Building the First Data Centers in Space

Why does it matter? Because one number separates every viable orbital business from science fiction

The whole case for space data centers lives in a single break-even calculation. Space-based solar needs $50/kg launch cost to work. Orbital compute needs $500/kg. That second number is nearly within reach. The gap explains why StarCloud exists, why Philip Johnston abandoned his original idea after two months, and why every other orbital business category is still waiting.

• Space solar loses 95% of its energy in wireless transmission back to Earth — been "almost viable" since the 1940s, won't be viable soon • Data centers are the only orbital payload that skips re-entry — the expensive chokepoint that makes asteroid mining, space manufacturing, and space hotels all unworkable • Legacy aerospace quoted $75–100M for what StarCloud built and launched for $2M • Terrestrial data center construction is being politically blocked in New York and several other states — space may soon be the only legal place to build at scale

Space-based solar destroys 95% of its energy in transit — and data centers were always the right answer

Space-based solar loses 95% of its energy in transmission from orbit to Earth. That's not a solvable engineering problem — it's physics. StarCloud spent its first two months trying to solve it anyway, operating as Lumen Orbit, before concluding the right move was to stop transmitting energy and keep the computation in orbit instead.

The break-even launch cost for space solar: around $50/kg. For data centers in orbit: $500/kg — "much closer to reality," as Johnston put it. Both ideas would have seemed equally futuristic in 2023; one of them is nearly fundable today.

By late 2023, almost all new energy projects were being built for data centers. If solar power generated in orbit was going to feed data centers on Earth anyway, the data centers might as well stay in orbit. They spent a month on the recalculation, then renamed the company.

Run the break-even launch cost calculation before committing to any orbital business idea. If the number is $50/kg, it's science fiction. If it's $500/kg or better, it might be investable now.

Re-entry is the constraint that makes asteroid mining, space manufacturing, and space hotels all unworkable — and data centers sidestep it entirely

Re-entry is the cost that makes nearly every orbital business unworkable — and it doesn't get cheaper no matter how far launch costs fall.

Varda is doing space manufacturing. Astroforge is mining asteroids. Space hotels are in the pipeline. All of them require the product to come back to Earth: ore needs to land, manufactured goods need delivery, hotel guests need to return. Re-entry is extraordinarily expensive and technically brutal — a cost paid on every cycle.

Data centers escape it entirely. They process requests from Earth and send results back as bits over Starlink laser terminals. Nothing physical comes back. That single structural advantage — no re-entry requirement — is what makes orbital compute the logical first commercial product in space while everything else waits.

Aerospace primes quoted $75–100M. StarCloud did it for $2M, including the launch.

One aerospace prime quoted $75 to $100 million to build what StarCloud launched for $2 million. The gap isn't dangerous corner-cutting — it's procurement convention.

Space-grade rad-hard components carry massive premiums from an assumption that every part must be individually certified for the space environment. StarCloud uses automotive-grade off-the-shelf hardware instead, ships batches to particle accelerators at Brookhaven National Lab and a proton cyclotron in Knoxville, and selects whichever survive. They believe this is part of how SpaceX drove down satellite costs too.

At 5am the night before shipping the satellite, co-founder Addy was dunking an H100 GPU in an ice bath for thermal cycling — hot air guns to melt the phase-change wax, then back into the bath. Johnston's verdict: "It is a miracle that it works." StarCloud now claims to be the only team in the world with test data on exactly where H100s, B200s, and H200s fail under high-velocity protons and heavy ions.

Book the launch before you know what you're building — the deadline does the work that planning never will

Book the launch before you know what you're building — before the team is full, before the payload is decided, before the satellite design exists. The deadline forces more engineering progress than any planning process.

January 1st, 2024: StarCloud founded. January 2nd: $300,000 committed to a SpaceX rideshare, 18 months out. No design, no confirmed payload, just a date and a check.

Something has to be on that rocket — which drives the question, which drives the engineering. The original plan was a Jetson chip, a conservative choice that had flown on satellites before. When Addy joined from SpaceX, he called it lame and swapped in an H100. That upgrade was only possible because the booking surfaced a concrete problem early enough to actually solve it.

Hardware founders: find your equivalent of booking a launch.

The real moat may not be cost — it may be being the last legal option

StarCloud originally needed to beat terrestrial data centers on price. That case is still being assembled. But New York has now blocked new data center construction. Several other states are following, for reasons Johnston describes as "more like vibes than anything grounded in science," as AI compute simultaneously becomes a declared national security priority.

If those curves cross — political restriction accelerating, compute demand exploding — the company building in orbit wins not by beating terrestrial economics but by being the last buildable option. The strongest moat isn't always cost. Sometimes it's being the only thing left standing.

---

Topics: space tech, data centers, AI infrastructure, hard tech, YC startups, SpaceX, orbital computing, energy, deep tech fundraising, startup strategy