How Do Orbital Data Centers Dissipate Heat? The Physics

How do orbital data centers dissipate heat? It is the single best question to ask about the “space AI” thesis, because the answer determines what kind of chips can fly, which suppliers matter, and how much of Dylan Jovine’s “Elon’s 100X AI Stock” pitch is physics versus marketing. We read the full presentation, and the heat question turns out to be the hinge of the whole argument.

The Counterintuitive Problem: Space Is an Insulator

The pitch sells the cold of orbit first: “Space is 250 degrees below zero, and it never warms up.” That is true, and it sounds like free cooling. But there is a catch the presentation addresses directly, and it is the crux of the physics. On Earth, a data center moves heat off its chips with air or liquid, carrying it away by convection. In orbit there is no air. A vacuum cannot carry heat anywhere. The only way waste heat leaves a satellite is thermal radiation: the facility must glow its heat into space across large radiator panels.

So the honest answer to the question is: orbital data centers dissipate heat through radiation, across radiator surfaces that scale with the amount of compute you fly. The pitch puts it plainly when it notes a chip in a vacuum can only be cooled by “enormous” radiators. Every watt a chip burns as waste heat is a watt that has to be radiated away, which means more panel mass, more launch cost, and less compute per kilogram. The 250-below-zero environment does not eliminate cooling engineering; it changes it from a plumbing problem to a surface-area problem.

Why This Makes Power Efficiency the Whole Game

Follow the logic one step and you land on Jovine’s investment thesis. If radiator size bounds orbital compute, then the most power-efficient chip architecture is the gate on what AI you can fly at all. The pitch’s phrasing: the teased supplier provides “the only chip that can run AI in orbit without cooking itself.” That is a sales claim about an unnamed company, and we flag it as such; the identification is unconfirmed, and we do not assert a name. But the underlying engineering claim is sound: efficiency is king in orbit in a way it never was on the ground, because on the ground you can always buy more air conditioning and grid power is the constraint, while in orbit you buy radiator mass with every extra watt of heat.

This is also why the terrestrial setup is described as broken in the pitch: the four biggest hyperscalers are pouring $725 billion into AI infrastructure in 2026, while Morgan Stanley projects a 49-gigawatt power shortfall for data centers by 2028, Google waits up to 12 years for grid hookups, and cities from Denver to Tulsa have paused or banned new data center construction. Musk’s stated answer, quoted twice in the pitch: “The cheapest place to put AI will be in space within 36 months or less.” Orbital solar helps on the supply side (“a solar panel in space generates roughly eight times the energy of the same panel on the ground”); radiation is the only exit on the exhaust side. The AI1 satellite design in the pitch, with “solar wings stretch[ing] 230 feet from tip to tip,” longer than a Boeing 747’s wingspan, is essentially a giant power-and-radiator bus wrapped around a compute core.

The Supplier Logic Built on Top of the Physics

The pitch’s money line connects the physics to the portfolio: “Elon doesn’t own that chip. He licenses it. That means, every time a Tesla rolls off the line… Every time SpaceX launches an AI1 satellite… This company gets paid.” Jovine says Musk has been “quietly paying this obscure firm since 2012,” and with the FCC request to go from roughly 10,000 satellites to one million, the bill to that firm “could soon get a LOT bigger. 100X bigger.”

That chain deserves honest parsing. The royalty-per-unit model is real and well documented as a business structure. The switching-cost argument is real: leaving an architecture means “rewriting every line of software in every Tesla and every satellite he owns.” The unverifiable parts are the specifics: which company, what royalty rate, whether the data center royalties really “more than doubled last year,” and whether the December 8th lockup expiry, which frees “roughly 800 million insider shares” of SpaceX, actually redirects capital into suppliers as Jovine argues. We examined that event in our SpaceX lockup expiration coverage, and the broader theme in our orbital data center investment piece.

What Checks Out and What to Watch

The physics is the strongest part of this pitch: radiation cooling, orbital solar yields, and the efficiency-first design constraint are all legitimate engineering. The race is real too, with the Nvidia-backed Starcloud having trained the first AI model in space and Google’s Project Suncatcher in motion. What to watch with a cool eye: the “36 months” timeline is Musk’s, not a construction schedule; the one-million-satellite figure is a filed request, not a deployed fleet; and the teased chip supplier is unnamed, with a fingerprint (a chip-IP licensor on per-unit royalties, a Tesla relationship since 2012) that we treat as unconfirmed.

The offer around the thesis is $49 for a year of Behind the Markets, with a six-month full-refund guarantee and four research reports including the one naming the stock. If the thermal physics angle interests you, the full presentation is here.

NewsletterVetter is an independent publication. We receive compensation from some of the services we review through affiliate links. Nothing on this site is investment advice. Always do your own research.