State of the Industry
State of Subsea AI Infrastructure 2026
A bookmarkable industry brief for subsea and ocean-adjacent AI infrastructure. Updated over time so researchers, operators, and journalists can return to one page for the current picture.
· 16 min read
By Seabase Editorial
← Articles index · Home · Infrastructure · Nori · About
Topics: Infrastructure · AI Inference · Cooling · Geography
Edition: 2026 Q3 brief. Short URL: /industry. This page is meant to be revisited. When the industry moves, the sections below should move with it.
Ocean compute is no longer only a research curiosity. Microsoft's Project Natick made underwater servers familiar. Startups, national programs, and marine engineering firms continue to propose systems that put compute in or on the water. Headlines often collapse those efforts into one story. They are not one architecture.
Bookmark this page for the category map. Use the statistics and glossary pages for the numbers and definitions.
How this brief works
- It is industry analysis, not a product brochure.
- It separates demonstrated work from concept marketing.
- It points to deeper essays and the living statistics page for ranges.
- Material updates should change the modified date and edition label.
Companion references: AI Infrastructure Statistics, What Is a Subsea Data Center?, and the glossary.
Major subsea and ocean-adjacent deployments
Visible activity still clusters into a few types: research pods, floating or barge systems, nearshore submerged modules, and marine-cooled hybrids. Public verification of commercial scale, uptime, and customer workloads varies by project.
Microsoft Project Natick
The best-known research program. Sealed underwater vessels, multi-month operation, retrieval, and public discussion of reliability and cooling. Best read as a proof point, not a finished product line. Detail: Why Microsoft Put a Data Center Underwater and Beyond Project Natick.
Subsea Cloud
Associated with submerged data-center concepts and public advocacy for underwater siting. Evaluate demonstrated deployments, interconnect plans, and service models separately from concept marketing.
China coastal and Hainan-linked projects
Chinese organizations have publicized underwater data-center work tied to coastal deployment narratives, including projects discussed in connection with Hainan. These efforts matter as evidence of industrial and state interest beyond Western research demos. Independent verification of scale and customer use still varies by source.
Nautilus Data Technologies
Focused on water-based cooling and marine or barge-adjacent facility concepts. Related to ocean compute, but not identical to fully submerged IT modules. A reminder that "water-cooled" and "underwater" are often blended in headlines.
Watchlist for the next quarter
- Any repeatable module retrieval and redeployment campaigns (not only first installs)
- Published interconnect and fiber paths for nearshore sites
- Environmental telemetry shared with regulators or research partners
- Customer-facing capacity claims with measurable SLAs
New GPU generations
Each accelerator generation raises rack power, thermal density, and networking requirements. That pressure is one reason liquid cooling and unusual siting ideas keep returning. The practical question for ocean systems is not whether GPUs get hotter. It is whether the facility model can service, power, and network them near demand.
- Track accelerator TDP and multi-GPU node designs from primary vendor materials.
- Watch whether liquid cooling becomes default rather than optional in dense AI halls.
- Ask how marine designs handle accelerator refresh cycles without rare vessel campaigns.
Related glossary: NVLink, NVSwitch, InfiniBand.
Rack power trends
Dense AI racks have moved from classic enterprise densities into tens of kilowatts, with high-density designs pushing toward and beyond roughly 80–100 kW in some planning envelopes. Exact numbers depend on generation and cooling. Living ranges: AI Infrastructure Statistics.
For subsea systems, higher rack power increases the importance of heat-exchanger design, materials, biofouling management, and shore power quality.
Cooling trends
Land AI infrastructure is shifting toward direct liquid cooling, and coolant temperatures are increasing with newer platforms. Ocean concepts use seawater as the stable heat sink for those liquid loops. The commercial thermal argument is the resulting ΔT and heat-transfer efficiency, paired with modular, retrievable infrastructure that can evolve with AI hardware. Both paths still need disciplined internal loops, monitoring, and service procedures.
| Trend | Why it matters | Read more |
|---|---|---|
| DLC adoption on land | Enables denser racks | Direct Liquid Cooling glossary |
| Seawater heat rejection | Can reduce land cooling plant | Underwater cooling costs essay |
| Hybrid marine cooling | Blurs water-cooled vs underwater | Floating vs subsea essays |
| Environmental telemetry | Needed for permits and trust | Environmental Accountability |
See Can Underwater Data Centers Reduce Cooling Costs? and Direct Liquid Cooling.
AI inference demand
Training can concentrate where power is cheap. Inference, agents, and interactive applications increasingly care about geography. Multi-step pipelines compound network delay. That is why coastal metros, cable landings, and regional capacity keep appearing in serious plans.
Related: Why AI Inference Changes Infrastructure Geography, The Latency Tax, Persistent Regional AI.
Major announcements to track
This section is a recurring watchlist, not a complete news feed. Each quarter, prioritize announcements that change architecture, regulation, or demonstrated operations:
- New underwater or floating installations with retrieval evidence
- Partnerships between compute operators and marine contractors
- Power or SMR announcements tied to coastal AI campuses
- Cable system and landing-station expansions near demand centers
- National or municipal AI infrastructure strategies that mention marine siting
Regulatory developments
Ocean compute sits at the intersection of data-center practice and marine use. Expect scrutiny on thermal discharge, acoustics, materials, seabed disturbance, cable routes, and emergency recovery. Continuous monitoring is becoming part of the credibility test, not an optional add-on.
Related: Environmental Accountability and environmental research.
Cable landing station growth
Subsea fiber remains the backbone of international traffic. Landing density and route diversity shape which coastal regions can host low-latency, high-bandwidth AI infrastructure. Proximity to a cable landing station helps, but terrestrial routing and peering still decide outcomes.
Related: coastal metros and Metro AI Infrastructure.
Architecture map
| Approach | What it usually means | Primary tension |
|---|---|---|
| Research pod | Sealed vessel for experiment and retrieval | Not a product fleet |
| Floating facility | IT on barge, vessel, or platform | Motion, weather, surface security |
| Modular subsea | Sealed modules on or near seabed | Serviceability and connectors |
| Marine-cooled hybrid | Land or pier IT with seawater cooling | Still land-constrained for halls |
| Energy-first offshore | Compute beside remote generation | Fiber, logistics, latency |
Comparisons: Floating vs. Subsea Data Centers, Floating vs Fixed Subsea Infrastructure, ocean compute landscape.
Open questions
- Can modules be manufactured and redeployed like a product line?
- What maintenance tempo is economically sustainable?
- How do sites secure firm power without isolating themselves from fiber?
- What environmental telemetry satisfies regulators over multi-year operations?
- Which workloads actually benefit from marine siting versus land edge sites?
Where Seabase's approach differs
Seabase is one participant in this category. Its published approach emphasizes modular subsea infrastructure near coastal demand, power-source flexibility, and terrestrial fiber, rather than attaching the business to a single offshore generator or a one-off research vessel. That is a positioning choice inside the map above, not a claim that other architectures cannot work for some use cases.
When evaluating any approach, including Seabase, weigh serviceability, interconnect, environmental monitoring, and demonstrated engineering maturity.
Versioned reports in this library
- State of Subsea AI Infrastructure 2026 (this brief)
- AI Infrastructure Statistics 2026 (living tables)
- Reports index (named editions and companions)
As Seabase publishes additional primary findings, those updates should feed this brief and the statistics page rather than living only as isolated posts.
Next step
To discuss coastal or subsea AI infrastructure in a specific region, contact Seabase.
Related research
- AI Infrastructure Statistics: Power, Cooling, Latency, and Networking
A living reference of AI infrastructure statistics: rack power density, PUE ranges, cooling methods, metro latency bounds, and cable landing context. Updated over time for researchers and operators.
- What Is a Subsea Data Center? (Complete Guide)
A complete guide to subsea and underwater data centers: history, benefits, challenges, power, networking, cooling, maintenance, environmental concerns, floating vs subsea, and AI inference.
- Ocean Compute: The Emerging Landscape of Subsea, Floating, and Offshore AI Infrastructure
A map of ocean-compute companies and architectures, and where Seabase fits as modular subsea infrastructure near coastal demand, power-source flexible, and connected to terrestrial fiber.
- Why Microsoft Put a Data Center Underwater
What Microsoft Project Natick tested, why the company put servers underwater, what the results suggested about reliability and cooling, and what commercial subsea compute still has to prove.
- Floating vs. Subsea Data Centers: Why Seabase Chose Subsea
Floating data centers can reduce land and cooling constraints, but remain exposed surface marine assets. Seabase explains why modular subsea infrastructure better fits its goals for metropolitan proximity, stable operation, depth-based cooling, serviceability, and global scale.