Scenarios · 2026–2032 · a model, not a forecast
Irish data centres in 2032: four possible futures
By 2032 Irish data centres could use between 6.7 and 27.6 TWh of electricity a year, 20–50% of metered electricity, against 7.7 TWh (23.2%) in 2025 (CSO). Even with no new grid connections, use reaches 12.9 TWh as contracts signed before 2021 fill. StackPath scenario model, 10 October 2026: scenarios, not forecasts.
Updated 10 October 2026 · data build 2026-10-10 · how we know · corrections
The four futures Scenario
Data centres' share of Ireland's metered electricity in 2032 · the tick on each bar is 2025, measured (23.2%)
Frozen. Nothing new connects to the grid. Operators fill contracts signed before 2021, mostly with denser AI racks on existing Dublin campuses.
How it gets there
- 2026Contracts signed before 2021 fill up: +595 MW of average load by 2032
- —No new grid connections for data centres
Everything Frozen assumes
Legacy (pre-2021) contracts only: ~600 MW of the 1.0-1.4 GW unused headroom fills by 2032, mostly AI densification in existing Dublin campuses. No new connections under CRU/2025/236; offshore merchant projects slip past 2032.
- Legacy contract holders (Amazon, Microsoft, Meta and pre-2021 colocation) fill ~600 MW of 1.0-1.4 GW unused headroom, mostly by densifying with AI racks (12.87 TWh by 2032).
- The grid copes with a DC peak ~1,543 MW in 2032 (3 MW above EirGrid's AIRAA median path) without new firm-capacity obligations, because legacy contracts are exempt from bring-your-own-power.
- The electricity sector still overshoots its 2026-30 budget (27.0 Mt vs 20 Mt) because renewables build stays at its recent pace.
- Little new AI capacity is built in Ireland: new hyperscale AI campuses go elsewhere (recent announcements: Spain, Finland, Denmark, Portugal).
Measured: CSO MEC02, to 2025 · scenarios: StackPath model, Oct 2026 · shaded band = the range of the four futures · click a line to pick it
What Frozen would mean by 2032
1.7× what they used in 2025, and 1.4× what every Irish home used that year.
CSO MEC02, 2025; scenario model
About 3 in every 10 units metered in Ireland, against 23.2% in 2025.
CSO MEC02; other demand grows 1% a year (assumption)
No load above what EirGrid already expects, so no extra call on spare capacity. Data centres would burn 17.1% of the gas used that day.
GNI Winter Outlook 2025/26; CRU202657; scenario model
35% over the sector's ceiling. Every future overshoots, the lowest by 15%: data centres decide by how much. Their own share: 7.1 Mt.
Sectoral Emissions Ceilings 2022; EPA 2025; scenario model
None needed: nothing new connects, and contracts signed before 2021 are exempt from the 80% rule.
CRU/2025/236; PlanSight permitted pool; scenario model
53% of the 2020 peak (14,300 direct construction jobs).
DETE/KPMG, Jun 2026 (55 jobs per MW built); scenario model
Where the growth comes from Scenario
In Managed growth, 50% of the increase (5.2 of 10.5 TWh) comes from contracts signed before 2021, which need no new connection. In Frozen, all of it does.
TWh in 2032. The model adds average load (MW) in each segment each year; we split each future's change from 2025 in proportion. Energy parks and AI campuses are new connections, so each must bring its own power and new renewables under CRU/2025/236.
Year by year: Frozen Scenario
Firm power: the peak, and who covers it
The peak reaches 1,543 MW by 2032, 3 MW above EirGrid's expected path. Contracts signed before 2021 don't have to bring their own power.
EirGrid AIRAA 2026–35 (959 MVA in 2024, 1,870 MVA median in 2035); CRU/2025/236
Gas on a 1-in-50 winter day
Data centres would burn 17.1% of the gas used on that day by 2032 (12% in 2026). No call on spare capacity: load stays within EirGrid's expected path.
GNI Winter Outlook 2025/26 (325.5 GWh/day); CRU202657 (10% spare, best estimate)
The 80% renewables rule
Not needed: nothing new connects, and contracts signed before 2021 are exempt.
No requirement in this future.
CRU/2025/236 s.3.5; PlanSight permitted pool; EirGrid ORESS 1 yields
Electricity-sector CO₂ against its ceiling
27 Mt over 2026–30 against a 20 Mt ceiling, 35% over. Data centres' grid share is 7.1 Mt, plus 0.54 Mt from their own gas plant.
Sectoral Emissions Ceilings 2022; EPA 2025 (6.3 Mt); 2031–35 ceilings not yet set
Investment each year
€22bn to grow over 2026–32, 50% of it IT hardware, nearly all imported. Refreshing today's servers adds €17.3bn.
DETE/KPMG building cost per MW; hardware, grid and generation costs per MW unchecked
Jobs
Construction peaks at 7,615 direct jobs. Direct operating jobs reach 4,888 by 2032.
DETE/KPMG, Jun 2026: 55 construction and 1.8–2.1 operating jobs per MW; 2020 peak 14,300
What breaks first Scenario
The carbon ceiling breaks first, in every future, from 2026, whatever data centres do. After that: Decline nothing else breaks; Frozen nothing else breaks; Managed growth passes the 2020 construction peak in 2029; AI hub passes the 2020 construction peak in 2028, then the 80% renewables rule in 2029, then spare gas capacity in 2030.
| Limit | Line | Decline | Frozen | Managed growth | AI hub |
|---|---|---|---|---|---|
| Electricity-sector CO₂, 2026–30 | 20 Mt ceiling | 22.9 Mtcrossed 2026 | 27 Mtcrossed 2026 | 27.2 Mtcrossed 2026 | 29.2 Mtcrossed 2026 |
| Spare gas capacity used, 1-in-50 day | 100% | 0% | 0% | 76% | 210%crossed 2030 |
| Permitted unsubsidised wind and solar contracted | all of today's pool | 0% | 0% | 29% | 98% |
| 80% rule, worst-year shortfall | none | met | met | met | 0.39 TWh shortcrossed 2029 |
| Construction workers at the peak | 2020 peak, 14,300 | 3,385 | 7,615 | 21,577crossed 2029 | 38,500crossed 2028 |
| Own power plant to build by 2032 | no set limit | 0 MW | 0 MW | 1,062 MW | 2,442 MW |
2032 values, except CO₂ (2026–30 total). "Crossed" is the first year a limit is passed: for CO₂, when cumulative emissions pass the cumulative ceiling. The 2020 construction peak is not a legal limit; it is the most the industry has done before.
What would tell us which future we're on Editorial judgement
None of these futures is a forecast. These dated decisions and releases will show which one Ireland is heading for.
| When | What | What it would tell us |
|---|---|---|
| pending (no hearing date published) | High Court judicial review of CRU2025236 (LEU connection policy): FIE, Friends of the Earth Ireland, ClientEarth v CRU | If the CRU's rules are quashed, new sites have no route to a grid connection, which leaves Frozen or Decline. If they are upheld, the growth futures stay open. |
| pending (consultation closed 5 Aug 2026; no date set) | CRU decision on interruptible gas connections above 50 MWth (CRU202657) | Sets the terms on which new sites' own gas plant connects. Managed growth needs 1,062 MW of own plant by 2032, AI hub 2,442 MW. |
| 21 September 2026 overdue | ACP decision on Oriel Wind Farm offshore wind (375 MW), case OA15.319799: 19 days past the statutory date, 28.5 months in planning | One of only two offshore wind farms that can sell to data centres. Managed growth needs it from 2031, AI hub from 2030. |
| 3 December 2026 | EU AI Gigafactories call (EuroHPC JU) closes, 17:00 CET | An Irish site in a bid would be an early sign of the AI hub case; no Irish site would make it less likely. |
| early December 2026 | Expected ACP decision on Arklow Bank Wind Park 2 (SSE) | The second offshore wind farm that can sell to data centres, with Oriel. Managed growth needs it from 2032, AI hub from 2031. |
| 25 March 2027 | 2030/31 T-4 capacity auction runs (bids close 10:00) | Shows whether gas plant linked to data centres wins capacity contracts: an early sign that new connections are coming. |
| early July 2027 | CSO Data Centres Metered Electricity Consumption 2026 (annual release) | The first hard number. About 8.0 TWh, Decline; 8.5 TWh, Frozen or Managed growth; 8.6 TWh, AI hub. The futures barely differ in 2026; they split from 2028. |
| Q2 2028 (assumed) | Celtic Interconnector (700 MW, Cork to Brittany) expected in service | Assumed in every future from 2028 (about 2 TWh a year of imports, unchecked). A delay means more gas burned and more CO₂ in all four. |
Dates: StackPath watch list (ACP case pages, CRU, SEMO/EirGrid, EuroHPC, CSO), as of 2026-10-10 · "What it would tell us" is our reading
All four, side by side, with 2025
| 2025 | Decline | Frozen | Managed growth | AI hub | |
|---|---|---|---|---|---|
| Data-centre electricity, 2032 | 7.7 TWh | 6.7 TWh | 12.9 TWh | 18.1 TWh | 27.6 TWh |
| Share of metered electricity | 23.2% | 19.8% | 32.2% | 40% | 50.4% |
| Average load | 875 MW | 765 MW | 1,470 MW | 2,070 MW | 3,145 MW |
| Installed IT capacity | 1.54 GW | 1.34 GW | 2.46 GW | 3.38 GW | 4.85 GW |
| Own firm power required (new sites) | 0 MW | 0 MW | 0 MW | 1,062 MW | 2,442 MW |
| Share of a 1-in-50 gas peak day | — | 9.7% | 17.1% | 22.6% | 30.7% |
| Spare gas capacity used | — | 0% | 0% | 76% | 210% |
| Electricity-sector CO₂, 2026–30 (ceiling 20 Mt) | 6.3 Mt in 2025 | 22.9 Mt | 27 Mt | 27.2 Mt | 29.2 Mt |
| Data centres' grid CO₂, 2026–30 | — | 5 Mt | 7.1 Mt | 7.4 Mt | 9.1 Mt |
| Renewables shortfall, worst year (full 80%) | — | 0 TWh | 0 TWh | 0 TWh | 0.39 TWh |
| Permitted unsubsidised pool contracted | — | 0% | 0% | 29% | 98% |
| Growth investment 2026–32 (excl. refresh) | — | €1.5bn | €22bn | €48.7bn | €105.8bn |
| Imported equipment 2026–32 (incl. server refresh) | — | €17.5bn | €32.1bn | €48.8bn | €91.4bn |
| Rates a year, 2032 | €72.6m (est.) | €60m | €128m | €183m | €271m |
| Direct operating jobs, 2032 | ~3,300 (2024) | 2,875 | 4,888 | 6,826 | 9,471 |
| Peak construction jobs | 14,300 (2020) | 3,385 | 7,615 | 21,577 | 38,500 |
2025 column: CSO MEC02 (electricity, share); DETE/KPMG (installed IT, 2025; jobs, 2024; 2020 construction peak); EPA provisional (CO₂); StackPath rates estimate from the Tailte Éireann valuation list. Average load = CSO GWh ÷ 8,760 hours. Scenario columns: StackPath model, Oct 2026.
What's checked and what's assumed
58 inputs: 27 checked, 6 derived, 18 assumed, 10 unchecked (3 are part one, part another). The unchecked ones are mostly costs per MW and can move the € results by 30% or more.
How the model works
- Load is modelled as annual-average MW added per segment (legacy fill, energy parks, AI campuses) and converted to IT MW with load-per-IT-MW factors; additions count for the whole year they are listed in.
- Share = DC TWh / (non-DC CSO metered demand grown at the non-DC rate + DC TWh). System demand uses Ember's basis (incl. losses).
- Firm capacity: DC peak = avg x 1.05, compared with a straight line between EirGrid AIRAA's 959 MVA (2024) and 1,870 MVA median (2035). New connections must bring dispatchable MW = MIC (CRU/2025/236); legacy contracts are exempt. 'systemFirmGapMW' = DC peak above the AIRAA path not covered by BYO plant.
- Supply balance: 2025 Ember actuals + baseline onshore/solar additions + DC CPPA renewables + DC-eligible offshore + imports (+Celtic from 2028) + hydro/bio + other fossil; gas is the residual (floor 4 TWh). Electricity-sector CO2 = gas TWh x a factor calibrated so 2025 matches EPA's 6.30 Mt.
- DC grid emissions are attributional (DC share of system demand x sector emissions). On-site: BYO gas MW x run hours / 42% x 56.6 tCO2/TJ, plus legacy licensed gas sites (assumed 500 h/yr from 2028).
- 80% rule: each energised cohort's requirement ramps linearly over 6 years (glide) and is also shown at the full 80%. Available = permitted onshore pool contracted at a scenario rate + DC-eligible merchant offshore (Oriel, Arklow Bank 2) at EirGrid's deemed yield.
- Peak day: DC load on a 1-in-50 day met 85% by gas at 50% efficiency, compared with GNI's 325.5 GWh/day. 'pctOfGniHeadroomUsed' counts only DC load above EirGrid's AIRAA median path, assuming GNI's forecast already embeds that path (assumption), against CRU202657's 10% best-estimate headroom.
- Investment: new IT MW x KPMG building cost/MW; IT hardware per MW (unchecked); grid per MW of new MIC; BYO firm plant; CPPA onshore/solar; offshore capex spread over 3 years before COD. Refresh of the 2025 fleet shown separately.
- Rates = index 2025 estimate + new IT MW x KPMG EUR 60k/MW. Jobs = KPMG FTE/MW (operational 1.8 hyperscale / 2.1 mix; construction 55 per MW built in the year). Imports = layer capex x assumed import shares.
Caveats
- This is a scenario model, not a forecast. Numbers marked 'unchecked' (hardware, grid, generation and renewables capex per MW; Celtic; solar yield) are general knowledge and can move results by 30%+.
- Average MW is not IT MW or MIC. The 2025 fleet draws 875 MW on average against ~1,543 MW of installed IT and ~2,000-2,400 MVA contracted.
- The onshore pool is shared: the same permitted wind/solar can win RESS instead, and PlanSight covers only ~54% of RESS projects, so the real pool may be larger.
- The supply balance is energy-only (TWh). It ignores hourly matching, dispatch-down, network constraints and interconnector availability in stress events.
- Electricity ceilings for 2031-35 are not set; 3.0 Mt/yr is an assumption scaled from the national carbon budget. Whether SEM-registered BYO plant counts in electricity or commercial buildings depends on registration.
- IT hardware is the largest and most uncertain line. Comext shows EUR 19.7bn of compute kit imported in the 12 months to Jul 2026 (EUR 4.75bn net of exports); much is re-exported, and customs value is not installed value.
- KPMG's capex and jobs per MW come from a survey and its input-output model; construction FTE/MW is 'inferred'. Rates per MW is KPMG's figure; this index's own estimate implies ~EUR 47k/MW IT.
- Cross-checks: Frozen's 2026-30 electricity-sector total (27 Mt) is above EPA's WAM projection (24.5 Mt) because baseline renewables build here follows recent pace, not plans. Modelled hardware-plus-equipment imports in Frozen (~EUR 4.6bn/yr) are close to Comext's 2025 net compute-kit imports (EUR 4.5bn).
- KPMG's 55 construction FTE/MW reflects past builds; AI campuses put more of their capex into imported hardware, so jobs per EUR invested will be lower.
- Comparisons with Spain (AWS EUR 33.7bn), Finland (Google EUR 13bn) and Portugal (Microsoft US$10bn) use announced totals over different periods.
Model built on this index's checked data (CSO MEC02, Ember, EirGrid AIRAA 2026-35, CRU/2025/236, CRU202657, GNI, EPA, Sectoral Emissions Ceilings 2022, DETE/KPMG 'The Value of Data Centres to Ireland' Jun 2026, Tailte Eireann, Eurostat Comext) plus labelled assumptions. Inputs: ireland.json built 2026-10-06
Questions people ask
- How much electricity will Irish data centres use in 2030?
- Between 7.2 and 19.8 TWh in 2030 across our four scenarios, against 7.7 TWh measured in 2025 (CSO). EirGrid's forecast, as cited by the CRU, is 14.6 TWh by 2034. Scenarios follow from stated assumptions; they are not predictions.
- Will data centres use half of Ireland's electricity?
- Only in the stretch 'AI hub' scenario, where they reach 50.4% of metered electricity by 2032. With no new connections ('Frozen') they reach 32.2%, because grid contracts signed before 2021 still have room to fill. They used 23.2% in 2025 (CSO).
- Why does data-centre use keep rising if new connections are frozen?
- Operators hold about 2,000–2,400 MVA of grid contracts (EirGrid) but drew about 875 MW on average in 2025 (CSO). Filling part of that headroom takes use to 12.9 TWh by 2032 without a single new connection, in our Frozen scenario.
- Can Ireland stay within its electricity carbon ceiling?
- Not in any of our scenarios, which assume renewables keep being built at their recent pace: the electricity sector emits 22.9 to 29.2 Mt of CO₂ over 2026–30 against a 20 Mt ceiling. Data centres change how far it overshoots, not whether it does.