Morgan Stanley Agrees SpaceX Can Deploy 8 Gigawatts of AI Data Center by 2028
by Brian Wang
Ignore the debate about AI data centers in Space for now. Whatever anyone says no one will believe it until the first thousand AI satellites are in space. The next 3-4 years are all about AI data centers on the ground. Morgan Stanley agrees with me that SpaceX can increase from 2 Gigawatts now to 8 Gigawatts in 2028.
Adam Jonas projects about 8 Gigawatts of SpaceX AI data center in 2028 and 16 Gigawatts by 2030. Very little of that AI data center comes from AI data centers in space.
Morgan Stanley has this in a 142 page report on SpaceX that has several material inconsistencies between its high-level assumptions about orbital compute scale, mass/payload, capex, and monetization and the known technical specifications for Starship launches, satellite power density, and real-world AI compute deals.
I agree that 8 gigawatts of AI data center is very doable for SpaceX in 2028-2030. They already have the natural gas turbines on order from Doosan in South Korea, all of the APR Energy production and 60-70% of the Solaris mobile turbines. Twelve 380 MW doosan turbines are on order and those are 4.5 GW. 1 GW per year from APR Energy and the Caterpillar joint venture is 6.5 GW plus the 2 Gigawatts that are already installed.
Adam of Morgan Stanley has SpaceX revenue of
$45 billion in 2026 (2 gigawatts end of year)
$319 billion in 2030 (16 gigawatts)
$3.3 trillion in 2040 (380 GW).
Adam project 2030 revenue top be $120 billion for the Starlink communication.
This means the projection is that AI data center revenue collapses to less than $12 billion per gigawatt per year.
Half of the 8 Gigawatts available at the end of 2028 would be 4 Gigawatts. Fully rented at current rates this would be $200 billion and there would be another $100 billion or more from communications and launch. There would be some monetization of the other half or SpaceX could just default to renting all 8 gigawatts to get $400 billion and have $100 billion from communication and launch. IF AI rental collapsed to $30 billion per year that would be $240 billion in 2028 for 8 gigawatts for just neocloud rental and $480 billion per year 16 gigawatts in 2030.
It becomes difficult to think of how SpaceX does not end up with far more revenue. Adam collapses the revenue per gigawatt in order to prevent the scenario
What happens to the AI data center economics per gigawatt? If SpaceX can build it ?
They have shown that they can build faster than anyone else and they have the power supply issues solved with suppliers and contracts for natural gas turbines that are already being installed.
