SPCX · Forward model · Model layer · Bear case
What has to happen in Model layer
Model as of
This page changes Model layer 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.
Model layer
Basis quarter$0
Final quarter$3.53B
Final revenue mix12%
Grok sold as a model rather than as a megawatt: tokens, seats and API calls, competing with Anthropic and OpenAI for a pool that booked a combined $18.3B in the June quarter. Modelled as market x share, because share is the only quantity the argument for this line is ever made in - and the finding is that the pool's growth rate matters more than the share does.
Last four quarters
2025 Q3
$0
Estimated
2025 Q4
$0
Estimated
2026 Q1
$0
Estimated
2026 Q2
$0
Reported
Grok API and token revenueGrok consumer and enterprise subscriptions
Capacity energised
18850 $M of market
at the basis quarter
$18,850M a quarter: Anthropic $11.6B plus OpenAI $6.7B plus Grok $0.55B. Press-reported, none of it audited.
Capacity added
1131 $M of market/qtr
changing +6.0% per quarter
$1,131M added in the first quarter, which is 6% of the pool - the rate the whole line is most sensitive to.
Utilisation
3%
gliding toward 18%
Share of the pool, not utilisation: Grok's $0.55B against $18.85B is 2.9% at the June quarter.
Revenue per $M of market
$1.00M/qtr
drifting 0.0% per quarter
$1M per unit, because the unit IS $1M of quarterly market. This input only rescales the pool.
Model layer
Latest: $3.53B (2031Q2E)
| Period | Value |
|---|---|
| 2025Q1 | $0.00 |
| 2025Q2 | $0.00 |
| 2025Q3 | $0.00 |
| 2025Q4 | $0.00 |
| 2026Q1 | $0.00 |
| 2026Q2 | $0.00 |
| 2026Q3E | $796M |
| 2026Q4E | $1.03B |
| 2027Q1E | $1.25B |
| 2027Q2E | $1.45B |
| 2027Q3E | $1.64B |
| 2027Q4E | $1.82B |
| 2028Q1E | $1.99B |
| 2028Q2E | $2.15B |
| 2028Q3E | $2.30B |
| 2028Q4E | $2.45B |
| 2029Q1E | $2.58B |
| 2029Q2E | $2.71B |
| 2029Q3E | $2.83B |
| 2029Q4E | $2.94B |
| 2030Q1E | $3.05B |
| 2030Q2E | $3.16B |
| 2030Q3E | $3.25B |
| 2030Q4E | $3.35B |
| 2031Q1E | $3.44B |
| 2031Q2E | $3.53B |
Assumptions & reasoning
- SpaceX does not break out Grok. There is no line in the segment disclosure that corresponds to this vertical, so it carries zero across every reported quarter and starts from nothing in the projection, like Space AI and Developer Platform.
- The $0.55B starting run rate is not new revenue. It is the capacity Grok already consumes, moved here out of Terrestrial AI's utilisation - fifteen points of 1,800 MW at $2.03M. Reported June-quarter revenue is unchanged; only its character is.
- Market x share is a landlord's model turned inside out. Terrestrial AI prices the building; this prices the tenant. Both are guesses about the same megawatts, which is why one of them had to give up fifteen points of utilisation.
- The pool growth rate is worth more than the share. Ten percent of a pool compounding at 10% a quarter beats a quarter of one compounding at 5%, and both beat half of a pool that stops - so a reader who wants to argue with this line should move the build rate, not the share.
- Capex at 8% terminal is software economics, and it is the assumption most likely to be wrong: if serving Grok needs megawatts SpaceX would otherwise lease, the cost of this line is Terrestrial AI's revenue rather than its own capex.