AI compute demand is doubling faster than the power grid can expand. The United States faces a projected 100-gigawatt deficit from data centers alone by 2030. Water consumption for data center cooling is doubling every 36 months. Land-use conflicts are pushing new construction into permitting delays measured in years.

These are physics constraints, not speculative problems.

Jeff Brown says the solution is in orbit.

The Space Data Center Thesis

SpaceX filed a confidential application with the Federal Communications Commission for an orbital data center system. The filing describes a constellation of satellites operating in low-earth orbit, each approximately 150 kilowatts of compute capacity, connected by laser interconnects and powered by solar arrays with liquid radiator cooling.

The filing is the piece that separates this thesis from speculation. A company does not file with the FCC for a system it does not intend to build. The 1-million-satellite scale is the kind of number that sounds like marketing until you remember that SpaceX already deployed 9,357 Starlink satellites, onboarded 9 million users, and turned Starlink into a profitable business.

The orbital data center applies the same playbook to compute instead of bandwidth. The same hidden-supplier structure drives the thesis. Starlink proved that SpaceX can launch and operate a massive satellite constellation at a scale no other company has matched. The orbital data center is what comes next.

Why Orbit

The argument for putting AI compute in space is about physics.

Terrestrial data centers consume enormous amounts of electricity and water. A single large data center can draw 100 megawatts — enough to power 80,000 homes. The heat generated by the compute equipment requires extensive cooling systems that consume water at industrial scale. The land required for a data center campus removes that land from other uses.

Space eliminates all three constraints at once. Solar power in orbit is continuous and unlimited. Cooling is passive — the near-absolute-zero vacuum of space is the ultimate heat sink. And with no land, there is no land-use conflict.

The barrier has always been launch cost. Sending equipment into orbit is expensive. The cost per kilogram has come down dramatically — from $10,000 during the Space Shuttle era to roughly $2,000 on the Falcon 9. Brown argues that Starship V3 will bring launch costs to $200 per kilogram, which is the threshold where orbital data centers become cheaper than terrestrial ones.

The gap between $2,000 and $200 is the entire thesis. If Starship V3 delivers, the economics work. If it does not, the orbital data center remains an engineering curiosity.

The 106X Claim

Brown says the orbital AI market represents a 106X opportunity. The number refers to the satellite count expansion versus the current Starlink constellation — roughly 1 million orbital data center satellites versus Starlink’s deployed base. The claim is about the scale of the infrastructure buildout, not the return on any single investment.

A 106X increase in satellite count means the market for space-based compute infrastructure is orders of magnitude larger than the current market for satellite-based broadband. The question for an investor is which companies supply that infrastructure.

The Radiation Problem

Orbital data centers face a problem that terrestrial data centers do not. The radiation environment in low-earth orbit destroys commercial silicon. Cosmic rays, solar particles, and trapped radiation belts degrade semiconductor performance over time. A standard data center chip that would last 20 years on Earth might fail in weeks in orbit.

The solution is radiation-hardened chips — semiconductors designed and manufactured to survive the space radiation environment. These chips use specialized manufacturing processes that make them more resistant to radiation damage. They are more expensive to produce and less powerful than their commercial equivalents, but they work in orbit.

The supplier of these chips is the bottleneck. Without rad-hard chips, orbital data centers cannot function. The company that makes them has already shipped billions of chips into the SpaceX supply chain. The relationship exists at scale.

SpaceX has confirmed the chip shortage in its IPO filing disclosure — an independent validation that the component constraint is real.

The Bigger Picture

The orbital data center thesis fits into a broader pattern that Brown has been tracking for years. The bottleneck in every technology transition is physical, whether that means the data centers cloud computing needed, the GPUs AI needed, or the rad-hard chips space-based compute needs. The companies that make the physical components are the picks-and-shovels plays.

SpaceX has already proven the satellite manufacturing model with Starlink. The orbital data center is the same factory producing a different product. The FCC filing is a real document and the chip constraint is a real problem. Whether Starship V3 can deliver $200 per kilogram launch costs — and which company has the chip relationship that makes the whole thing possible — is the open variable. The SpaceX IPO thesis and the gurus tracking it maps the infrastructure buildout this orbital data center thesis depends on.

The idea is worth sitting with. More campaign breakdowns in the Promo Watch.