Based on my research, orbital data centers are **technically possible but face significant challenges that make them questionable as a widespread solution in the near term**. Here's what I found: ## **Current Status: Possible but Early Stage** - **Starcloud** (formerly Lumen Orbit) successfully launched a satellite with an NVIDIA H100 GPU in November 2025 and trained an AI model in orbit—an industry first - **SpaceX** filed plans for a constellation of up to 1 million satellites for orbital computing in January 2026 - **Google's "Project Suncatcher"** and **Axiom Space** are also developing orbital data center prototypes - The market is projected to grow from $1.77 billion by 2029 to $39 billion by 2035 ## **Advantages** - **Continuous solar power** in certain orbits (sun-synchronous orbits get 99%+ sunlight) - **Natural radiative cooling** in vacuum—no water needed - **Avoids terrestrial constraints** like grid dependence, water scarcity, and zoning issues - **Reduced latency** for space-based applications (satellite data processing) ## **Major Challenges & Why Experts Are Skeptical** ### **1. The "Physics Wall" - Cooling is the Biggest Problem** - Space is a near-perfect insulator—heat can only be radiated away as infrared light - A 1 MW orbital data center would need **~1,200 m² of radiator area** (four tennis courts) - Radiators add massive mass and complexity, requiring plumbing and pumps that can fail - Running radiators hotter reduces area but shortens hardware life ### **2. Economic Reality** - **Launch costs**: Even at $3,000/kg (Falcon 9), launching thousands of compute satellites costs vastly more than terrestrial facilities - **Replacement cycles**: Radiation damage means frequent satellite replacements - **Limited repairability**: Most satellites can't be serviced in orbit ### **3. Technical & Operational Hurdles** - **Radiation damage**: Electronics degrade faster in LEO, requiring expensive radiation-hardened components - **Latency**: LEO adds 10-100ms round-trip latency vs. terrestrial fiber—problematic for real-time AI - **Orbital congestion**: Adding millions of satellites increases collision risk and Kessler cascade potential ### **4. Expert Opinion** - **Elon Musk** claims orbital AI could be cheaper than terrestrial within 2-3 years - **MIT professor Olivier de Weck**: A 100 MW facility would be 500-1,000× larger than the ISS—feasible but "not next year and certainly not in three years" - **Carnegie Mellon professor Brandon Lucia**: Calls Musk's timeline "an optimistic interpretation" - **Fortune/PCMag experts**: Major safety and technical hurdles remain ## **Verdict: Niche Applications Only** Orbital data centers **are possible** but **not a good idea for general-purpose computing** in the foreseeable future. They may make sense for: - **On-orbit processing** of satellite imagery to reduce downlink bandwidth - **Edge compute** for space-based instruments - **Highly secure, physically isolated computation** - **Scientific missions** requiring compute near sensors For mainstream AI training and cloud services, terrestrial data centers with improved efficiency, liquid cooling, and renewable energy remain the better economic and operational choice. The hype currently outpaces the practical reality.