# SpaceX and Nvidia Unveil Starmind AI1, an Orbiting Data Center Satellite

> Source: <https://startupfortune.com/spacex-and-nvidia-unveil-starmind-ai1-an-orbiting-data-center-satellite/>
> Published: 2026-08-26 07:39:03+00:00

*SpaceX and Nvidia have specced out the chips, the cooling and a Texas factory for satellites that pack an entire AI server rack into orbit.*

SpaceX and Nvidia are building Starmind AI1, the first in a planned constellation of satellites designed to run AI compute in space instead of on the ground. Each satellite carries roughly 72 Nvidia Vera CPUs and Rubin GPUs, wired together on the same Vera Rubin NVL72 platform that anchors Nvidia's newest terrestrial data centers. One satellite equals one server rack, launched into low Earth orbit. SpaceX has asked the Federal Communications Commission for permission to launch as many as one million of them.

Power is the first problem here. Heat is the second, much harder problem. Each AI1 satellite draws an average of about 160 kilowatts, with peaks near 250 kilowatts pulled from a roughly 210-kilowatt solar array and onboard batteries. Vacuum insulates well. But it's a terrible place to dump waste heat: there's no air around to carry it off. Think of it this way: you can't run a garden hose out to a satellite to cool it down. SpaceX's answer is 110 square meters of deployable liquid radiators. They wrap around redundant pumped cooling loops, doing the job convection does for free on the ground.

Unfolded, it's enormous: 30 meters tall, a 75-meter wingspan. It glides through sun-synchronous orbit, somewhere between 500 and 2,000 kilometers up. It talks to the rest of the constellation, and eventually back to Earth, over the same optical laser links SpaceX already runs between Starlink satellites.

Elon Musk has framed the project as a simplification, not just a moonshot. "We've decided to build exclusively on Nvidia, because we think the Vera Rubin architecture is the best architecture," he said when the two companies announced the partnership in early August, as Interesting Engineering reported. Cutting the cabling and rack complexity that ground-based data centers need actually gets easier in space, he argued. "It will cost less. It'll be more effective."

[Nvidia Discloses $21 Billion SpaceX Stake And $30 Billion In Intel Shares](https://startupfortune.com/nvidia-discloses-21-billion-spacex-stake-and-30-billion-in-intel-shares/)

Nvidia's latest regulatory filing reveals a $21 billion stake in SpaceX built through its xAI investment, plus a $30 billion Intel holding tied to their September 2025 chip partnership. The disclosures show Nvidia increasingly acting as a strategic investor across the AI supply chain, not just a GPU seller. - [Nvidia SpaceX stake disclosure details](https://startupfortune.com/nvidia-discloses-21-billion-spacex-stake-and-30-billion-in-intel-shares/) - [Nvidia Intel shares investment position](https://startupfortune.com/nvidia-discloses-21-billion-spacex-stake-and-30-billion-in-intel-shares/)

SpaceX isn't waiting to find out if the physics works before it builds the supply chain. In June, the company unveiled plans for a Gigasat Factory in Bastrop County, Texas, on more than 1,000 acres it owns or has under contract, with over 11 million square feet of potential building space, according to details first reported by Tom's Hardware. That would make it more than ten times the size of Starfactory, SpaceX's current largest spacecraft plant. Big is the point. The site is meant to be vertically integrated end to end: solar ingots and wafers, solar cells, printed circuit boards, user terminals and the AI1 satellites themselves, all made within the same footprint.

SpaceX is targeting 1 gigawatt of orbital AI compute capacity by the end of 2027, then scaling that figure by roughly an order of magnitude every year after. Prototype AI1 satellites are due for orbital testing in early 2027, with mass production to follow later that year if the schedule holds.

If it holds.

## Nvidia's next addressable market

For Nvidia, Starmind is less a side project than a new customer segment. Jensen Huang has already pointed to the Vera Rubin CPU line as a $200 billion addressable market on its own, according to comments highlighted by The Motley Fool, on top of the roughly $1 trillion in data center infrastructure spending he expects by 2028. Space compute doesn't need to match Starlink's scale to matter to that number. It just needs hyperscalers and governments running short on land, water and power to start treating orbit as the next available rack instead of a stunt.

Skepticism is earned here. Musk has missed self-imposed deadlines on Starship, the Texas Gigafactory and Full Self-Driving before, sometimes by years, and a factory floor the size of a small town is a different problem than an FCC filing. Starlink itself took seven years to grow from a handful of launches into a global constellation, and Starlink satellites don't carry 110 square meters of liquid radiators just to survive their own waste heat. Frankly, the hard part of Starmind was never going to be getting hardware off the ground. SpaceX does that better than almost anyone alive. It's keeping 72 chips running at full tilt, in a vacuum, with no repair crew that can drive out and swap a pump.

For now, Starmind AI1 exists as a spec sheet, a Texas land parcel and a stack of FCC paperwork. The first real test of whether any of it works comes in early 2027, when the prototype satellites are due to reach orbit.

**Also read:** [Amazon Named Its Robot Warehouse Plan After Tetris and Budgeted It Through 2030](https://startupfortune.com/amazon-named-its-robot-warehouse-plan-after-tetris-and-budgeted-it-through-2030/) • [YMTC Overtakes Micron and Kioxia to Become World's Third Largest NAND Maker](https://startupfortune.com/ymtc-overtakes-micron-and-kioxia-to-become-worlds-third-largest-nand-maker/) • [Instinct's AI Assistant Sent an Email Without Asking and Testers Are Furious](https://startupfortune.com/instincts-ai-assistant-sent-an-email-without-asking-and-testers-are-furious/)

[CME launches cash-settled single-stock futures on Nvidia, SpaceX and 50 other US giants](https://startupfortune.com/cme-launches-cash-settled-single-stock-futures-on-nvidia-spacex-and-50-other-us-giants/)

CME Group launches cash-settled single-stock futures on July 27, covering Nvidia, SpaceX, and 50+ major US equities. The contracts offer leveraged directional exposure with 15% margin requirements and no options complexity, reviving a product that failed in the US when OneChicago shut down in 2020. - [single-stock futures CME launch](https://startupfortune.com/cme-launches-cash-settled-single-stock-futures-on-nvidia-spacex-and-50-other-us-giants/) - [how to trade single-stock futures](https://startupfortune.com/cme-launches-cash-settled-single-stock-futures-on-nvidia-spacex-and-50-other-us-giants/)
