Going fully zero-carbon

Does the build need a wind park?

Solar is modular and quick to permit; a wind park is a far bigger siting and permitting undertaking. Comparing solar + wind + battery against solar + battery only (both firm, gas-backed, 2035): there is a ceiling. The no-wind system tops out near ~68% renewable — nights and multi-day cloud always fall to gas, and a battery cannot shift energy across days — while adding wind reaches ~94%. Below the wall, dropping wind costs little in the sunny US and a clearer premium in Europe. For a moderate renewable target, solar + battery alone reaches the target without wind and is much easier to build; high renewable fractions require wind (or long-duration storage / hydrogen).

Solar + battery + hydrogen (no wind)

To go fully zero-carbon without a wind park, replace the gas backstop with green hydrogen. All the options below are green — they differ only in how you get the H₂. Making it yourself (an electrolyser turning surplus renewables into H₂, a few percent bought) brings a wind-free zero-carbon datacenter to ~$145/MWh (EU) / $130 (US) by 2035; buying it all on the market (no electrolyser) is far dearer ($211 / $192). The self-made wind-free build is only ~$17–26 above the same build with a wind park ($119 / $113) — so the wind park, not the hydrogen, is the smaller lever here. In Europe that with-wind build is the cheapest option of all, since gas there is carbon-priced.

In the chart, the three green bars are the same green hydrogen; the only differences are whether there is a wind park and whether the H₂ is self-made or bought.

What does it cost the landscape?

Per gigawatt of always-on datacenter, the 2030 European gas-free build (6.3× solar + 1.8× wind + batteries + hydrogen) needs roughly:

• ~140–180 km² of solar plant (a ~12–13 km square) — the largest direct land use. In the sunnier US the same build needs only ~73–93 km².
• ~587 km² of wind-park spacing, of which only ~9 km² (~1–2%) is actually occupied by towers and roads — the rest stays farmland or open land.
• In exchange it avoids ~3.0 MtCO₂ every year versus running the same datacenter on gas — and burns no fuel, uses no combustion water, and needs no pipeline.

Two counterpoints. First, this is real land: siting must avoid protected areas (Natura 2000 and national designations), and the scan map's cells are screening averages, not permits. Second, the cheapest clean option of all — reservoir hydro — carries the largest ecological footprint per site: the siting ranking prices hydro at full new-build cost at places where big reservoirs already exist (Norway, Sweden, the Alps); damming new wild rivers carries an ecological cost not captured in the price. The land area is reported per GW so it can be weighed explicitly.

And the water?

Water use splits into two separate ledgers. The power side of this build is essentially water-free: solar, wind and batteries consume almost nothing (panel washing aside), the electrolyser's feedwater (~10–15 L per kg of H₂) is minor at these volumes, and the hydrogen turbine runs only through the rare lulls — while the gas plant the build replaces evaporates roughly 0.8 m³ of cooling water per MWh it generates. The datacenter side is a design choice, not a consequence of going off-grid: evaporative cooling consumes ~1–2 L per kWh of IT load, but dry (closed-loop) cooling cuts site water to near zero for a small efficiency penalty — a common choice at the sunny, dry sites the siting ranking ranks highest (Iberia, the islands), where water is scarcest.

Zero-carbon means zero combustion — not zero footprint

The model's carbon accounting is combustion-scope: the gas-free build burns nothing, so it scores zero. Manufacturing its panels, turbines and batteries still emits. On standard lifecycle intensities, scaled up by this build's own overbuild (the panels behind curtailed energy get manufactured too), the delivered power carries very roughly ~31–58 gCO₂e per kWh embodied in the solar and wind fleet — batteries and the electrolyser add a few grams more — versus ~490 for lifecycle gas. That is a ~88–94% cut, not 100%, and the residual falls further as the manufacturing itself decarbonises.

Overbuild ratios from the model's optimal builds (output/*_results.json); land densities: solar 35–45 MW/km² total plant area, onshore wind ~3 MW/km² spacing with ~1–2% direct occupation (NREL land-use studies; Denholm et al.); CO₂ at the model's CCGT intensity (0.345 tCO₂/MWh, combustion scope). Water: power-plant consumption from Macknick et al. 2012 (NREL); datacenter water-use figures from operator environmental reports (evaporative ~1 L/kWh). Lifecycle intensities: IPCC AR5 Annex III medians (utility solar 48, onshore wind 11, CCGT ~490 gCO₂e/kWh); modern PV supply chains run nearer 25.