A secondary summary of a Deutsche Bank note has SpaceX exiting 2026 at approximately $98 billion of annualised revenue, rounded to $100 billion, against a June-quarter run rate of $31.3 billion. Its breakdown is neocloud $48.0B, Starlink $13.2B, Cursor $12.0B, Starshield and enterprise $11.1B, Space $7.2B, advertising $3.5B and Grok/X/other $3.2B.
SpaceX management separately guided to at least $100 billion on the first post-IPO earnings call. CFO Bret Johnsen defined it as an annualised run rate based on expected revenue in December 2026, and Musk said it was what the company would reach "if we basically did nothing."
The management target is a forward-looking statement, not booked revenue. The Deutsche Bank note is not public, so its line items in this piece are attributed estimates taken from a secondary summary rather than independently verified contents of the note.
The interesting part is not whether $100 billion is big. It is what definition of "run rate" gets you there, because the definition is doing more work than any of the seven line items.
The $100 billion is a December, not a quarter
There are two annualisation methods in play and they are not the same arithmetic.
The $31 billion is the June quarter times four. SpaceX booked $7,814M of revenue in the quarter ended June 30, as reported, and $7.814B × 4 = $31.3B. Straightforward.
The $100 billion is not the December quarter times four. Johnsen said it was based on expected revenue in the month of December: a single month, annualised. That is why the target can coexist with full-year 2026 revenue far below $100 billion: the method captures the peak month at the end of the year rather than the year's average.
Which produces the number this whole piece turns on. To reach $100B ARR on a December × 12 basis, SpaceX has to bill:
$100B ÷ 12 = $8.33 billion in the month of December.
SpaceX's entire June quarter — three months, every segment, launch and Starlink and AI together — was $7.81 billion.
So the target requires one December that is 1.07× the whole of the June quarter. That is the claim in the units the company reports. The company said on the call that existing cloud agreements ramp later in the year and that an additional $6.7 billion of cloud-services revenue over six months begins ramping in October. Cursor is also included in the target. Even with that back-loading, this is a materially larger claim than "triples its run rate" sounds like in isolation.
It is also worth noting what kind of revenue is being called recurring. A launch is not recurring in any sense a software business would recognise, so $7B of "Space ARR" made of Falcon and Starship flights is a category the acronym does not really fit.
Half the target is one line, and it is a monetisation claim
Neocloud at approximately $48B is the largest line in the Deutsche Bank summary and therefore the assumption worth isolating.
SpaceX's Q2 release reported 1.4 GW of nameplate compute at June 30 and $2,194M of AI solutions and infrastructure revenue for the quarter. Annualising that broader revenue line and dividing it by period-end nameplate capacity gives approximately:
$2.194B × 4 ÷ 1,400 MW = $6.27M per nameplate MW-year.
Management said on the call that compute capacity should exceed 2 GW by year end. At exactly 2 GW, Deutsche Bank's $48B neocloud estimate would imply:
$48B ÷ 2,000 MW = $24.0M per nameplate MW-year.
That boundary comparison is approximately 3.8× the Q2 figure. But it is not a like-for-like price comparison, and it should not be read as one. The Q2 numerator covers a full quarter while the 1.4 GW denominator is a period-end figure; nameplate capacity is not the same as average utilised capacity; and SpaceX's AI solutions and infrastructure line may not match Deutsche Bank's definition of neocloud. Because management said capacity will be over 2 GW, the precise year-end figure could also be lower than $24M per MW-year.
The defensible conclusion is narrower: the $48B estimate requires a substantial increase in revenue generated per unit of reported nameplate capacity, much higher capacity than the stated threshold, or both. SpaceX does not disclose enough utilisation and contract-level revenue detail to separate those effects.
SpaceX also disclosed $14.1 billion of contracted sales signed in Q2 and a $47.5 billion backlog. Its release defines contracted sales as total contract value over the non-cancellable period, including revenue already recognised and deferred revenue. Those figures therefore cannot be treated as annual revenue without the contracts' duration and recognition schedules.
Cursor is another forward assumption
Deutsche Bank's summary assigns approximately $12B of the year-end run rate to Cursor. Management confirmed on the earnings call that Cursor is included in its $100B target but did not provide a separate December contribution. The $12B is therefore an analyst estimate, not company guidance or reported SpaceX revenue.
What our model says
Our SPCX model can be put on the same December × 12 basis, and it exits 2026 at $47.2 billion — with the Elon case, our most aggressive published scenario, reaching $53.5 billion. A little under half the target, and a bit over half at the top of our range.
Most of the difference is in compute. Neocloud and advertising account for $39.1 billion of the $52.7 billion gap between Deutsche Bank's approximate breakdown and our base case.
That makes compute monetisation the central disagreement. Our model assumptions produce far less revenue from the capacity expected online by year end. Another quarter of AI revenue and capacity disclosure will narrow that disagreement, though isolating price from utilisation will still require more detail than SpaceX currently reports.
Nothing in the Deutsche Bank summary is a new company disclosure, so we have not changed the model in response to it.
What to watch
- October's cloud-services ramp. SpaceX said the additional $6.7B agreement begins ramping then. The September and December quarters should show whether the required step-up has started.
- AI solutions and infrastructure revenue alongside nameplate compute. Reporting both again will show whether monetisation per unit of capacity is moving toward the level implicit in the target, even if utilisation remains unknown.
- Any utilisation or average-capacity disclosure. Without it, revenue per nameplate MW remains a blended indicator rather than a price.
- Cursor's contribution and whether it is broken out. Management includes Cursor in the target, while the $12B segment estimate comes from Deutsche Bank's model.
- Whether management continues to call the mixed total "ARR." The December-month annualisation is clear, but launches and other transactional revenue do not recur like subscriptions. A segment bridge would make the target easier to interpret.
SpaceX's reported figures are from its Q2 2026 earnings release: revenue $7,814M, AI solutions and infrastructure revenue $2,194M, 1.4 GW of period-end nameplate compute, $14.1B of contracted sales signed and a $47.5B backlog. Management's $100B target, its December-month basis, the additional $6.7B cloud-services agreement over six months, the inclusion of Cursor and the year-end capacity target of more than 2 GW are from the August 4 earnings call. The $31.3B Q2 run rate, $8.33B implied December revenue, $6.27M Q2 revenue per period-end nameplate MW-year and $24.0M at $48B over exactly 2 GW are our arithmetic. The Deutsche Bank breakdown is from a secondary summary of a non-public note and totals $98.1B before rounding: Space $7.2B, Starlink $13.2B, Starshield and enterprise $11.1B, neocloud $48.0B, Cursor $12.0B, advertising $3.5B and Grok/X/other $3.2B. Our SPCX model exits 2026 at $47.2B in the base case and $53.5B in the Elon case when put on the same December-month annualised basis; those are model outputs, not company forecasts.
