Musk Says SpaceX's Ground Data Centers Will Look Tiny Once Starmind Goes Orbital

Elon Musk is trying to reframe a debate that has been rattling SpaceX investors: how much the company can afford to spend on compute. His answer, posted to X, is that the market has the numbers wrong — and that the whole ground-based buildout is just the down payment on something far bigger in orbit.

Musk was reacting to a post from Milk Road Stocks that leaned on analyst Kyle Reidhead’s estimate of roughly $470 billion in SpaceX capex across the next five quarters. That figure was reverse-engineered from SpaceX’s own third-quarter earnings call, where management targeted 13 GW or more of compute capacity by the close of 2027.

Musk’s rebuttal was blunt: the market, in his view, badly underestimates how efficiently SpaceX converts capital into compute. He did not offer his own number.

The Math Behind the Worry

Investors have reason to run the projections. SpaceX closed the quarter with 1.7 GW of nameplate compute and told shareholders to expect $35 billion to $40 billion of capex in the fourth quarter alone. Reidhead’s breakdown has Nvidia taking somewhere between 83 and 89 percent of that spend.

Cash flow, though, changes the picture. CFO Bret Johnsen said during the call that fresh compute pays for itself in under a year at current pricing, and Musk suggested that payback window could tighten to six months during 2027. Reidhead’s model follows that logic: roughly $230 billion of the $470 billion could be covered by operating cash flow, with about $60 billion drawn from SpaceX's stockpile of around $90 billion, leaving some $180 billion to be borrowed.

The figures in play: ~$470B in projected capex over five quarters · 1.7 GW of compute today · 13 GW targeted by end-2027 · $40B debt raise already closed for GPUs.

Musk confirmed last week that SpaceX has already locked in a new $40 billion debt facility earmarked for graphics processors.

Ground Compute Is Just the Warm-Up

Here is where the story leaves the balance sheet and heads for orbit. The terrestrial clusters are not the finish line — they are supposed to fund the move upward.

Responding on Friday to charts drawn from a SpaceX model by Sequoia partner Shaun Maguire, Musk wrote that ground-based compute will be trivial against what the company deploys in space. He went further, projecting that orbital compute stood up each month will likely pass the terrestrial total around late 2028 and then pull far ahead in 2029.

Maguire’s model shows how brutal that ramp looks. Capex climbs from $88 billion this year to $290 billion in 2027 and $369 billion in 2028, with SpaceX burning through $77.8 billion of cash this year and $122.6 billion the next before swinging to $114.9 billion cash positive in 2028. Musk’s reply to that projection was short: “I think we can do better.”

Starmind and the Starship Bottleneck

The space side of the plan runs through Starmind. SpaceX intends to field as many as one million AI satellites, flying in clusters of roughly ten and stitched together with about 10 terabits of bandwidth. Volume production of thousands of units could begin as early as late 2027 at the company’s Gigasat factory in Bastrop, Texas, per SpaceX’s space safety page. Musk has said each Starmind craft will generate more than 250 kW of solar power.

None of it flies without launch cadence. Maguire’s model assumes 250 Starship flights in 2028 — a steep climb from two this year and 38 planned for next year.

The nearer test is financial, not orbital. SpaceX has guided to $33 billion to $37 billion in first-quarter 2027 revenue, more than double the $14.58 billion booked in Q3, and Musk reshared that guidance on Sunday morning.

For readers tracking this from the EV and energy side, the signal is hard to miss: Musk is positioning AI compute — first on the ground, then above the atmosphere — as the core growth engine around SpaceX, with all the capital intensity that implies.