The relentless expansion of cloud capacity has collided with finite land, power, and regulatory headroom in key metros, pushing infrastructure engineering toward an unconventional frontier: the ocean. Recent progress in floating and subsea data centers—led by South Korean shipbuilders and evidenced by operating pilots—indicates a methodical convergence of maritime engineering and digital utilities. This analysis examines the technical maturity, operational data, market dynamics, and strategic implications for hyperscale operators, with a focus on Amazon Web Services as the largest prospective tenant of such capacity.
Technical Foundations: A 50 MW Blueprint from Samsung Heavy Industries
At the core of the commercial proposition is Samsung Heavy Industries’ (SHI) 50 MW-class floating data center design, which secured Approval in Principle from both the American Bureau of Shipping and Lloyd’s Register in April 2026 1. The design borrows heavily from SHI’s experience with floating liquefied natural gas (FLNG) facilities—a direct transfer of offshore structural and systems integration expertise 1. Its power architecture is engineered for operational flexibility: external subsea cables provide a grid connection, while onboard solid oxide fuel cells (SOFCs) fueled by LNG can reduce the power interconnection burden and offer a dispatchable alternative 1. SHI has further hardened the concept through a partnership with Supermicro to tailor server hardware for the marine environment, addressing vibration, humidity, and corrosion from the component level up 1.
Commercial intent is evident. SHI has signed a memorandum of understanding with Greek shipowner Capital Clean Energy Carriers and reports a growing pipeline of inbound collaboration inquiries from global fleet operators 1. The target deployment regions—coastal cities and island markets with severe power and land constraints—align with the most acute data center siting challenges 1. The broader taxonomy of designs includes barge-mounted, ship-converted, subsea, and fixed offshore structures, each with distinct operational trade-offs 1.
Operational Benchmarks: Pilots Reduce Uncertainty
While no floating data center at the 50 MW scale is yet in service, smaller active installations are generating the operational data that engineers crave. The most instructive reference is the Shanghai HiCloud Technology subsea data center, a 24 MW facility approximately 10 kilometers offshore from Lingang, which began commercial operations around May 2026 1. Using offshore wind power and direct seawater cooling, the facility reports a 22.8% reduction in power consumption compared to equivalent land-based data centers, along with a substantial decrease in freshwater withdrawal 1. In the United States, Nautilus Data Technologies operates a smaller 6.5 MW barge-mounted proof of concept at the Port of Stockton, California, demonstrating viability in a riverine environment 1.
Microsoft’s well-publicized Project Natick, an underwater data center capsule, is now widely regarded as a commercial non-starter, serving as a cautionary example that contrasts with the surface and near-shore approaches gaining traction 1,3. Separately, Japanese shipping major Mitsui O.S.K. Lines is evaluating the conversion of decommissioned vessels into data centers, further broadening the potential supply pool 1.
Friction Points: Technical, Environmental, and Regulatory Hurdles
Despite encouraging pilot data, several friction points must be resolved before floating data centers can become a routine part of the capacity portfolio. The marine environment introduces persistent risks: vibration, inclination, salinity, and humidity that challenge server stability; saltwater corrosion of exposed components; and bandwidth constraints arising from maritime connectivity 1. Environmental scrutiny focuses on ocean thermal pollution from heated cooling discharge, a non-trivial concern in sensitive coastal ecosystems 1. Forward-looking concepts, such as using waste heat for desalination, may mitigate some impacts but remain conceptual 1. Physical security—from water intrusion to piracy—adds an operational complexity not present in terrestrial facilities 1.
The most significant gating factor, however, is the absence of a standardized permitting and regulatory framework. Industry observers consistently cite this as the primary barrier to broad commercialization 1. Without predictable compliance pathways, the capital deployment timelines remain uncertain. Moreover, the industry needs a substantially larger corpus of long-term operational data to validate reliability, mean time between failures, and total cost of ownership at scale 1. Near-term revenue contribution from FDCs is expected to be minimal, and there is legitimate market skepticism about whether data center hardware can generate sufficient return before technological obsolescence forces a refresh cycle 1,9.
Supply Chain Constraints: Korean Shipyards at Capacity
The practical ability to build floating data centers is tightly coupled to the health of the Korean shipbuilding industry, which is currently operating at or above full capacity. Major yards report utilization rates at 100% or higher, driven by an extraordinary surge in liquefied natural gas carrier orders—16 contracted year-to-date through mid-May, compared to only seven in all of 2025 2. The top three Korean shipbuilders secured $19.1 billion in contracts from January to mid-May, putting them on pace to exceed the prior year’s $36.3 billion intake 2. Further tightening the supply picture, the U.S. Department of Defense is conducting a $1.85 billion feasibility study on outsourcing warship construction to South Korea and Japan, while the ultra-high-voltage transformer backlog of Korean manufacturers—Hyosung Heavy Industries, HD Hyundai Electric, and LS Electric—stands at KRW 32 trillion ($21.3 billion), underscoring the breadth of industrial demand 2.
For floating data centers, this capacity crunch means that shipyard slots are scarce and pricing power rests with builders. Any hyperscale operator considering FDC procurement must factor in potential cost escalation and schedule uncertainty, similar to the dynamics observed in the offshore energy sector during upcycles.
Business Model and Capital Trends
The emerging business model for floating data centers mirrors the asset-leasing structures common in maritime and infrastructure finance: shipowners (or specialized infrastructure funds) retain ownership of the asset and lease capacity to cloud operators under long-term contracts. This decoupling offers a more stable revenue stream than the cyclical freight markets and has attracted interest from global fleet operators and financial sponsors 1. Capital costs per vessel are estimated at $100–200 million, a figure that, while significant in isolation, compares favorably against the multi-billion-dollar outlays for terrestrial hyperscale campuses that must service gigawatt-scale demands 1,4.
Broader data center financing trends provide additional context. J.P. Morgan strategist Tarek Hamid projects a stabilization of project financing around 2028, while JLL reports that global data center transaction volumes reached a record $73 billion in 2025 5,6. Emergent modular data center models—requiring letters of intent before capital deployment—are enabling multiple capital partners to share project economics, a structure that could be directly applied to FDCs 5. This alignment of capital availability and technological maturation creates a plausible window for first commercial units in the 2028–2030 timeframe.
Implications for Amazon Web Services
For AWS, the world’s largest cloud infrastructure provider, floating data centers represent not a revolutionary shift but a logical extension of the infrastructure toolkit—a new type of “site” that can bypass the land-acquisition and regulatory bottlenecks now manifesting in cities such as Seattle, where a one-year moratorium on new data centers was enacted 1,8. The demonstrated 22.8% power consumption reduction from HiCloud, if sustained at scale, would directly improve energy efficiency metrics and lower Power Purchase Agreement costs, both of which are direct contributors to operational margin and sustainability targets.
However, the path to adoption is laden with engineering and commercial trade-offs. The $100–200 million unit cost must be weighed against the uncertainty of operating in an unproven regulatory environment and the need to secure scarce shipyard capacity. While early engagement could yield first-mover advantages in underserved coastal markets, a premature commitment risks locking in a technology that may not yet match the reliability of terrestrial builds. Given that the financial community expects project financing to stabilize around 2028, coinciding with the likely maturation of FDC technology, the next 24–36 months represent a prudent evaluation window for Amazon. During this period, the company could assess operational data from the HiCloud and Nautilus installations, engage shipbuilders and classification societies in design partnerships, and position itself to negotiate favorable terms with owner-operators—similar to its approach in renewable energy procurement.
More broadly, the push to diversify hardware and supply chain resilience, as evidenced by ongoing efforts to expand supplier bases 7, suggests that floating data centers fit within a pattern of infrastructure optionality that AWS has consistently pursued. The convergence of maritime engineering and digital infrastructure is a methodical progression, not a speculative leap. For a company that measures success in throughput per dollar and mean time between failures, the FDC represents a potentially elegant—if still nascent—solution to the friction of land-based scaling.