Grid studies for datacenters

Datacenter grid connection studies

A connection offer in a congested zone comes with restrictions written as rules and tables. We turn them into an hourly power limit for your site, run your load against it, and size what covers the shortfall.

5
Datacenter sites studied
30 to 150 MW
Connection size per site
2 countries
Netherlands and Belgium
8760
Hourly limits per site, per year

What a grid connection study answers

Four questions decide whether a site works, and none of them are answered by the headline capacity on the connection offer.

The five steps of a datacenter grid connection study Connection terms become an hourly power limit; the limit is run against a datacenter load model with IT load and cooling separated; that produces an hourly energy deficit and a congestion heatmap; the deficit sizes battery storage and back-up generation against their permit limits. 1 Connection terms Elia EDS parameters or NL congestion research 2 Hourly power limit per site and per build-out phase 3 Datacenter load model IT load and cooling, real weather years 4 Hourly deficit energy the site cannot draw, plus heatmap 5 Back-up sizing BESS, gas or diesel, against permit limits How a connection study is built Every step is hourly, so the answer holds per site and per year rather than as an annual average.
The five steps of a connection study. Every step is hourly, so the answer holds per site and per build-out phase rather than as an annual average.

How often the connection is restricted

The grid operator sets out the restriction regime for a flexible connection in its own documents, per zone, per direction and per build-out phase. We convert that into an hourly power limit for the connection point, for every year of the build horizon, so the restrictions become something a load model can be run against.

What the site actually draws

A datacenter load is not flat. IT load carries a day and week rhythm, and cooling tracks ambient temperature, so the site peak lands in summer and moves between weather years. Modelling IT load and cooling separately changes the deficit the restrictions produce, which a flat annual average hides.

What covers the deficit

Battery storage, gas gensets and diesel gensets, each sized against the restricted hours rather than against nameplate capacity. Permit conditions decide as much as cost here: a diesel genset is cheaper to install but capped in running hours, so it stops being an option once the restrictions get long.

Which candidate site to take

Candidate connection points ranked by residual capacity and expected restrictions over the build horizon, so siting is decided on the hours the grid will allow rather than on queue position alone.

Grid connection capacity is not IT capacity

PUE, power usage effectiveness, is the ratio of total facility power to IT power. A site drawing 78 MW at the meter to run 60 MW of servers has a PUE of 1.30. The remaining 18 MW is cooling, UPS losses, fans and lighting: real load that the grid connection has to carry and that earns nothing.

That makes PUE a capacity question before it is an efficiency question. A grid connection is sized in total facility power, so the IT capacity you can install behind it is the connection divided by the design PUE. Two developers with the same 100 MW connection can end up 24 MW apart in revenue earning capacity.

IT capacity available behind a 100 MW grid connection at design PUE from 1.10 to 1.50 A grid connection carries total facility power, so the installable IT capacity is the connection divided by the design PUE. At a PUE of 1.10 a 100 MW connection supports 90.9 MW of IT. At 1.50 it supports 66.7 MW, a difference of 24.2 MW of revenue earning capacity on the same connection. 100 MW grid connection PUE 1.10 90.9 MW IT 9.1 MW overhead PUE 1.20 83.3 MW IT 16.7 MW overhead PUE 1.30 76.9 MW IT 23.1 MW overhead PUE 1.40 71.4 MW IT 28.6 MW overhead PUE 1.50 66.7 MW IT 33.3 MW overhead 24.2 MW of IT capacity, decided by design PUE alone IT load Cooling and auxiliaries
Installable IT capacity behind a 100 MW grid connection. The connection carries total facility power, so IT capacity is the connection divided by the design PUE.

The connection has to carry the peak hour

Design PUE is the figure at the design ambient temperature, and in the Dutch and Belgian climate that temperature is reached on a handful of summer afternoons. Averaged over the year the same site runs at a PUE closer to 1.19, because the temperature dependent part of the cooling load is small in winter.

Which cuts both ways. Annual energy is set by the average, so an annual figure understates what the connection must be able to deliver. The connection has to cover the peak hour, and the peak hour sits above the design PUE, not below it.

Annual datacenter load profile: flat IT load with a temperature driven cooling layer Daily mean site load for a 60 MW IT installation at a design PUE of 1.3. IT load is nearly flat all year. Cooling and auxiliaries swing with ambient temperature, so total site load peaks in August and troughs in winter. 0 15 30 45 60 75 90 MW 80 MW grid connection peak hour 79 MW Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec IT load (60 MW nominal) Cooling and auxiliaries Peak hour of each day
Daily mean site load for a 60 MW IT installation at a design PUE of 1.30. IT load barely moves across the year. Cooling and auxiliaries follow ambient temperature, so the peak hour lands in summer and comes within 1 MW of the 80 MW connection.

This is also why a restriction on the connection does not translate into lost IT capacity one for one. The hours a restriction bites, and the load the site was drawing in those hours, decide the cost. Both are hourly questions, which is where the study starts.

Netherlands and Belgium

The two grids publish their restrictions in different forms, so the first step of the study differs by country. What comes out is the same hourly limit.

Belgium: Elia EDS

Elia states the flexibility attached to a connection through EDS parameters: the share of time under preventive flex, the share under curative flex, the share of active energy affected and the flexible energy per year. We turn those bands into an hourly profile per site and per build-out phase.

Elia EDS analysis ->

Netherlands: congestion research and BirdFlow

TenneT and the regional operators publish congestion research per area, which sets out whether capacity exists at all. For the hours themselves we run BirdFlow, our power-flow model of the Dutch transmission grid, which gives the expected hourly restrictions per substation from 2027 onwards.

BirdFlow ->

Reference: five sites for a global datacenter developer

The studies were commissioned by a datacenter developer active worldwide, covering five planned sites in the Netherlands and Belgium between 30 and 150 MW. Site names, locations and results stay with the client. What we can describe is the method and the shape of the answer.

For each site we translated the grid operator's restriction terms into hourly power limits per build-out phase, ran a datacenter load model with separate IT and cooling components against them, and reported the resulting energy deficit as a share of the full connection. Across the sites and phases, restrictions ran from 2% to 34% of the year. The deficit was then covered with combinations of battery storage, gas gensets and diesel gensets, compared on capital cost and on the permit conditions attached to each.

The outcome per site was a ranking, a set of congestion heatmaps showing when in the year and the day the restrictions fall, and the back-up capacity each site needs to stay at full load.

Get the restriction profile for your connection point

Send us the location, the capacity you are requesting and the connection documents you have, and we come back with the expected hourly restrictions and what it takes to cover them.