It is easy to get excited about SpaceX. Its mission — to make life multiplanetary — sounds like science fiction brought into the present. Reusable rockets, Starlink, Mars ambitions, and the rapid pace of launch innovation make it one of the most fascinating companies in the world.
But from a modeller's perspective, the interesting question is not whether SpaceX is exciting. The more important question is: how does that excitement translate into economic value?
That is where financial modelling becomes useful. A model forces us to move from narrative to numbers. It asks: what are the key business segments, what drives revenue, what margins are possible, how much capital is required, and what valuation could the business justify?
For this exercise, I used AI to help build a first-cut financial model of SpaceX. Compared with my earlier attempts, I wrote a more detailed prompt, and the output from Claude was significantly better. It produced a segmental model with operating drivers, revenue assumptions, cost estimates and EBITDA by business line — all in less than 20 minutes.
That was impressive. But it also raised a more important question:
When AI produces a sophisticated-looking model, how should a modeller judge whether the output actually makes sense?
A model is only as convincing as the assumptions beneath it, so I judge an AI-built one no differently from my own. Four questions do most of the work.
Do the drivers make sense?
Are the growth rates, ARPU and margins defensible against history and against comparable businesses?
Does the model hang together?
Do the three statements tie, and is the capital intensity reflected honestly on the balance sheet?
Does the output survive a reality check?
How do the projected earnings and cash flows compare against companies that already exist at similar scale?
What does the valuation require you to believe?
Working back from the share price, what set of assumptions would have to hold for the price to be justified?
The rest of this piece runs SpaceX through those four checks.
The Three Economic Engines
SpaceX's revenue mix in the model breaks into three distinct segments, each with very different economics, certainty, and contribution to value.
Connectivity 61% of revenue, 109% of EBITDA in 2025
The segment exceeds 100% of group EBITDA only because the loss-making AI segment pulls the group total below Connectivity's own.
The core revenue drivers are Starlink broadband subscribers, consumer ARPU, and enterprise and government revenue. SpaceX had 8.9 million subscribers at the end of 2025, rising to 10.3 million by March 2026 — subscriber numbers growing rapidly (102% in 2025) even as ARPU falls as the company expands into newer, more price-sensitive geographies. Enterprise and Government revenue has grown quickly too — up 63% in 2024 and 51% in 2025.
Space 22% of revenue, 10% of EBITDA in 2025
Revenue here comes from launch services and launch and development. The company expects Space to grow more slowly than the group, because it functions less as a profit centre in its own right and more as a strategic enabler for Connectivity and AI. Over time, launch and development is expected to make up a larger share of Space revenue.
AI Segment 17% of revenue, loss-making at the EBITDA level
This is the most debatable segment. It combines Grok, advertising revenue on X, and AI infrastructure revenue. Through 2025 the bulk came from X advertising, but from 2026 the growth is expected to come from AI Solutions and Infrastructure.
It is best treated as the largest conceptual assumption in the model rather than an established SpaceX business line.
What Claude Produced
Claude produced a detailed operating model with segmental drivers across Connectivity, Space and the AI segment. What made it impressive was that it did not simply forecast revenue at the top line; it built revenue up from operating drivers — launches and revenue per launch, subscribers and ARPU, enterprise and government revenue, compute capacity and AI revenue per gigawatt.
That is the right direction. A good model should not just produce a number; it should explain the business story behind the number.
| Segment | Key Driver | Why It Matters |
|---|---|---|
| Connectivity | Subscribers, ARPU, enterprise & government revenue | Determines whether Starlink becomes the main cash engine. |
| Space | Launches, revenue per launch, launch & development revenue | Shows whether launch scales profitably or mainly enables other businesses. |
| AI Segment | Compute draw, AI revenue per GW, advertising revenue | Creates the biggest upside, but carries the most modelling uncertainty. |
Here are the resulting revenue and EBITDA estimates from Claude:
Where I Changed the Assumptions
After reviewing Claude's assumptions, I made the model more conservative. I reduced revenue expectations across all three segments while leaving most of the cost drivers broadly similar. The biggest revisions were to the pace of launch growth, the ARPU trajectory in Connectivity, and the scale of future AI revenue.
This step matters: the value of AI is that it produces a credible first draft quickly. The value of the modeller is to challenge that draft.
| Metric | My Estimate | Claude's Estimate | Interpretation |
|---|---|---|---|
| Revenue | $55.3bn | $65.0bn | Claude is more optimistic on growth. |
| EBITDA | $33.1bn | $36.7bn | Margins stay high in both cases. |
| Net Income | $10.6bn | $13.7bn | Meaningful, but small against the implied valuation. |
| CFO | $37.2bn | $42.1bn | Strong cash generation, subject to margin and working-capital assumptions. |
| Capex | $33.1bn | $20.9bn | The largest practical difference — I assume far higher investment intensity. |
| Total Assets | $255.5bn | $202.7bn | Higher capex drives a much larger asset base. |
The Balance Sheet Matters
The income statement can make SpaceX look extraordinarily attractive; the balance sheet tells a more demanding story. The largest item is property, plant and equipment — exactly what you would expect for a company building launch capacity, satellites, connectivity infrastructure and, potentially, compute infrastructure.
This is where my model and Claude's diverged most. Given the Q1 2026 run-rate, I assumed capex would be materially higher than Claude had estimated, and I also reflected the 2026 equity raise, which Claude had not. These are not cosmetic changes — they flow through leverage, asset intensity, free cash flow and, ultimately, valuation.
Valuation: Impressive Business, Demanding Price
Even on 2030 projections, SpaceX looks expensive. On my estimates the implied 2030 P/E is 164x; on Claude's more optimistic numbers it is still 127x. Both multiples are anchored to SpaceX's targeted IPO valuation of roughly $1.75 trillion — so even five years out, and even on the more bullish numbers, the company is being asked to grow into a price the market is setting today.
That does not make the price impossible to justify. It means the valuation requires a very large future opportunity set: the market would not merely be paying for 2030 earnings, but for the possibility that SpaceX becomes a platform spanning connectivity, launch, AI infrastructure and, eventually, space-based activity.
| Valuation Metric | My Estimate | Claude's Estimate | Comment |
|---|---|---|---|
| 2030 P/E | 164.4x | 127.3x | Demanding even on forward projections. |
| Price / CFO | 47.1x | 41.6x | Cash flow helps, but the price still assumes long-duration growth. |
Sanity Check: Are the Projections Too Optimistic?
A useful test is to set SpaceX's projected 2030 profits and cash flows against the largest companies in the world today.
| Company | Market Cap ($bn) | Net Income ($bn) | Operating Cash Flow ($bn) |
|---|---|---|---|
| Alphabet (YE Feb 2026) | 4,450 | 132 | 165 |
| NVIDIA (YE Jan 2026) | 4,960 | 73 | 64 |
| Apple (YE Sep 2025) | 4,510 | 112 | 111 |
| Microsoft (YE Jun 2025) | 3,100 | 102 | 136 |
| Saudi Aramco (YE Dec 2025) | 1,730 | 93 | 136 |
| Amazon (YE Dec 2025) | 2,650 | 78 | 140 |
| SpaceX (FY30E) | 1,750 | ~11 | ~37 |
On that basis the projections do not look absurd — several companies already generate far higher net income and operating cash flow than the model gives SpaceX in 2030. The comparison cuts both ways, though. In absolute terms the 2030 earnings estimate is not wildly implausible. But these companies already earn many times what SpaceX is projected to make in 2030 — a reminder of how much growth would still be required, well beyond 2030, to justify the price being paid today.
The Real Swing Factor: AI
The biggest swing factor is the AI segment. In my 2030 model, Connectivity assumes 41 million subscribers paying an average of $37.60 a month, plus $13.4 billion of enterprise and government broadband revenue. It is not hard to argue that AI services could eventually reach a similar scale, if enough users and enterprises pay for productivity tools or AI infrastructure.
That is possible, but it is also the least certain part of the model. Unlike Starlink — where there is a clear chain from satellites to subscribers to ARPU — AI revenue depends on product adoption, competitive positioning, infrastructure economics, pricing power, and the evolving relationship between SpaceX, X and xAI. This is where the model needs the most careful scenario analysis.
The Vision Premium
Then there is Mars. SpaceX is not valued like a conventional transport or telecom company because its ambition is not conventional: to lower the cost of reaching space, build a global connectivity network, and ultimately make space travel routine.
The difficulty for a modeller is that this vision is both real and hard to quantify. If it materialises, today's projections may prove conservative. If it takes longer than hoped, the model may overstate how quickly ambition turns into cash flow.
What This Exercise Taught Me About AI Modelling
The clearest lesson is that AI can create a sophisticated first draft, but it cannot replace judgement. Claude was genuinely useful — it provided structure, segmental logic and a working set of assumptions. But the real modelling work began after the output appeared: deciding whether the segments are correctly defined, whether the drivers are commercially sensible, whether the capex is realistic, and whether the valuation is supported by the economics.
Conclusion
SpaceX is exactly the kind of company that makes modelling both exciting and dangerous. The story is extraordinary, the addressable markets are huge, and the pace of execution has already surprised many. But a powerful story can also make a model too easy to believe.
My revised model is more conservative than Claude's, especially on revenue growth and capex, and even so the business scales meaningfully by 2030. The question was never whether SpaceX can become much larger. It is whether the current price already assumes too much of that future.
Strip out the optionality and the core business — broadband and launch — does not come close to justifying roughly $1.75 trillion on any assumptions. Working back from that price, you have to believe one of two things: that the AI segment becomes a second Connectivity, earning at something like Starlink's scale within a few years, or that the space travel vision is real enough to carry a valuation no current cash flow can support.
Neither is impossible; neither is something a model can prove today. And that is the honest end of the exercise — at this price you are not buying the business in the model, you are buying everything the model cannot yet capture. Whether that is a bet worth making is no longer a modelling question. It is a question of conviction.
That, ultimately, is the value of the modeller's lens: it does not kill the story, it tests how much of the story is already in the price.