Brian Wang SpaceX was going to launch about 5,000 to 8,000 Version 3 communications satellites in 2027 and then another 20,000 Version 3 communication satellites in 2028. This was the old plan until it was confirmed that AI data centers in space would work and would be easier to build. They are raising $75 billion or may $85 billion from the IPO and they will speed up launch plans and satellite build. They will sacrifice and slightly slowdown deployment of V3 communication (high speed internet) and direct to cellphone communications. — nextbigfuture (@nextbigfuture) June 11, 2026 Adding 5,000 V3 satellites is like adding 60,000 V2 mini satellites. This enables them to add six times as many customers. They can already go to 30-50 million customers from the 12 million they have today. The FCC permitted an increase in power flux density which as increase in transmission power. This lets more customers get better service. They also bought more spectrum from Echostar which can be used for direct to cellphone and is critical for direct to cellphone. The Echostar midband spectrum also improves high speed internet service and capacity. SpaceX can get by adding 2000 V3 satellites for high speed internet and 2000 satellites for direct to cellphone. By pushing hard with the $75 billion they can add more launch towers and launch sites and upgrade launch facilities to get two to three times the Starship launches in 2027 and in 2028. They can get to $30 billion per year of Starlink high speed even without adding new satellites and a moderate increase in V3 satellites ensures they can get to 200 million customer capacity globally for $90-120 billion per year in high speed internet and $40-100 billion in direct to cellphone annual revenue. Screenshot 1.5 Gigawatt of AI data center revenue and data center revenue on Earth with Rubin chips can be twice as much as the Colussus 2 revenue potential using B200. The Rubin chips have ten times the AI inference capacity. This should rent for twice as much. 550K B300 chips using 1.3 Gigawatts of energy and at google rental rates of $11 billion per year for 110k B200 chips is $55 billion per year for all of colossus 2. It would be $110 billion per year for a Rubin chip version of Colossus 2 or about $75 billion per year per gigawatt. IF SpaceX could sell its own AI via XAI Grok Code and Cursor, then they could fully use and monetize 10X the inference for $440+ billion per year for each gigawatt of ground AI data center or Space data center. ONCE SPACEX IS PUTTING 4-10+ GW PER YEAR OF AI DATA CENTER ON THE GROUND OR IN SPACE IT IS GAME OVER – $400 Billion to Trillions per Year 130 V3 Starship launches is enough for launches for 1 gigawatt of Version 1 AI satellite and that would be $75 billion per year or more in revenue. It would be $400 billion per year if they can get Cursor and Grok Code competitive or leading. This is about 2 to 3 times the incremental communication revenue even when they are renting out the AI to others. It is over 10 times the revenue if they get their own coding and AI agent systems competitive. The first dozen or so launches of communication add more critical incremental capability. But it also takes time to make the dishes and sell them. Customers for communications can only be added so quickly. 1 GW PER YEAR WAS ALREADY 20% OF ALL AI DATA CENTER IN 2025 and 2026 Screenshot