The largest underused power source on Earth is also the coldest, quietest place to put a data center.
The ocean carries roughly two hundred times more energy than the world consumes — and we have built almost nothing to catch it.
Wave energy density is concentrated near the surface and decays rapidly with depth. A 1-km stretch of west-facing North Atlantic coastline can carry 40–70 kW per metre of wavefront — comparable to a small wind farm, available on schedules wind cannot match.
The bottleneck has never been resource. It is survivability, mooring, power-takeoff reliability, and the cost of being at sea.
A wave converter has to be sensitive enough to absorb a 1-metre swell and strong enough to survive a 15-metre storm — three orders of magnitude in between.
Three orders of magnitude.
The cost of a wave-energy converter is set not by the average sea state, but by the worst sea state it must survive. Modern designs use active detuning, submergence, feathering and predictive control — tuning the power-takeoff to each incoming wave forecast.
Source: DOE Marine Energy Cost Reduction Pathways, 2025.
Five families of wave-energy converter, each suited to a different sea state and shoreline.
From wavefront to GPU rack, in seven hops.
Co-located generation
24-unit point-absorber array delivers ≈ 18 MW average / 70 MW peak through a single subsea hub. No long export cable — the data hall sits within 2 km of the lead converter.
Ocean as heat sink
Direct seawater cooling at 8–14 °C eliminates evaporative cooling towers, drops PUE below 1.10, and saves ≈ 1.4 ML / day of freshwater versus a comparable inland facility.
Buffered, not islanded
Onshore battery (40 MWh) absorbs minute-scale wave variance and 6-hour tidal beat. The grid covers seasonal lows; surplus is exported during winter swell maxima.
Built like a substation, run like a DC
Modular 6 MW pods, liquid-to-chip cooling, two-tier security, latency < 4 ms to nearest tier-1 metro. Designed for AI training workloads where 24/7 uptime is shaped by SLA, not loss-of-grid.
Specifications for the first SeaPower facility, sized for a single AI-training tenant.
Pelagos-1 · 60 MW IT
Why wave, not wind, for compute
Wave power lags wind by 6–18 hours and storms by 12–24 hours, so a coastal pairing smooths variance dramatically. Where wind alone gives 32 % capacity factor, wave + wind + battery reaches 71 % at our reference site.
The 60 MW step
60 MW is the sweet spot for a single hyperscaler training tenant: enough to host one frontier-class run end-to-end, small enough to permit and finance against a single coastal lease.
Built where data centers won't go
West-facing coasts have low industrial competition for land, abundant grid-injection capacity, and natural environmental envelopes (wind, salt, fog) that a marine-grade facility can absorb without exotic hardening.
A short list of coastlines where wave resource, grid, latency and permitting align.
Cornwall, UK
Bristol Channel and Wave Hub footprint. Existing 33 kV grid takeoff, deep marine industrial heritage, ≤ 6 ms to London-Slough.
North Iberia
Asturias to Galicia. Bimep test site, EU Innovation Fund eligible, complementary to Spanish solar via winter peaking.
Oregon, USA
PacWave South leasehold. Latency tier-1 to Hillsboro-Portland fiber. DOE Powering the Blue Economy partnership pipeline.
Central Chile
Bío-Bío to Valdivia. Among the highest year-round wave power in the world; HVDC corridor under development.
SW Australia
Albany / Augusta. Carnegie test legacy, AEMO grid spare capacity, sub-sea cable to Singapore in planning.
Wairarapa, NZ
Cook Strait corridor. Combines tidal-stream and wave; Transpower 110 kV interconnect within 8 km of coast.
Wave power doesn't beat solar on cost. Co-located AI compute changes the question.
Three levers move wave from outlier to competitive.
1. Captive offtake. A co-located DC pays a power price benchmarked against grid + curtailment, not against wholesale.
2. Capex amortisation. Sharing substation, cooling intake, civil works and permitting between energy and compute halves balance-of-plant.
3. Cooling credit. Direct seawater cooling at PUE 1.08 vs inland PUE 1.40 yields ~ 23 % more compute per delivered MWh.
Sources: IRENA Renewable Power Generation Costs 2024 · EU Blue Economy Report 2025 · OES Annual Report 2024 · SeaPower internal model.
Four phases. Modular by design, so each one funds the next.
Resource & site
- GIS screening across 6 candidate coastlines
- Two priority sites under exclusive lease
- 1-year wave buoy + bathymetry campaigns
- Reference DC tenant LOI signed
Single-unit pilot
- 1 × 1 MW point absorber + onshore POC
- 4 MW container DC bolted to substation
- 12-month survival & availability proof
- Environmental monitoring under OES protocols
Pelagos-1 · 60 MW
- 24-unit array · subsea hub · 33 kV export
- 60 MW IT data center, single tenant
- 40 MWh BESS for variance smoothing
- First commercial offtake at < €120 / MWh
Pelagos series
- 3 × 60 MW campuses across Atlantic basins
- Hybrid wave + offshore wind topologies
- Open-source siting & survival framework
- 1 GW order book by 2033