The supply chain thesis behind Jeff Brown’s Orbital AI pitch comes down to two components nobody outside the satellite industry talks about. Radiation-hardened chips and laser interconnects. One solves the physics problem of running silicon in orbit. The other solves the data problem of linking thousands of satellites into a single compute fabric.
The campaign-level framing and the trust check on Brown’s engineering background are covered in the parent Orbital AI analysis; this piece breaks down the supply chain layer, which is where the investment thesis actually lives.
What an orbital data center needs
A terrestrial data center is a building full of server racks, connected by fiber optic cables to the internet, powered by the grid, and cooled by water and air. An orbital data center is the same architecture with every component re-engineered for space.
Each satellite in the constellation carries roughly 150 kilowatts of compute capacity. That is about the draw of a small commercial building, packed into a satellite bus. The compute racks run on silicon. The power comes from solar arrays. The cooling is passive radiation into the vacuum of space through liquid radiator loops. The networking happens through laser interconnects rather than fiber. Every one of those components has a space-grade version that is different from the terrestrial equivalent.
The supply chain thesis asks which of those components is the bottleneck. Brown’s argument is that it is the chips.
Why rad-hard chips are the bottleneck
Commercial silicon does not survive in orbit. The radiation environment in low-earth orbit includes cosmic rays, solar particle events, and trapped radiation belts. A standard data center chip designed for a terrestrial facility might operate for 20 years on Earth. The same chip, placed in a satellite, degrades over weeks.
The mechanism is well-documented. High-energy particles flip bits in memory, corrupting stored data. Ionizing radiation accumulates in semiconductor junctions, degrading transistor performance until the chip fails. The total ionizing dose a chip receives in orbit depends on altitude, shielding, and solar activity, and the numbers at the altitudes SpaceX targets are high enough that commercial-grade silicon is not a viable option.
Radiation-hardened chips solve this through specialized manufacturing processes. Insulated substrates, redundant circuitry, and wider transistor geometries make the chips resistant to radiation damage. The trade-off is that rad-hard chips are more expensive to produce and run slower than their commercial equivalents. A rad-hard processor might be a generation or two behind a top-end Nvidia GPU in raw performance, but it works in orbit, and the commercial part does not.
That is the bottleneck. Orbital data centers cannot run on the same silicon that terrestrial facilities use. They require a specialized component manufactured by a small number of companies with the qualified processes to produce it. The supplier that has already shipped rad-hard chips into the SpaceX supply chain at scale is the structural position in this thesis.
The laser interconnect network
The second component is the data link between satellites. A terrestrial data center networks its racks with fiber optic cables running at hundreds of gigabits per second. In orbit, there is no fiber. The satellites have to talk to each other through free-space optical links, laser beams carried between satellites across hundreds or thousands of kilometers of vacuum.
SpaceX already operates laser interconnects in the Starlink constellation. The technology is proven at scale. Each Starlink satellite carries four optical interconnect terminals, and the network routes traffic between satellites without bouncing down to ground stations. The orbital data center architecture extends the same networking model to compute traffic.
The supply chain question is who makes the optical terminals and the underlying components. The laser interconnect market is smaller than the rad-hard chip market, and the qualified suppliers are a handful of companies with heritage in satellite communications. The investment thesis treats this as a secondary component rather than the primary bottleneck, because the laser interconnect technology is already deployed in Starlink, while the rad-hard compute silicon is a new requirement driven specifically by the orbital data center architecture.
The $200 per kilogram threshold
The economics that make any of this viable come down to launch cost. During the Space Shuttle era, sending a kilogram to orbit cost roughly $10,000. On the Falcon 9 today, the figure is down to about $2,000. Brown’s thesis is that Starship V3, the fully reusable next-generation vehicle SpaceX is developing, will bring that cost to $200 per kilogram.
At $2,000 per kilogram, an orbital data center cannot compete with a terrestrial one on cost. The launch expense overwhelms the operational savings from free solar power and passive cooling. At $200 per kilogram, the math inverts. The orbital facility becomes cheaper to deploy and operate than a ground facility with equivalent compute capacity, for the classes of compute that can tolerate the latency of an uplink.
Starship V3 has not yet demonstrated the $200 per kilogram figure in operational flight. The number is an engineering projection based on the vehicle’s designed payload capacity and the cost savings from full reusability. Projections are not flight data. The thesis inherits the risk that Starship V3 underperforms its spec, launches less frequently than projected, or costs more per mission than the design assumes. If the threshold is not reached, the orbital data center remains an engineering possibility that has not yet closed its economics.
The picks-and-shovels logic
The supply chain structure here is the same one Brown has used across his career. Find the enabling component that makes the headline technology possible, identify the company that manufactures it, and invest before the market prices in the demand. The method traces across the Nvidia, Bitcoin, and Tesla calls in his published track record. Past performance does not guarantee future results.
In the Orbital AI thesis, the headline technology is the orbital data center. The enabling component is the rad-hard chip. The company that makes it is the picks-and-shovels position, because every operator of an orbital compute constellation needs the same component from the same small set of qualified suppliers. That is the structural argument, independent of which specific company Brown names in the paid report.
The variable the thesis inherits is whether Starship V3 delivers the launch cost the economics require. The filing, the physics, and the component constraint are all real. The launch cost threshold is the open question, and it is the one variable the supply chain logic cannot answer from inside the supply chain. The component supplier exists whether the constellation deploys at full scale or not, but the demand curve that makes the supplier a thesis depends on the constellation actually flying, and that depends on a vehicle that has not yet proven its cost in operational service.
The idea is worth sitting with. The structural case for a rad-hard chip supplier to a space-based compute buildout does not require the 106X satellite count to materialize on the timeline the presentation implies. It requires the architecture to be built at any meaningful scale, and the FCC filing is the evidence that SpaceX intends to build it. Which specific company occupies the bottleneck position is the question the paid report answers, and it is a separate question from whether the bottleneck itself is real.
The Promo Watch board tracks the rest of the current campaigns we cover.