Off/Grid
Guide

How to read this tool.

Every screen answers the same question in a different way: given this weather, this build and this compute load, how often does the data center go dark? Here is the vocabulary, and the order to read it in.

Four steps

The workflow, start to finish.

01

Pick a site

On Sites, hover a row and the globe swings over its country. On Instrument, the left rail selects it directly. Colour is the verdict: cyan viable, amber marginal, red not viable.

02

Set the load

The IT load buttons (10 / 25 / 50 MW) switch the whole scenario. Everything below — LOLP, deficit hours, optimal sizings — recomputes for that load.

03

Dial the build

Turbines, solar and battery sliders move through the simulated grid. The big number is the LOLP of that exact combination; the CAPEX card is what it costs.

04

Compare the cost of reliability

The frontier row shows the cheapest build for each reliability band. Read it right to left: going from 10 % to 0.1 % is where most of the money goes.

Vocabulary

Six terms, and nothing else to learn.

LOLP%
Loss-of-load probability

The share of the 8 760 hours in the year where production plus battery cannot cover the load. 1 % = 88 hours of blackout per year. It is the headline number everywhere in this tool.

LOLP · ref build%
The comparison number

LOLP at one identical build for every site — 40 turbines, 100 MWp of solar, 800 MWh of storage — under a 50 MW IT load. Because the build is fixed, the ranking measures the weather alone. Push the sliders to the maximum and most sites reach 0 %, which is why the index uses this reference instead.

Deficit hoursh / yr
The same number, in hours

LOLP × 8 760. Easier to reason about for operations: how many hours of the year compute has to stop or fall back to a generator.

CAPEXM€
Cost of the build

Turbines × €8 M (6 MW each) + solar × €0.7 M per MWp + battery × €0.25 M per MWh. Land, civil works and the halls themselves are excluded.

CAPEX ≤ 1 %M€
Cost of reliability

The cheapest sizing in the grid that holds LOLP under 1 % at 50 MW. A dash means no sizing we simulate reaches that band — the site is weather-limited, not budget-limited.

PUEratio
Power usage effectiveness

Total power drawn per watt of compute. Derived from a Carnot-limited chiller, so cold sites sit near 1.10 (free cooling) and hot sites climb above 1.30 — which is why Seville needs more turbines than Dublin for the same racks.

Verdict bands

What the colours mean.

  • ≤ 0.1 %≈ 9 h/yr — approaches continuous, latency-sensitive service.
  • ≤ 1 %≈ 88 h/yr — fine for AI training, not for real-time inference SLAs.
  • 1 – 10 %Interruptible batch compute only, with schedule slack.
  • > 10 %The site cannot power itself off-grid at any sizing we simulate.

Several sites reach 0.00 % at their largest sizing — that is a real result, not a missing value. It means wind and storage alone cover every hour of 2023 at that build; the interesting question there becomes CAPEX, not reliability.