PaperlayerView exact revision on arXiv

Source-linked AI summary

Computing at Sea: Floating and Offshore Data Centres as a Pathway to Sustainable AI Infrastructure

Cheng Siong Chin, Jianhua Zhang, M. Venkateshkumar

arXiv:2609.12511v1eess.SYcs.AIcs.SY

TL;DR

Rising AI and data-centre demand is intensifying electricity, cooling, freshwater, and siting constraints. This article examines floating and offshore data centres, where marine cooling and renewable access shape energy use and deployment. Demonstrations report lower server failure rates and no freshwater consumption, while regulatory and maintenance constraints remain.

  • Problem

    Rising data-centre electricity demand is converging with cooling, freshwater, land, and grid constraints, prompting questions about sustainable infrastructure siting.

    Source: §I · p. 1

  • Method

    The article examines offshore architectures, energy and cooling performance, deployment evidence, and environmental, economic, engineering, and regulatory considerations.

    Source: §III · p. 2, §V · p. 4, §VI · p. 5

  • Results

    Project Natick’s server failure rates were eightfold lower than land-based equivalents, with zero freshwater consumption and no unplanned maintenance during 26 months of operation.

    Source: §VII · p. 6

  • Takeaways & Limitations

    The article concludes that technical viability is largely established, while remaining barriers center on regulation, maritime governance, grid interconnection, and environmental coordination.

    Source: §IX · p. 7

    Offshore facilities occupy an unresolved regulatory intersection, so regulated data may need compliant jurisdictions and offshore use may be limited to workloads unconstrained by sovereignty obligations.

    Source: §VIII · p. 6

Abstract

from arXiv · show

The rapid expansion of artificial intelligence is transforming data centres into one of the world's fastest-growing sources of electricity demand. As AI systems scale in size and capability, the physical infrastructure supporting computation is approaching critical limits in energy availability, cooling capacity, land use, freshwater consumption, and carbon management. Conventional land-based data centres are increasingly constrained by urban land competition, grid congestion, environmental pressures, and lengthy permitting processes, raising fundamental questions about where future computing infrastructure can sustainably exist. This article examines floating and offshore data centres as an emerging alternative model for digital infrastructure. By relocating computation to marine environments, offshore systems can exploit the ocean's natural cooling capacity, reduce freshwater dependence, and enable direct integration with offshore renewable energy resources such as wind, wave, and tidal power. Early deployments have demonstrated the potential for significantly improved energy efficiency and operational reliability compared with conventional facilities, while also opening new possibilities for distributed and resilient computing architectures. The article explores how offshore computing may reshape the future relationship between electrification, renewable energy, and large-scale AI infrastructure. It analyses the opportunities and trade-offs associated with marine deployment, including environmental impacts, engineering design challenges, economic feasibility, and regulatory governance. Rather than treating offshore data centres as experimental novelties, the article presents them as part of a broader systems-level transition in how society may power, cool, and sustain the next generation of computational growth.

I. INTRODUCTION

Rapidly rising data-centre and AI electricity demand is converging with cooling, freshwater, land, and grid constraints. Early offshore deployments suggest marine facilities can reduce freshwater use and improve energy efficiency and reliability.

I. INTRODUCTION

  • 300–380 TWh of electricity was consumed by hyperscale data centres in 2023, with the IEA projecting around 945 TWh by 2030 as demand grows approximately 15% annually from 2024,.In the United States, data centres consumed 176 TWh in 2023, with estimates as high as 580 TWh for 2028.Source: §I · p. 1, Table I · p. 2
  • Cooling typically uses 30–40% of conventional data-centre electricity, while freshwater use and limited land and grid capacity are already prompting local and regulatory pressure,.A West Des Moines cluster supporting GPT-4 training used roughly 6% of its local water district’s monthly supply in one month.Source: §I · p. 1
  • Marine facilities can use seawater for cooling without freshwater draw and can access offshore wind, wave, tidal, and ocean-thermal energy resources.The passage presents floating and seafloor systems as alternatives to increasingly constrained land-based facilities.Source: §I · p. 1
  • Project Natick operated for two years off Scotland’s Orkney Islands with server failure rates below comparable land-based systems and no freshwater consumption.Recent commercial installations in China reported PUE figures below 1.3.Source: §I · p. 2

II. FLOATING DATA CENTER ARCHITECTURE

Floating and subsea data-centre architectures pair marine deployment with offshore energy and cooling. They range from sealed seafloor capsules to modular containerised systems, with co-location and modularity shaping their operating model.

II. FLOATING DATA CENTER ARCHITECTURE

  • Offshore wind, wave, and tidal generation can feed a collocated subsea data centre, linking marine power sources with computing infrastructure.The figure presents reliability, rapid deployment, and zero freshwater use as advantages, and depicts lower infrastructure cost and firm power as outcomes.Source: §II · p. 2
  • Project Natick’s sealed capsule housed 864 servers and 27.6 petabytes of storage at 35.6 m depth, using passive seawater cooling and a nitrogen atmosphere.The nitrogen interior eliminated corrosion and reduced humidity to near zero; server failure rates were approximately eightfold lower than comparable land-based facilities.Source: §II · p. 2
  • Containerised modules can be assembled and stress-tested onshore, deployed within weeks, and added or removed incrementally as demand changes.A 20-rack module contrasts with 18–36 months for greenfield hyperscale construction and decouples investment from demand forecasting.Source: §II · p. 2
  • Co-designing offshore energy and computing can shorten transmission paths, combine wind with wave generation, and pair OTEC electricity with chilled water for servers.The passage describes these co-location strategies as moving platforms toward self-contained energy islands less dependent on onshore grids.Source: §II · p. 2

III. ENERGY CONSUMPTION COMPARISON

Offshore seawater cooling is associated with lower PUE than legacy air-cooled facilities, with Project Natick reported near 1.07. The section compares these values across data-centre generations.

III. ENERGY CONSUMPTION COMPARISON

  • The subsea concept figure links offshore wind, tidal, and wave power to a collocated data centre and depicts reliability, deployment time, and freshwater use as advantages.It presents the “marine powerhouse” vision as yielding lower infrastructure cost and stable, firm power.Source: Figure 1 · p. 3
  • Approximately 1.07 PUE was estimated for Project Natick, compared with 1.11–1.15 for commercial surface-floating platforms and 1.09 for Google’s global fleet average in 2025,,.The global data-centre average was approximately 1.58 in 2023. Table II summarizes PUE values across generations.Source: §III · pp. 2–3

Cooling Energy Fraction and Freshwater Elimination

Seawater-cooled offshore facilities reduce cooling energy and freshwater use relative to conventional cooling-tower systems. Offshore operation also changes non-IT overhead and introduces transmission and latency considerations.

Cooling Energy Fraction and Freshwater Elimination

  • 30% of facility energy goes to mechanical refrigeration in traditional air-cooled data centres, compared with 3–8% for seawater-cooled offshore facilities and approximately 2–3% for passive underwater capsules.Cooling towers use 1.8–2.5 L/kWh of freshwater, while seawater-cooled offshore facilities have essentially zero freshwater WUE.Source: Cooling Energy Fraction and Freshwater… · p. 3
  • Automated, periodically serviced offshore platforms reduce background energy loads such as lighting, security, and administration by avoiding continuous human occupancy.The passage attributes direct PUE improvement to eliminating incidental energy demands, rather than server or cooling-chemistry advances.Source: Cooling Energy Fraction and Freshwater… · p. 3
  • For nearshore deployments 50–200 km offshore, total transmission losses remain comparable to equivalent onshore AC transmission, while 50–100 km offshore adds approximately 0.5–1.0 ms roundtrip fibre-propagation latency.The passage says AI training tolerates inter-node delays up to 100 ms, whereas workloads needing sub-millisecond responses require proximity-optimised placement.Source: Cooling Energy Fraction and Freshwater… · pp. 3–4

IV. RENEWABLE ENERGY INTEGRATION

Offshore renewable integration combines co-located marine generation with cooling and storage to support substantial IT loads. Wind, wave, tidal, and OTEC resources offer complementary power and thermal-management roles, while hybrid architectures are presented as a route toward stabilising marine microgrids.

IV. RENEWABLE ENERGY INTEGRATION

  • A five-turbine, 75 MW wind array yields 30–41 MW average output at 40–55% capacity factors, enough for a 30–35 MW IT load with short-duration battery storage.Offshore turbines can reduce transmission distances and associated losses when mounted on or near the platform.Source: §IV · p. 4
  • In tropical and subtropical waters, OTEC can provide electricity and chilled water from the same temperature gradient, while wave generation can complement wind output for several hours after winds drop.The passage places OTEC's integration pathway at approximately 20°N–20°S and gives wave energy a 20–35% capacity factor.Source: §IV · p. 4
  • Tidal stream generators have 35–45% capacity factors in strong tidal regimes, and Orkney's tidal and wind resources powered Project Natick exclusively.The European Marine Energy Centre in Orkney was a primary factor in site selection for the project.Source: §IV · p. 4
  • A hybrid power architecture combines multiple generation sources, battery storage, shore-cable backup, and real-time energy management to maintain IT power quality.AI-driven workload scheduling is described as adding grid responsiveness, while data centres can absorb generation variability and stabilise microgrids.Source: §IV · p. 4

V. ENVIRONMENTAL IMPACT

The environmental case includes potential lifecycle benefits from improved server reliability and freshwater-free cooling, alongside installation and thermal risks that require site-specific assessment. The section also frames subsea development as a staged technology pathway toward reducing computing's environmental impact.

V. ENVIRONMENTAL IMPACT

  • Eightfold lower failure rates in Project Natick imply reduced hardware manufacturing over facility lifetimes, with embodied emissions estimated at 300–800 kg CO2-equivalent per server.The passage says these lifecycle savings are not yet captured by conventional carbon accounting frameworks.Source: §V · p. 4
  • Thermal effluent from seawater cooling is typically 3–5°C above ambient, while seafloor anchoring disturbs benthic communities during installation.Hard substrates have been observed to be rapidly colonised and function as artificial reefs; deep-water intakes require screening and constrained flow velocities to minimise entrainment risks.Source: §V · pp. 4–5
  • The subsea readiness roadmap runs from commercial seawater cooling and surface platforms at TRL 8–9 to conceptual compute archipelagos at TRL 2–3, targeting zero-freshwater, high-efficiency operation and clean offshore power.Intermediate stages include sealed nitrogen-filled capsules at TRL 6–7 and autonomous ROV maintenance and integration at TRL 3–5.Source: Figure 2 · p. 5
  • Most maritime jurisdictions mandate Environmental Impact Assessments, which provide ecological risk-management mechanisms and enforceable monitoring obligations during operation.Source: §V · p. 5
  • A 100 MW data centre dissipates approximately 85–90 MW of heat; open-water discharge is described as negligible at current scales, but enclosed waters can face ecologically significant temperature anomalies.Thermal dispersion modelling is warranted where circulation is restricted or thermally sensitive ecosystems may be affected.Source: §V · p. 5

VI. ECONOMIC CONSIDERATIONS

Offshore projects trade land acquisition for platform and cable costs, with modular deployment and shared infrastructure affecting capital economics. The paper describes potential ten-year operating-cost savings in energy- and water-intensive markets, while identifying maintenance access as a continuing disadvantage.

VI. ECONOMIC CONSIDERATIONS

  • Urban land acquisition accounts for 15–25% of development cost at prime co-location sites, while offshore platforms are estimated at USD 15–25 million per 10 MW and broadly comparable after land premiums.Source: §VI · p. 5
  • A 1 MW floating barge module is estimated at USD 3 million before cable infrastructure; submarine cabling is the principal offshore-specific capital cost, but shared infrastructure can improve per-operator economics.Source: §VI · p. 5
  • Over ten years, operating savings in energy- and water-cost-intensive markets can offset the initial premium for offshore platforms and submarine cables, producing a favourable total cost of ownership trajectory.Markets named include Singapore, California, Arizona, and Japan.Source: §VI · p. 5
  • Offshore maintenance access is a principal operating-cost disadvantage: surface platforms need vessel servicing, while submerged capsules may require ROV intervention or retrieval for major failures.Complementary strategies include subsystem redundancy, AI-assisted remote diagnostics, and modular swap-out architectures; Project Natick ran at least 26 months without intervention.Source: §VI · p. 5
  • Modular offshore systems can be commissioned within 6–12 months of contract award, compared with 18–36 months from groundbreaking to full capacity for land-based hyperscale construction.The passage links the faster trajectory to operators' pressure to provision AI compute amid planning, grid-connection, and labour constraints.Source: §VI · p. 6

VII. INDUSTRY CASE STUDIES

The case studies range from Project Natick's submerged capsule trial to China's operational floating platform and Hamina's long-running seawater cooling. A commercial underwater capsule design also targets coastal edge computing and content delivery.

VII. INDUSTRY CASE STUDIES

  • Project Natick's Phase 2 operated for 26 months at 35 m depth with 12 racks, 864 servers, and 27.6 petabytes of storage; failure rates were eightfold lower than land-based equivalents, with zero freshwater use and no unplanned maintenance.The nitrogen atmosphere eliminated corrosion, thermal stability reduced solder fatigue, and the absence of human activity removed physical disturbance as a failure mode.Source: §VII · p. 6
  • Hainan's offshore floating data centre entered service in 2023 with about 400 server racks and PUE below 1.15; additional deployments are advancing along the Pearl River Delta coast.Seawater from 20 m depth circulates through plate heat exchangers and returns 3–4°C above ambient; the government has designated offshore data infrastructure a strategic priority.Source: §VII · p. 6
  • Hamina has used Gulf of Finland seawater cooling for over a decade with zero freshwater consumption and reported PUE of 1.09, providing longitudinal evidence for investment decisions in offshore deployment.The repurposed paper mill facility is not itself floating; a USD 670 million capacity expansion followed in 2022.Source: §VII · p. 6
  • Subsea Cloud's proposed underwater capsules use diver-accessible maintenance hatches for component replacement without full retrieval and target edge computing and content delivery for coastal markets.Initial commercial deployments were targeted for 2025–2026; the passage claims structural advantages in latency and cost positioning relative to land-based alternatives.Source: §VII · p. 6

VIII. CHALLENGES AND FUTURE DIRECTIONS

Offshore data centres face maintenance, regulatory, and grid-integration constraints, alongside opportunities in flexible energy use and distributed computing. Future scaling also depends on enabling technologies and coordinated operations.

VIII. CHALLENGES AND FUTURE DIRECTIONS

  • Unplanned submerged failures require full capsule retrieval using specialist vessels and ROVs over several days, despite planned five-year service intervals.The passage identifies subsystem redundancy, robotic modular replacement, and predictive diagnostics as design priorities; surface platforms remain weather-vulnerable and need marine vessel support.Source: §VIII · p. 6
  • Offshore IT jurisdiction within exclusive economic zones remains legally untested in most jurisdictions, while data-residency rules may require regulated workloads to be routed to compliant jurisdictions.The passage identifies latency-insensitive workloads such as AI training, rendering, archival storage, and scientific computing as possible offshore uses.Source: §VIII · p. 6
  • Mainland-grid integration requires bidirectional submarine cables, power-quality management, and participation in ancillary service markets.As controllable, continuous loads, data centres can provide demand flexibility; direct access to offshore wind is described as amplifying those benefits.Source: §VIII · p. 6
  • Offshore operations could coordinate pump scheduling, renewable forecasts, and training workloads with ocean conditions and energy availability.Scaling also depends on terabit-scale submarine cables, autonomous underwater robots, pressure-tolerant immersion cooling, and autonomous service vessels; distributed compute arrays are envisioned for coastal megacities, with Southeast Asia proposed as a proving ground.Source: §VIII · p. 7

IX. CONCLUSION

The conclusion presents offshore data centres as a technically viable option amid rising AI demand, but says their potential depends on institutional coordination and workforce development. It identifies the ocean’s thermal and energy resources as underused, with offshore computing’s future contingent on governance and education evolving alongside technology.

IX. CONCLUSION

  • Project Natick, commercial offshore deployments, and long-term seawater cooling operations indicate that technical viability is largely established.The conclusion says the remaining barriers are coordination-based rather than engineering-centric.Source: §IX · p. 7
  • Key barriers span regulation, maritime governance, grid interconnection, and environmental oversight, requiring joint action by policymakers, regulators, utilities, and researchers.Source: §IX · p. 7
  • Priorities include lifecycle carbon and water accounting, assessment of offshore-renewable integration benefits, and ecological validation of marine thermal use.Source: §IX · p. 7
  • Realising offshore computing’s potential also depends on developing an interdisciplinary workforce across marine engineering, subsea systems, power integration, and data-centre operations.The conclusion says these capabilities are currently fragmented across separate domains.Source: §IX · p. 7
Loading 2609.12511v1…