Where in Europe is clean power cheapest?

Candidate locations ranked by the cheapest 24/7 carbon-free delivered power, each using its best clean resource: firm geothermal (Iceland) or big hydro (Norway, Sweden, the Alps, Greenland) where they exist; everywhere else a gas-free solar + wind + battery build, firmed by green hydrogen or pumped storage (PHS). (2030, firm · real ERA5 weather.)

Firm clean baseload has the lowest delivered cost. Nordic/Alpine hydro (~$46/MWh) and Icelandic geothermal (~$63) cost less than every build-it-yourself sun-and-wind site and sit below gas.

Both firmings are shown for every sun+wind site, because which one a site can use is itself geographic: off-river PHS needs terrain with head (availability from the ANU Global Pumped Hydro Atlas), while H₂ works anywhere. Where strong sun+wind co-locates with pumped-storage terrain — the Iberian sierras and mountainous islands (Tarifa, Sines, Sicily, Crete, Gran Canaria) — PHS firms far cheaper than H₂ (~80% round-trip efficiency vs ~35%, so far less overbuild is wasted). Flat sites (Jutland, the Dover Strait) have no cheap PHS and fall back on dearer H₂; the open circle on each bar marks the firming not chosen.

Two caveats. (1) Wind and PHS terrain don't always co-locate: Romania's wind is on the flat Black Sea coast while its PHS is inland in the Carpathians, so its site is H₂-firmed; Switzerland is wind-poor even with world-class PHS — its lowest-cost clean resource is conventional hydro. (2) In carbon-priced Europe, partial-gas isn't the cheap option: "75% RE + gas" undercuts fully-clean only at the flat H₂-firmed sites; where PHS makes clean firming cheap, 100% zero-carbon is cheaper.

Each sun+wind figure reflects the exact ERA5 grid cell at the site's coordinates, so very localized wind regimes (e.g. the Tarifa jet) can be under-captured — treat the ranking as directional. Geothermal/hydro costs: IRENA 2023 installed costs ($4,589/kW geothermal, $2,806/kW hydro); pumped storage: NREL ATB / DOE-PNNL (RTE 0.80, ~50-yr life). From tools/build_eu_siting.py.

Scanning the whole continent

The nine sun+wind candidates above were chosen by hand. To remove the hand-selection, the same gas-free build (solar + wind + battery + self-made hydrogen) was computed at every ~1° land cell of Europe — 978 cells — on real hourly ERA5 weather (2019–2021), with each cell's solar and wind costs re-anchored to its real capacity factors. EU technology costs are used everywhere, so the map isolates geography: resource quality and weather structure, not national policy. Median cell: ~$156/MWh at 2030.

The raw map is led by wind, not sun. The best cells on raw cell weather (~$95–101/MWh) trace the North Sea and Baltic coasts and islands — Danish and Pomeranian shores, the Estonian and Swedish Baltic islands, Orkney and the Faroes — where a ~0.5 wind capacity factor costs less than Mediterranean sun. The expensive interior band — and the very worst cells, the sheltered Scandinavian inland valleys (~$245) — is what a datacenter pays for being far from wind. Firm hydro (~$46) still costs less than every cell on the map.

Sea wind must be priced at offshore capex, which changes the ranking. The raw top cells above are only 20–40% land: their measured wind is North Sea/Baltic sea wind, and the map prices it at onshore capex. Re-pricing every cell below 60% land at European fixed-bottom offshore costs ($110/MWh levelised at CF 0.50 — the UK AR7 clearing level — with offshore's slower ~10% learning) lifts the raw top-10 from ~$95–101 to ~$158–169/MWh. The build-it-yourself ranking is led instead by the cheapest mostly-land coastal cells at their onshore pricing (54°N, 15°E ($119), 51°N, 4°W ($119), 55°N, 6°W ($120), 47°N, 2°W ($120)) — and the north–south gap largely disappears: the cheapest mostly-land southern cells (43°N, 9°W — windy Galicia — at $122; 43°N, 3°E at $126) sit within the scan's screening noise of the northern top cells. What survives every pricing: coastal wind has a lower cost than inland sun, the sheltered continental interior is the most expensive, and no DIY cell approaches firm hydro (~$46) or the PHS-firmed southern sites above ($70–93). (Waters needing floating turbines — the Norwegian Trench — would be ~2.4× dearer still; not modelled.)

Can two capacity factors predict the price?

Almost — and the gap is the informative part. A transparent least-squares fit on mean solar and wind capacity factor alone predicts the dispatch model's cost with R² 0.94 (typical error ~$5/MWh) on held-out cells; adding simple weather-structure statistics (the depth of the worst 5- and 14-day sun+wind drought, sun–wind correlation, winter-solar share) improves it to R² 0.96 (~$4/MWh). Both the formula and its miss are both informative: annual averages carry most of the signal, and what they miss is exactly the multi-day Dunkelflaute a 24/7 datacenter must ride through. The full coefficients are in output/eu_scan_results.json.

The cheapest cells found by the scan

The same build across 14 markets — Europe vs the US

For contrast with the cheap-gas US, the same firm off-grid build is computed at seven large EU countries and seven US states — the biggest data-center markets in each region — on real ERA5 weather (2015–2025, 11 years), one grid point per location, every real year a dispatch sample (so the curves carry real year-to-year variability). Within a region only the renewable resource differs — gas, carbon and technology costs are the region default — so this isolates how much where you build, and whether you add a wind park, move the cost.

Does adding a wind park lower the cost? (gas-backed)

Solar is quick to permit; a wind park is a far bigger siting undertaking. Both builds are optimised to the same renewable target at each site — the most a solar + battery + gas system can reach without wind (~55–68%, shown in each panel) — and the wind build may add a wind park only if that lowers cost. So the blue line is never above the no-wind orange line: it dips below where wind lowers the cost (the United Kingdom; Texas, Iowa) and merges with it where wind is too weak to add (Arizona, California, Italy — ~2% capacity factor at the modelled point, so the optimiser builds almost none). A reading note: each location is sampled at one representative point — usually its datacenter hub (Ashburn, Silicon Valley, Phoenix, Milan) — not the country's windiest terrain. Italy's 0.02 wind CF is real for Milan's becalmed Po Valley; Italy's actual wind fleet, on southern ridgelines, averages ~0.2. Read each row as "a datacenter at this hub", not a national wind verdict. Grey dashed: the gas baseline; purple dash-dot: a small modular (nuclear) reactor — competitive in carbon-priced Europe, undercut by renewables+gas in the cheap-gas US.

Fully zero-carbon: self-made hydrogen, with and without wind

To drop gas entirely, the backstop becomes green hydrogen made from surplus renewables (the small residual bought from the market). Both builds are zero-carbon by construction — solar + battery + hydrogen vs the same plus a wind park — on the same 2015–2025 weather. The gap between the lines is what the wind park buys (2035): United Kingdom ~$55/MWh, Sweden ~$35/MWh, Iowa ~$30/MWh; ~$0 where wind is scarce (Spain, Virginia, Italy, California, Arizona), where the easier-to-permit no-wind build is the cheaper one. The purple dash-dot small modular reactor is the other firm zero-carbon option — often competitive, especially in carbon-priced Europe. Grey dashed: the (emitting) gas baseline, for reference.