Musk wants ai servers in orbit: starship must first stop burning money
Elon Musk’s weekend tweet about orbital data centers has Fidelity running spreadsheets on kilowatt-hours in zero-g. The bet: if Starship can loft 100 t at a price that doesn’t make accountants weep, the vacuum above Texas becomes the cheapest real estate in computing.
The spreadsheet that orbits
Karin Fronczke, the Fidelity investor who has tracked SpaceX since 2015, keeps a model that translates every successful landing into cents shaved off a kWh. She showed it to Bloomberg TV on Monday. The sheet turns green only when launch cost drops below the price of terrestrial electrons plus the NIMBY lawsuits Musk is trying to escape. It is still red.
SpaceX is about to light Starship for a third time. A success would let engineers stop talking in conditional tense. A fireball on the pad would push orbital server racks into the same drawer as Mars condos and hyperloops—fun renderings, lousy IRR.
From 13 flights to every 48 hours
Fronczke’s slide deck is brutal. In 2015 the Falcon 9 flew 13 times; this year a booster leaves Cape Canaveral every other day, hauling 80-90 % of the planet’s off-world cargo. Starlink, once a side hustle, just crossed 10 million subscribers. The cadence is already there; the reuse is not. Starship must survive re-entry, belly-flop, catch the chopsticks and do it again the next morning. No silicon customer will ship a $200 million GPU cluster on a single-use rocket.
Musk’s answer is vertical integration squared. Terafab, the new Austin foundry he announced Saturday, will churn out custom silicon for robots, cars and, yes, orbital racks. Tesla handles the wafers; SpaceX handles the launch window. The plant rises a short drive from both headquarters, letting engineers swap engineers instead of emails.
The vacuum as landlord
Orbital data centers would sip unlimited solar power, dump heat into the infinite black and never hear a zoning complaint. The concept is old; NASA sketched it in the 1970s. The difference is a spreadsheet that now says “maybe” instead of “never.” Fronczke underlines the caveat: “Conceptually it makes sense. The question is profitability.”
The clock ticks. Every rival cloud provider is buying land near cheap hydro dams. Every delay gives terrestrial grids time to green themselves. If Starship doesn’t nail the next landing, the vacuum stays empty and the servers stay on mud.
Fronczke already knows the number that ends the debate. She will not quote it on camera. But she smiles when she says Fidelity updates the cell after every launch. The next ignition is weeks away. The color of that cell depends on flame, steel and a catch tower standing in the Texas fog.