Every transformative infrastructure wave in history has had a moment when the binding constraint disappeared. In the early 1900s, it was the power grid that freed factories from building their own generators. Right now, AI is hitting its own version of that wall — and the solution being built is not on the ground. It is in orbit.
The core problem is straightforward: AI data centers require land, power, and water in massive quantities, and all three are becoming acutely scarce. Bloomberg estimates that nearly half of all planned AI data center projects in the United States will be delayed this year due to power grid constraints alone. Interconnection queues now stretch three to five years. Microsoft, Amazon, and Alphabet collectively have the capital and the chips — they just cannot get a power connection fast enough to put them to work.
Enter orbital compute. Solar panels in low Earth orbit receive roughly 1,400 watts per square meter of raw energy, compared to roughly 1,000 watts at Earth’s surface under ideal conditions — and with none of the night cycles, weather losses, or atmospheric drag that slash terrestrial solar averages to a fraction of that. The deep cold of space provides essentially free heat dissipation, eliminating the billions of gallons of water that Earth-based data centers burn through annually just to keep GPUs from melting. And satellite-based compute sits closer to the data it processes: a massive share of AI workloads analyze satellite imagery, defense telemetry, and weather data that originates in orbit anyway.
This is no longer science fiction. In November 2025, Starcloud launched the first Nvidia H100 GPU into orbit. In December, that satellite trained a language model in space. In March 2026, Nvidia unveiled the Vera Rubin Space-1 module — a chip platform purpose-built for orbital data centers — at its GPU Technology Conference. Jensen Huang said on stage: “Space computing, the final frontier, has arrived.”
Then in April, SpaceX confidentially filed with the SEC for an IPO targeting a $1.75 trillion valuation — what would be the largest public offering in market history — with orbital compute as the central investment thesis. The company had already acquired xAI in February 2026 in the largest private merger on record, combining Elon Musk’s rocket infrastructure with his AI ambitions under one roof.
The economics are still expensive: running a GPU-hour in orbit costs roughly $142 today versus about $1 on Earth, and about 60% of that premium is pure launch cost. But launch costs have fallen over 90% in a decade, and SpaceX’s Starship — designed specifically to drive costs toward $10 per kilogram to orbit — changes the math dramatically if it hits its targets. The thesis is a bet on launch cost curves continuing their trajectory, not on orbital compute being cheap today.
For investors, the SpaceX IPO — if it proceeds — would be the defining event of the decade. It is rare that a company this large, this capital-intensive, and this genuinely transformative reaches the public market. The companies positioned in the orbital compute supply chain — satellite components, high-efficiency space-grade solar, laser communication hardware — are worth watching closely now, before the IPO story fully captures mainstream attention.
