There is a number that explains why Phantom Space bought a thermal radiator company.
Rejecting one kilowatt of heat in orbit requires roughly 2.5 square meters of radiator surface. Not because of engineering preference, but because of physics: there is no air in space, no convection, no rivers, only vacuum and the cold of the cosmic background. A 100-kilowatt AI processing payload, the scale the industry keeps announcing, needs a radiator the size of a small studio apartment. That is not a packaging problem. That is a hard ceiling on what orbital compute can actually do. Phantom Space's answer is to stay small. It is building 10-kilowatt-class micro data centers, not 100-kilowatt inference farms. And in early April, it acquired Thermal Management Technologies, the Utah company that makes the radiators that make 10 kilowatts survivable.
The acquisition looks like vertical integration. It is also a physics lesson. TMT founder Scott Schick has worked with Phantom CEO Jim Cantrell for over 40 years, dating back to their time together at Space Dynamics Lab at Utah State University. He stays on as TMT's general manager. The two companies had already worked together on a customer spacecraft delivered in 2023, which gave Phantom enough confidence to buy the whole thing.
"High-density AI or data-processing payloads can produce tens or even hundreds of kilowatts, which must be collected, transported and dissipated efficiently without adding too much mass or compromising the spacecraft's structural integrity," Cantrell told SpaceNews. TMT integrates thermal and structural functions into shared hardware, promising more compact designs than separate radiator panels. Cantrell describes the result as shaving months off development time by letting Phantom's thermal, structural, bus, payload, and power teams iterate together instead of passing documents across corporate boundaries.
The TMT deal is the third acquisition in Phantom's vertical stack. It bought StratSpace in 2021 for services, and it acquired Vector Launch assets in February 2026, funded by a previously unannounced Series B that closed late 2025 in the eight-figure range. The pattern is the same every time: identify a dependency, buy the company that fills it, eliminate the supplier negotiation. That logic works. The question is whether Phantom has the capital and time to make it work twice. The company is building a rocket and a constellation simultaneously. Daytona is a 180-kilogram-to-LEO vehicle with nine Hadley engines on the first stage and one Hadley Vacuum on the second, targeting a maiden flight in the second half of 2026 that has already slipped from earlier schedules. The Block I demonstration mission for Phantom Cloud is targeted for the second half of 2027, two spacecraft launched by a third party, ahead of a 66-satellite constellation that has not yet been filed with the FCC.
Phantom Cloud itself is positioned differently from Starcloud and SpaceX's orbital computing ambitions. Those companies are targeting full AI inference at scale, which requires solving power generation, thermal management, and radiation hardening simultaneously at data-center density. Cantrell frames Phantom as an open orbital data platform for data backhaul, edge processing, and cloud-like storage accessible to government agencies, commercial Earth observation operators, scientific missions, and satellite owners. Early partners include Ubotica for edge AI processing, Assured Space Access for RF communications payloads, and Secured2 for quantum-secure data protection. The scope is narrower, and that narrowness may be the point: edge processing workloads don't need the power density that makes orbital thermal management a crisis.
This is the unsexy problem at the center of the orbital computing race. The announcements are easy. The heat is hard. Demidov of Beyond Earth Ventures wrote recently in SpaceNews that the industry is missing power as its biggest variable, and he is right. On Earth, AI hyperscalers are signing multi-gigawatt nuclear power purchase agreements because grids cannot keep up. In orbit, the physics of solar power are more favorable, but converting that advantage into reliable kilowatts for computing hardware requires engineering that does not yet exist at the scales being announced. Phantom is attacking one constraint directly by owning its thermal solution. Whether it can solve the rest on a startup budget and a delayed timeline is the question that will determine whether this vertical stack is strategy or slideware.
The 2027 demonstration mission will answer some of it. Phantom says it is already seeing strong interest, with larger commitments under discussion for the full 66-satellite constellation. Cantrell is careful about specifications, which is appropriate given how many times the industry has announced its way past physics. The radiator math does not lie. Ten kilowatts is survivable. One hundred kilowatts, at today's thermal technology, is not.
