SPCX · Forward model · Space AI · Bear case
What has to happen in Space AI
Model as of
This page changes Space AI inside the complete SPCX model. Every other vertical, corporate programme and valuation assumption stays in the selected portfolio case.
Shares this vertical and portfolio case. Slider and horizon edits stay in your browser.
Space AI
Basis quarter$0
Final quarter$1.74B
Final revenue mix6%
Orbital data centres: compute in vacuum, cooled by radiation and powered by uninterrupted sunlight, launched on Starship at a marginal cost only SpaceX has. Pre-revenue today and the only vertical here that is pure option value - it is worth nothing at all if Starship cadence slips, and it is worth more than the rest of the model combined if it works.
Last four quarters
2025 Q3
$0
Estimated
2025 Q4
$0
Estimated
2026 Q1
$0
Estimated
2026 Q2
$0
Reported
Orbital compute capacity leasingDemonstration and government contracts
Megawatts energised
0 MW
at the basis quarter
Starts at zero. There is no orbital compute in service today - this line begins from nothing.
Flights
12/qtr
ramping toward 91/qtr (1.0 a day)
12 Starship flights in the quarter Space AI opens — roughly one a week, across all missions.
Share of manifest
50%
flights given to this line
Half of STARSHIP flights. At the ceiling that is ~291 MW a quarter — 156 flights to the gigawatt.
Satellites per flight
40
160 kW each, so 6.4 MW a flight
40 compute satellites per Starship load — the V3 bay volume, not the mass limit.
Deployed per quarter
38 MW
at the launch-quarter cadence
160 kW average electrical power per satellite. Thermal rejection, not solar area, is what caps this.
Utilisation
55%
gliding toward 75%
55% at first light. Latency rules out most workloads, so it competes for training and batch only.
Revenue per MW
$2.50M/qtr
drifting −2.0% per quarter
$2.5M per MW a quarter, a premium to the $2.03M ground price for free cooling and constant power.
Space AI
Latest: $1.74B (2031Q2E)
| Period | Value |
|---|---|
| 2025Q1 | $0.00 |
| 2025Q2 | $0.00 |
| 2025Q3 | $0.00 |
| 2025Q4 | $0.00 |
| 2026Q1 | $0.00 |
| 2026Q2 | $0.00 |
| 2026Q3E | $0.00 |
| 2026Q4E | $0.00 |
| 2027Q1E | $0.00 |
| 2027Q2E | $0.00 |
| 2027Q3E | $0.00 |
| 2027Q4E | $0.00 |
| 2028Q1E | $73M |
| 2028Q2E | $171M |
| 2028Q3E | $288M |
| 2028Q4E | $420M |
| 2029Q1E | $560M |
| 2029Q2E | $706M |
| 2029Q3E | $853M |
| 2029Q4E | $998M |
| 2030Q1E | $1.14B |
| 2030Q2E | $1.28B |
| 2030Q3E | $1.40B |
| 2030Q4E | $1.52B |
| 2031Q1E | $1.63B |
| 2031Q2E | $1.74B |
Assumptions & reasoning
- Deployment is derived from launch cadence, not assumed directly: flights per quarter x the share of the manifest given to compute x 40 satellites x 160 kW. At one flight a day with half the manifest, that is about 290 MW a quarter — 156 flights to the gigawatt.
- Cadence ramps toward one flight a day and stops there. One a day with half the manifest is about 291 MW a quarter, so the base case is a claim that orbital compute settles at roughly a gigawatt every three and a half quarters — not that cadence keeps compounding. Faster or higher belongs in the bull and Elon cases.
- This replaced a fixed 3 MW a quarter compounding at 15%, which reached only ~120 MW by 2031 — roughly a tenth of what the flight arithmetic supports. The old number was not a view about launch capability; it was a placeholder that had never been reconciled against one.
- Utilisation stays well below the ground business. Latency rules out inference and anything interactive, so this competes for training and batch work only, and it prices at a premium to ground only while power and permitting are the binding constraint there.
- The swing factor is cadence, and it is unproven. Every megawatt needs a launch slot, so if Starship holds at one flight a week rather than one a day this line stays a rounding error. Move the ceiling slider to see it: the assumption is doing more work here than anywhere else in the model.
- The manifest is not a single pool. This share is of Starship flights, while the Space vertical counts external customer missions across Falcon and Starship both. Do not read the two flight numbers as competing for the same slots — but do treat a high compute share as an implicit claim that Starlink deployment has moved off Starship or become far more efficient per flight.