Elon Musk and Jeff Bezos have publicly advocated deploying data centers in orbital platforms to sidestep Earth-based power constraints and thermal bottlenecks for AI training. Musk has suggested orbital infrastructure could bypass terrestrial energy limitations, while Bezos has positioned Blue Origin as a logistics provider for such ventures. Business Insider reported that physics researchers are raising hard objections: cooling systems in vacuum require radiative heat dissipation at vastly larger surface areas, launch costs remain $1,500-$15,000 per kilogram depending on provider, and reliability metrics for mission-critical compute in radiation-heavy LEO orbit lag decades behind ground-based data centers.

This represents billionaire-driven technological escapism dressed as innovation. The unit economics are brutally unfavorable. A single NVIDIA H100 GPU costs $40,000 and weighs roughly 2kg. Launching it to orbit via SpaceX Falcon 9 costs $30,000-$300,000 in marginal launch fees alone, before integration, redundancy, thermal management, and replacement cycles. A terrestrial data center pod costs $5-15 million and houses 1,000+ GPUs. Amortized cost per FLOP in orbit is orders of magnitude worse. Moreover, thermal engineering in vacuum is unsolved at scale: radiators must be massive, deployable, and survive micrometeorite impact. Musk and Bezos are leveraging their aerospace credentials to attract venture capital and government subsidies for ventures that lose on first-principles physics.

Loses: Bezos (Blue Origin), Musk (SpaceX's marginal launch economics worsen if they absorb orbital data center risk), venture firms backing space compute startups. Wins: Terrestrial hyperscalers (MSFT, GOOG, TSMC) deepening moat through cheaper density. Neutral: governments can extract subsidies for 'space infrastructure' while knowing the economics don't work.

Watch whether any orbital data center project launches with >10 MW continuous power draw. Watch SpaceX's actual pricing for payloads >50 tons to GEO or LEO. Watch NVIDIA's Q-filing language on 'emerging compute architectures'. If real capital flows exceed $500M by 2027, it signals regulatory capture rather than genuine technical viability. The tell: no commercial data center operator will voluntarily abandon a $1M/MW/year terrestrial facility for $100M+/MW orbital equivalent.