Orbital AI arrives in 2028. Not 2030. Not 2035. Not 2038. It needs two things. First, launch below roughly $700 per kilogram. That's the cost point where orbit beats a terrestrial build. Starship gets there, and I don't think it even needs full reusability to do it. Second, hardware built for space. That's what SpaceX and Nvidia are developing now: a space-ready Vera Rubin NVL72 rack, with each satellite carrying about 250 kilowatts. Target completion is Q4 2027. The scale math is simple. 250 kilowatts per satellite, 4,000 satellites, one gigawatt. A few hundred Starship flights. After that, the constraint isn't technology. It's how many Starships SpaceX can build and fly. Elon usually overpromises on timelines. This time the incentive is trillions of dollars, and I think he hits it. $SPCX on

SpaceX + xAI orbital AI supremacy: 100 GW added every year. 1M tons to orbit (Starship V4 @ 200t/flight = 1 launch/hour). Terafab chips + solar arrays, 24/7 power, vacuum cooling, Starlink mesh. Physics, win's.

A lot of experts still think orbital data centers are impossible....or something that’s not going to happen in the near future Elon just confirmed he is highly confident SpaceX will be launching Nvidia VR NLV72 AI computers in space next year SpaceX already designed a space-optimized version of this system with Nvidia. Elon said it’s significantly simpler, lower cost, denser and lighter than a traditional rack Vera Rubin NVL72: • 72 Rubin GPUs • 36 Vera CPUs • 20.7 TB of HBM4 • 3,600 PFLOPS of NVFP4 compute Then there’s Starmind AI1: • 75-meter wingspan....wider than a Boeing 747 • Up to 250 kW peak / 175 kW average compute • Sun-synchronous orbit for continuous solar power • Radiative cooling directly into space • High-bandwidth laser links connecting into Starlink SpaceX is targeting the first Vera Rubin systems in orbit next year....with significant scale already targeted for 2028 Starlink already operates the largest satellite constellation in history, serving internet around the world SpaceX has already solved the major engineering problems needed to make orbital AI compute work....now the challenge left is deploying it at massive scale Yeah....Starlink was also dismissed by so-called experts as something that would never work at this scale Fast forward to today, SpaceX basically owns the infrastructure layer of space....and has the monstrous infrastructure to dominate what comes next This is what Elon does Take something everyone calls impossible, turn it into an engineering problem....solve it and deploy it at massive scale

SpaceX ($SPCX) wants to launch orbital AI data centers as early as next year using Nvidia ($NVDA) chips.But experts say the logistics are brutal. One estimate says a 1M-satellite network could require more than 9 Starship launches per day just to keep chips updated.

BREAKING: SpaceX’s new FCC filing reveals that each Starmind orbital AI data center could weigh up to 4,000 kg, roughly twice the ~2,000 kg figure reported for a Starlink V3 satellite. SpaceX is seeking FCC approval for a constellation of up to 1 million Starmind satellites.

Sam Altman and OpenAI will be fucked thinking orbital data centers aren’t a thing or that it will take a very long time. Just watch all the large AI companies partner with SpaceX for compute in space. • Anthropic: Locked in the entire Colossus 1 supercluster (~300MW+) at roughly $1.25B per month AND actively pursuing multiple gigawatts of orbital AI compute via SpaceX’s satellites • Google: Signed a $920 million-per-month deal for ~110k Nvidia GPUs/compute capacity through 2029 and are in deep talks with SpaceX to launch and test their own orbital data centers. I bet other tech AI companies (Meta, etc.) are also lining up for early access as well… SpaceX’s million-satellite orbital AI constellation, powered by Starship, endless solar, and free vacuum cooling, is going to make this a reality faster than most realize. And anyone who still thinks the best place to put these is on Earth will eventually become compute constrained and left behind. Mark my words.

SpaceX’s Starmind AI satellite plans just got even crazier New FCC details reveal each orbital AI data center can weigh up to 4,000 kg.....roughly twice the reported mass of a Starlink V3 satellite SpaceX is seeking authority for up to 1,000,000 satellites Massive AI compute is moving toward orbit