SpaceX is partnering with NVIDIA to launch AI data centers into orbit, accelerating a timeline that could put American-built space compute online by late next year — while Washington regulators still can't figure out how to govern what's already on the ground.

The Starmind project, detailed in a recent SpaceX FCC filing, aims to deploy up to a million satellites in Low Earth Orbit, each packed with roughly 72 NVIDIA Vera Rubin GPUs and Vera CPUs. SpaceX stated on X that the AI1 satellites will use NVIDIA's Vera Rubin NVL72 rackscale system — what NVIDIA calls its Space-1 module, delivering 25 times the AI compute of the older H100 GPU. Musk called it the "best architecture" available on the company's Q2 earnings call, saying the NVL72 VR computer design is "a much better design than, say, having a standard rack-style design."

Each satellite is expected to crank out 175kW of computing power on average, beaming results back to Earth via high-speed laser links through the Starlink network. By flying a sun-synchronous orbit — facing the sun 98 percent of the time — the satellites' massive solar arrays can generate 210kW of power, sidestepping the land, cooling, and grid constraints choking terrestrial data centers. Heat gets dumped through 1,700-square-foot liquid radiators releasing it "freely into the vacuum of space."

Engadget framed the story around skepticism — technical hurdles, cost constraints, scaling issues — and called Musk's timeline "optimistic, even for the notoriously optimistic Musk." International Business Times, by contrast, led with SpaceX's $100 billion Louisiana launch facility announcement and buried the accelerated Starmind timeline several paragraphs deep. IBTimes reported that Musk now says the first AI1 satellite will launch in Q4 of next year, with "significant scale in 2028" — a year ahead of the previous 2028 target.

The Louisiana Starbase facility, set to begin construction next year with a first Starship launch projected for 2029, will support "thousands of Starship flights a year" and create more than 3,000 new jobs, according to SpaceX. That is real capital and real employment — not a subsidy, not a grant, not a green-energy boondoggle.

Meanwhile, the power problem that Starmind aims to solve in orbit is also being tackled on the ground. Applied Atomics, co-founded by former SpaceX engineer Ben Kellie, is reviving the dormant mPower small modular reactor project outside Lynchburg, Virginia — a facility that already absorbed $400 million in investment before it was abandoned in 2017. Kellie told The Washington Post the existing infrastructure gives them "a leg up" in an industry where new reactor tech takes years and billions. The AI boom's electricity demands have made nuclear competitive again, without a federal mandate in sight.

Eventually, Starmind is expected to shift to Terafab, a $119 billion joint chip manufacturing venture between SpaceX and Tesla — further vertical integration, further independence from supply chains controlled by foreign fabs.

The question isn't whether SpaceX can hit its timeline. It's whether the regulatory state will get out of the way long enough to let them try.