Gridcog for data centres

Data centre power modelling: grid connections, storage and onsite generation

A data centre can get power years sooner by accepting a flexible grid connection, whether timed, phased or curtailed in real time, and covering the gap with battery storage, onsite generation or flexible compute. Gridcog models the connection, the load and every onsite asset together, so developers can size each option and see what it costs before investing.

Key points

  • 125 GW

    In Great Britain, demand connection applications grew from 41 GW to 125 GW in under a year, with data centre projects accounting for at least 80 GW (Ofgem, 29 July 2026).

  • 5–7 years

    Grid connections can take five to seven years, against plans to bring new data centres online within two (encoord, December 2025).

  • Ireland

    In Ireland, larger data centres must bring dispatchable generation or storage matching their maximum import capacity to connect (CRU, 16 December 2025).

  • £795m

    In Gridcog modelling of a 100 MW data centre in Slough, a microgrid portfolio with a smaller connection cut 15-year cost from about £1.9bn on full grid supply to £795m.

How can a data centre get power sooner?

By connecting on flexible terms instead of waiting for a firm connection.

A firm connection obliges the network operator to build whatever reinforcement full capacity needs, and that reinforcement causes most of the delay. A flexible connection removes the guarantee for some hours or conditions, so the site connects earlier and usually pays less.

Four ways to connect sooner

Option 01 of 04

Timed or scheduled

Import capped in agreed hours, days or seasons

Grid import allowedIllustrative, not to scale

Notice before reduction

Known in advance

Best fit

Workloads that can be shifted or batched

Option 02 of 04

Flexible now, firm later

Restricted import until reinforcement is done, then firm

Grid import allowedIllustrative, not to scale

Notice before reduction

Known; capacity steps up

Best fit

Space let to tenants on a firm service level

Option 03 of 04

Real-time curtailment (active network management)

Automated control reduces import when the network nears its limit

Grid import allowedIllustrative, not to scale

Notice before reduction

Limited; two minutes to acknowledge in Ireland

Best fit

Sites with significant onsite generation already committed

Option 04 of 04

Bring your own generation

Onsite storage or generation picks up the load when grid supply drops

Grid import allowedOnsite storage or generationIllustrative, not to scale

Notice before reduction

Covers any of the above

Best fit

Mandatory for larger sites in Ireland

Backup generation and batteries are already part of most data centre designs for resilience, and the same equipment can ride through a curtailment event. The real cost is any extra generation or storage needed so compute never drops, plus the fuel and cycling to run it. That cost rises as notice falls.

How do you model a flexible connection before investing?

Apply the connection as a Site Import Schedule, then autosize storage or generation against it and compare scenarios.

Gridcog Energy Analyst Joshua Elliott did this for a 100 MW data centre on an example winter peak timed connection.

Joshua Elliott, Energy Analyst at Gridcog

Joshua Elliott

Energy Analyst

  1. Connection only

    What changes

    Winter peak import limit, no supporting generation

    Result

    Periods of unserved load during evening peaks

  2. Autosized standalone battery

    What changes

    Battery charges at midday within the import limit and discharges at the winter peaks

    Result

    Unserved load removed; peak-time network costs avoided

  3. Autosized battery and solar

    What changes

    Battery discharges in the same intervals, partly from solar

    Result

    Lower grid import costs again

Timed and phased limits are set in the schedule editor; real-time curtailment is uploaded as interval data so its randomness is reflected.

Read flexible grid connections for data centres: GB and Ireland

Want to test your site's connection options?

Model your project in Gridcog

What does a data centre's power strategy cost over its life?

It depends far more on the mix of grid, onsite generation and storage than on any one asset.

In Gridcog modelling of a hypothetical 100 MW data centre in Slough:

A microgrid portfolio with a smaller connection cut 15-year cost from about £1.9bn to £795m

15-year cost by power strategy, hypothetical 100 MW data centre in Slough (£)

Bar chart. Full grid supply: About £1.9bn; Part of the connection swapped for a 50 MW gas genset: About £1.79bn; Microgrid portfolio of solar, wind and storage with a smaller connection: About £795m
15-year cost by power strategy, hypothetical 100 MW data centre in Slough (£)
Power strategy15-year cost
Full grid supplyAbout £1.9bn
Part of the connection swapped for a 50 MW gas gensetAbout £1.79bn
Microgrid portfolio of solar, wind and storage with a smaller connectionAbout £795m
Source: Gridcog modellingGridcog

Should a data centre use a genset or a fuel cell for onsite power?

It depends on the system.

A solid oxide fuel cell (SOFC) is more efficient and cheaper to run, but costs far more to build and can't follow prices the way a genset does. We modelled a 100 MW UK data centre with a 50 MW grid connection, co-located solar and storage, and either a gas genset or an SOFC.

Gas genset

Cheaper to build and follows prices

SOFC

More efficient and cheaper to run

What matters most for your site?

Pick one to see which comes out ahead in our model.The SOFC. About 60% efficient against 40%, and £277m less to run over 20 years in our model.The gas genset. £1,000/kW to build against £5,000/kW for the SOFC.The gas genset. It ramps at 20% per minute against 1%, so it can chase price spikes and back off when solar is strong.

About 40%

Electrical efficiency

Higher is better

About 60%
£1,000/kW

Capex

Lower is better

£5,000/kW
20% per minute

Ramp rate

Higher is better

1% per minute
£1,412m

20-year opex

Lower is better

£1,135m
Purple marks the stronger figure on each measure. Source: Gridcog modellingGridcog

What the model showed

The SOFC cut 20-year opex by £277m but ran like a baseload asset; the genset ramped for price spikes and backed off when solar was strong.

Read SOFC vs gas genset for UK data centre power

What can you model for a data centre site?

The load and its connection, with battery storage, gensets and fuel cells, solar and wind, and flexible compute, on one site or as a microgrid.

Frequently asked questions

Can't see your question?

Get in touch

In Great Britain, yes for most arrangements. Since a 2023 rule change, active network management connections carry a contractual ceiling on annual curtailment hours, with compensation due beyond it. Technical Limits connections, where the congestion is on the transmission network, and Ireland's real-time curtailment arrangements carry no such ceiling and no compensation.

A phased connection lets capacity scale up over time in line with a data centre's build-out and ramp-up. Ofgem's February 2026 call for input on demand connections reported clear developer appetite for phased arrangements, and Ireland's flexible demand connections already convert to firm once reinforcement is complete.

Larger ones do. Under the CRU's December 2025 decision, they must build dispatchable generation or storage matching their contracted capacity, sell it into the wholesale market, and cover 80% of annual demand with new Irish renewables within six years. Meeting this brings exemption from emergency curtailment.

As of August 2026, Australia plans a national framework requiring large data centres to offset their demand. The AEMC's advice proposes certificate and firm-contract obligations, market registration and incentives to flex and co-locate. Behind-the-meter generation and storage would reduce the certificate obligation, and flexibility and storage would reduce the firming requirement.

Yes, for some non-critical workloads. In a five-day trial in December 2025 at Nebius's London data centre, Emerald AI, EPRI, National Grid, Nebius and NVIDIA cut power demand by up to 40% while critical workloads ran as normal. Batch inference and overnight processing are the likeliest loads to shift.

Yes. You can add generators of any fuel type with their operating parameters, including ramp rates, heat rate curves and must-run periods. Model them with a constrained grid connection, solar and battery storage, then compare scenarios on cost, emissions and unserved load in the GB electricity market and 40+ others.

Request a live demo

See Gridcog on your own project.

Tell us a little about what you're modelling and we'll shape the demo around it. Bring your own price curves and assumptions, or use ours.

The team behind Gridcog

Fabian Le Gay Brereton, Chief Executive Officer and Co-Founder of Gridcog

Fabian Le Gay Brereton

Chief Executive Officer & Co-Founder

Previously co-founded Greensense, acquired by ERM Power (now Shell Energy).

Pete Tickler, Chief Product Officer and Co-Founder of Gridcog

Pete Tickler

Chief Product Officer & Co-Founder

Co-founded Greensense and led digital innovation at ERM Power (now Shell Energy).

Genna Boyle, Chief Commercial Officer and Head of EMEA at Gridcog

Genna Boyle

Chief Commercial Officer & Head of EMEA

More than 10 years in energy strategy, trading and renewables, with senior roles at Battery Optimiser and Limejump.