The Orbital Data Center Hype
In 2025–2026, a wave of announcements has put orbital data centers in the spotlight. SpaceX filed for a constellation of up to 1 million satellites. Google launched Project Suncatcher. NVIDIA posted job listings for space-based compute. But the economics tell a very different story.
SpaceX Filing
In January 2026, SpaceX filed with the FCC for an orbital data center constellation of up to 1 million satellites in low Earth orbit. The scale is unprecedented — but so is the cost.
Google Suncatcher
Google's "moonshot" Project Suncatcher proposes solar-powered satellite constellations equipped with TPUs and free-space optical links. Their own whitepaper admits energy costs of $14,700/kW/year in orbit vs $570–$3,000 on the ground.
Industry Skepticism
AWS CEO Matt Gorman stated at a recent event: "There are not enough rockets to launch a million satellites yet. The cost of getting a payload in space today is massive. It is just not economical."
"A lot of the energy here is FOMO and aesthetic futurism, not a grounded value proposition. People jump straight to hardware and hand-wave the business case, as if the economics are self-evident. They aren't."
— Andrew McCalip, Space Engineer & Orbital Economics Analyst
The Hard Numbers: 1 GW Data Center
Based on Andrew McCalip's first-principles model and TechCrunch analysis (February 2026). All figures use conservative estimates for a 1 GW facility over 5 years.
Orbital Data Center
$51.1B
Underwater Data Center
$5–8B
6–10×
Cheaper Than Orbital
An underwater data center costs 6–10× less than an orbital equivalent for the same 1 GW capacity. And unlike orbital systems, it uses proven technology that exists today.
Cooling: The Decisive Factor
Cooling is the single largest operational cost for any data center. In space, it's a nightmare. Underwater, it's essentially free.
Space: The Thermal Nightmare
Space is a vacuum. There is no air, no water, no convection. The only way to dissipate heat is through radiation — which requires massive radiator panels that add enormous weight and cost.
Every kilogram of radiator must be launched to orbit at $500–$3,600/kg. For a 1 GW facility generating hundreds of megawatts of waste heat, the radiator mass alone would be tens of thousands of tons.
GPU die temperatures must stay below 85°C, but in direct sunlight in orbit, external surfaces can reach 120°C+. The thermal management challenge is the single biggest unsolved problem for orbital compute.
Underwater: Nature's Perfect Coolant
Water has a thermal conductivity 25× higher than air and a heat capacity 4× greater. At depth, ocean temperatures are a constant 2–4°C — a free, infinite heat sink.
China's Highlander project demonstrated a 90% reduction in cooling energy compared to land-based data centers. Microsoft's Project Natick achieved similar results with passive seawater cooling.
No chillers. No cooling towers. No freshwater consumption. The ocean does what billions of dollars in cooling infrastructure cannot — and it does it for free, 24/7, forever.
90%
Cooling energy reduction (underwater vs land)
$14,700
Cost per kW/year for energy in orbit
$570
Cost per kW/year for energy on ground (low end)
Reliability & Failure Rates
Microsoft's Project Natick proved that underwater data centers are dramatically more reliable than land-based ones. In space, the environment is far more hostile.
8×
Fewer Failures Underwater
Microsoft's Northern Isles datacenter had a failure rate one-eighth of land-based equivalents after 2 years on the seafloor.
9%/yr
GPU Failure Rate in Orbit
Cosmic radiation causes bit flips and hardware degradation. Meta reports 9% annual GPU failure rates even on the ground — in orbit, it's worse.
2.5%/yr
Solar Cell Degradation
Solar panels in LEO degrade 2.5% per year from radiation and atomic oxygen, requiring constant replacement launches.
Why underwater is more reliable: The sealed nitrogen atmosphere eliminates corrosion from oxygen and humidity. The absence of human access means no accidental bumps or component disturbances. Constant temperature eliminates thermal cycling stress. And unlike orbit, if something does fail, a maintenance submarine can reach the facility in hours — not never.
Complete Comparison Matrix
A comprehensive side-by-side comparison across every critical metric. Conservative estimates used throughout.
| Metric | Orbital DC | Underwater DC |
|---|---|---|
| Total Cost (1 GW, 5yr) | $51.1 Billion | $5–8 Billion |
| Cost per Watt | $51.10/W | ~$8/W |
| LCOE (Energy Cost) | $1,167/MWh | ~$200/MWh |
| Cooling Method | Radiative only (vacuum) | Passive seawater (free) |
| Cooling Energy Savings | N/A (massive radiators needed) | 90% reduction vs land |
| Server Failure Rate | High (radiation + thermal cycling) | 8× lower than land (Microsoft) |
| GPU Failure Rate | 9%+/year (cosmic radiation) | <1.5%/year (sealed environment) |
| Latency to Users | 4–40ms (LEO to GEO) | <1ms (coastal, near population) |
| Deployment Time | Years (launch campaigns) | Months (modular pods) |
| Maintenance Access | Impossible (disposable) | Submarine in hours |
| Physical Security | Orbital debris, solar storms | Excellent (depth = shield) |
| Power Source | Solar (degrades 2.5%/yr) | Grid, offshore wind, SMR |
| Freshwater Consumption | N/A | Zero (seawater cooling) |
| Land Footprint | N/A | Zero (seafloor) |
| PUE (Efficiency) | N/A (different model) | ~1.05 (near-perfect) |
| Technology Readiness | Experimental (TRL 3–5) | Proven (TRL 7–9, Microsoft/China) |
| Scalability | Limited by launch capacity | Modular, unlimited |
Latency: The Speed of Light Problem
For AI inference, financial trading, and real-time applications, latency is critical. The physics of light-speed delay makes orbital compute fundamentally slower.
Orbital Latency
Light travels at 300,000 km/s. A satellite in LEO at 550 km altitude adds a minimum 3.7ms round-trip — and that's the absolute physical minimum with no processing delay.
For GEO orbits (35,786 km), latency jumps to ~240ms round-trip. For AI inference requiring multiple round trips, this makes orbital compute fundamentally uncompetitive for latency-sensitive applications.
Underwater Latency
Underwater data centers sit on the seafloor near coastal population centers. Connected via submarine fiber optic cables — the same infrastructure that carries 99% of intercontinental internet traffic.
Latency is sub-millisecond to nearby cities. Subsea Cloud reported a 98% reduction in latency for their pressure-equalized pods compared to remote land-based facilities.
Proven, Not Theoretical
Underwater data centers are not theoretical. They have been built, deployed, operated, and validated by the world's largest technology companies.
Microsoft Project Natick
2015–2020 — Two phases of testing. 864 servers deployed on the seafloor near Scotland's Orkney Islands for 2 full years.
Result: 8× fewer failures than land-based equivalents. Powered by 100% renewable energy from tidal and wind sources.
China Highlander (Hainan)
2025 — World's first commercial underwater data center. Budget: $223 million. Deployed off Hainan Island.
Achieves 90% cooling energy reduction and 40% total energy savings. Powered by offshore wind energy.
Subsea Cloud
2024–2025 — Pressure-equalized pods for underwater colocation. Reports 40% energy savings and 98% latency reduction.
Modular design allows rapid deployment and scaling. No land acquisition, no freshwater, no cooling towers required.
Technology Readiness Level: Underwater data centers are at TRL 7–9 (system prototype demonstrated in operational environment to actual system proven in operational environment). Orbital data centers remain at TRL 3–5 (experimental proof of concept to component validation in relevant environment). The gap is 5–10 years of development.
Environmental Comparison
As the world demands sustainable computing, the environmental calculus is clear.
Orbital: The Hidden Costs
Launching 1 million satellites requires thousands of rocket launches, each burning hundreds of tons of propellant. The total propellant mass for a 1 GW orbital DC is estimated at 6,750 tons per launch campaign.
Deorbited satellites burn up in the atmosphere, depositing aluminum oxide particles in the stratosphere — a growing concern for ozone depletion.
Space debris from failed or decommissioned satellites poses a cascading collision risk (Kessler Syndrome) that could render entire orbital bands unusable.
Underwater: Minimal Footprint
Zero freshwater consumption. Zero land footprint. 90% less cooling energy. 40% total energy reduction compared to land-based facilities.
Microsoft's Natick project showed that sealed underwater containers can operate for years with zero environmental impact on surrounding marine ecosystems.
Powered by offshore renewable energy (wind, tidal, wave), underwater data centers can achieve true carbon-neutral operation — something orbital systems cannot claim due to launch emissions.
Maintenance & Repairability
When hardware fails — and it will — the ability to repair it determines the economic viability of the entire system.
Orbital: Disposable by Design
There is no maintenance in orbit. When a satellite fails, it is deorbited and burned up. A replacement must be manufactured and launched — at a cost of $500–$3,600 per kilogram.
With a 9% annual GPU failure rate and 2.5% solar degradation, a 1 GW constellation would require continuous replacement launches costing billions per year.
Underwater: Accessible & Repairable
Underwater data centers can be retrieved by crane ship in hours. Microsoft retrieved their Natick pod after 2 years for analysis and component inspection.
Modular pod design allows individual units to be swapped without affecting the rest of the facility. Maintenance cost is a fraction of the original deployment — not a complete rebuild.
The Cunabula Salutis Advantage
Our deep-water city infrastructure provides the perfect platform for next-generation underwater data centers — with capabilities no standalone subsea pod can match.
Integrated Power
Our Small Modular Reactors (SMR) provide dedicated, reliable power at the point of use. No submarine cables stretching hundreds of kilometers. No dependency on surface weather for renewable energy. Constant, clean, abundant power.
Deep-Water Cooling
At 300–3,000m depth, water temperature is a constant 1–4°C. Combined with our pressure-equalized hull design, we achieve cooling efficiency that surface-level subsea pods cannot match. PUE approaching 1.01.
Physical Security
Data centers inside our deep-water bastions are protected by hundreds of meters of water — immune to physical attack, natural disasters, and electromagnetic pulse. The ultimate air-gapped secure facility.
Permanent Staff
Unlike standalone subsea pods, our underwater cities have permanent human residents. Hardware maintenance, upgrades, and emergency repairs can be performed on-site — no expensive surface vessel operations required.
Revenue Generation
Hosting data centers provides a continuous revenue stream that helps fund the underwater city itself. Tech companies pay premium rates for the most reliable, secure, and energy-efficient compute infrastructure on Earth.
Scalability
Our modular architecture allows data center capacity to scale with demand. New compute pods can be added to existing city infrastructure without building entirely new facilities — true elastic scaling.
The Bottom Line
Orbital Data Center (1 GW, 5yr)
$51.1 Billion
Experimental technology. No maintenance possible. Massive environmental impact. Years to deploy. Physics-limited cooling.
Underwater Data Center (1 GW, 5yr)
$5–8 Billion
Proven technology. 8× more reliable. 90% cooling savings. Months to deploy. Repairable. Sustainable.
The orbital data center is a solution looking for a problem. The underwater data center is a proven solution to the biggest problems facing the computing industry today: energy consumption, cooling costs, reliability, and sustainability. The question is not whether underwater data centers will become mainstream — it's how quickly the industry will recognize that the answer was always beneath the surface.
Sources
- TechCrunch — "Why the economics of orbital AI are so brutal" (February 11, 2026)
- Andrew McCalip — Economics of Orbital vs Terrestrial Data Centers (andrewmccalip.com/space-datacenters)
- Microsoft — "Project Natick: Underwater datacenter is reliable, practical, and uses energy sustainably" (September 2020)
- SpaceX FCC Filing — Orbital data center constellation of up to 1 million satellites (January 31, 2026)
- Google Research — "Project Suncatcher: Space-based AI infrastructure" (November 2025)
- Tom's Hardware / Barron's — China Highlander underwater data center, 90% cooling energy reduction (October 2025)
- IEEE Spectrum — "How Stupid Would It Be to Put Data Centers in Space?" (2025)
- Wired — "China Dives in on the World's First Wind-Powered Undersea Data Center" (October 2025)