**Major tech firms are advancing orbital AI compute to bypass terrestrial power and cooling constraints, but qualifying deployments remain years away.** SpaceX filed for up to one million AI1 satellites in early 2026 with detailed specs released in June, targeting prototypes in early 2027 and commercial operations later that year, while leveraging Starlink heritage for solar arrays and radiative cooling. Starcloud demonstrated the first orbital LLM training on an NVIDIA H100 in November 2025 and recently raised $250 million toward an 88,000-satellite constellation. Google’s Project Suncatcher and Blue Origin’s Project Sunrise have similar FCC filings and 2027 prototype targets, with companies like Orbital and Sophia Space pursuing modular GPU platforms and patents for passive heat rejection in vacuum. Current low implied probabilities for 2026–2027 reflect the gap between single-GPU demos and the market’s 100+ data-center-grade accelerator threshold, plus pending regulatory approvals, launch cadence limits, and economic hurdles for gigawatt-scale LEO constellations. Key catalysts ahead include 2027 prototype launches and FCC decisions that could accelerate or delay first qualifying orbital data centers.