SPCX · Forward model · Terrestrial AI · Bear case
What has to happen in Terrestrial AI
Model as of
This page changes Terrestrial 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.
Terrestrial AI
Basis quarter$2.56B
Final quarter$13.41B
Implied CAGR+39%
Final revenue mix44%
Ground-based compute leasing is the explosive line, and it is capacity-gated: revenue tracks energised megawatts times what is contracted times price. xAI/Grok and X advertising ride on top. The highest-variance vertical in the model - the contracts are real, the pricing power past the first tranche is not yet proven.
Last four quarters
2025 Q3
$850M
Estimated
2025 Q4
$870M
Estimated
2026 Q1
$744M
Estimated
2026 Q2
$2.56B
Reported
Compute infrastructure / cloud leasingX advertising and platform
Megawatts energised
1800 MW
at the basis quarter
1,800 MW energised. Revenue is gated by what is switched on, not by demand — contracts land well ahead of power.
Megawatts added
400 MW/qtr
changing +10.0% per quarter
400 MW a quarter, the rate implied by roughly $14B of quarterly capex at current build costs.
Utilisation
55%
gliding toward 67%
55% of energised capacity is leased to third parties. The other 15 points run xAI, now carried in Model layer.
Revenue per MW
$2.03M/qtr
drifting −1.5% per quarter
$2.03M per MW per quarter (~$8.1M a year), reconciling $2.56B of Q2 revenue against 1,800 MW at 70%.
Terrestrial AI
Latest: $13.41B (2031Q2E)
| Period | Value |
|---|---|
| 2025Q1 | $700M |
| 2025Q2 | $780M |
| 2025Q3 | $850M |
| 2025Q4 | $870M |
| 2026Q1 | $744M |
| 2026Q2 | $2.56B |
| 2026Q3E | $2.42B |
| 2026Q4E | $2.85B |
| 2027Q1E | $3.30B |
| 2027Q2E | $3.76B |
| 2027Q3E | $4.23B |
| 2027Q4E | $4.71B |
| 2028Q1E | $5.21B |
| 2028Q2E | $5.73B |
| 2028Q3E | $6.26B |
| 2028Q4E | $6.80B |
| 2029Q1E | $7.36B |
| 2029Q2E | $7.94B |
| 2029Q3E | $8.54B |
| 2029Q4E | $9.16B |
| 2030Q1E | $9.81B |
| 2030Q2E | $10.47B |
| 2030Q3E | $11.17B |
| 2030Q4E | $11.89B |
| 2031Q1E | $12.63B |
| 2031Q2E | $13.41B |
Assumptions & reasoning
- Modelled on energised capacity x utilisation x price because revenue here is supply-gated, not demand-gated. The contracts are signed well ahead of the power being available, so what is switched on sets the ceiling — a demand-side growth rate would be modelling the wrong constraint.
- The build rate is the engine: 400 MW a quarter compounding at 10%. That is derived from roughly $14B of quarterly capex at current build costs, which is why capex intensity starts at an extreme 520% of segment revenue and decays as revenue catches up with the asset base.
- Price is assumed to fall 1.5% a quarter. Every hyperscaler is building at once, so the pricing power visible in the first tranche of contracts is unlikely to survive multi-gigawatt supply. Holding price flat is the single most flattering change available in this model.
- This is the highest-variance segment and the one that decides the answer: it goes from a third of revenue to over half, and its present value is deeply negative inside the horizon because the capex lands years before the cash. If the utilisation target is wrong, the valuation is wrong.
- Orbital compute was previously carried here as unmodelled optionality. It is now its own vertical, so every megawatt in this line is on the ground.
- Fifteen points of utilisation were moved to the Model layer vertical. 1,800 MW x 15% x $2.03M is $548M a quarter - the capacity Grok runs on. Internal inference is not a third-party lease, so billing it here and again as model revenue would count the same megawatt twice.