Space Exploration Technologies directed $15.8 billion of its $18.4 billion second-quarter capital expenditure toward artificial intelligence, a allocation that makes the rocket company’s AI spend larger than the entire annual budget of several national space programs. Starlink received $1.4 billion and the launch business $1.2 billion. The figures, disclosed in the company’s quarterly report, show where management believes the next growth lever sits, even if the most exotic part of that bet remains on the drawing board.
The orbital pitch
Orbital data centers are framed as a solution to the two constraints choking terrestrial AI build-out: power and cooling. In orbit, solar energy is uninterrupted and heat can be radiated directly to space, removing the need for water-intensive cooling plants. The concept also aligns with a future space economy that would need compute where the assets already fly. Whether the physics advantage translates into economic return is unproven, and the filing offers no breakdown of how much of the $15.8 billion is earmarked for the orbital variant versus ground infrastructure.
The terrestrial pivot
On the Q2 conference call, Elon Musk spent the bulk of his AI remarks on terrestrial data centers, arguing that SpaceX’s rocketry and thermal-management expertise gives it an engineering edge on Earth. Space-based centers received only passing mention in the prepared materials and the call transcript. That emphasis suggests the company is securing its AI footing on the ground before committing to the orbital leap, and it raises the possibility that the $15.8 billion is largely a conventional infrastructure play dressed in a more futuristic narrative.
The downside cushion
Even a total write-off of the orbital effort would not cripple the balance sheet. Starlink continues to generate recurring revenue from a subscriber base that now exceeds two million, and the launch manifest remains the industry’s most reliable. The AI capex, while outsized, is funded from operating cash flow rather than fresh debt. Investors are effectively financing a call option on orbital compute with the profits from a business that already works.
What to watch next
The next quarterly filing will reveal whether the AI capex run-rate holds or whether the orbital line item begins to appear as a distinct category. A shift in Musk’s public commentary, from terrestrial engineering advantages to orbital deployment timelines, would signal that the experimental phase is ending. Until then, the $15.8 billion is a very large bet on a very long horizon, paid for by a rocket company that has made a habit of turning long horizons into revenue.