result is never stronger than the weakest input it came from.
- Can I quote an E0 or E1 figure as a property of a physical asset?
- No. An E0 or E1 figure is assumed or modelled, not measured. If you quote one, say so.