Starlink and launch are real, profitable franchises. The price pays for an AI compute business that has to keep renting at today’s rates. I own that bet as a conviction call, and I show the numbers that argue against it.
Data through Q2 2026. Q3 results are due Oct 20, 2026, and about 1.3B shares unlock on Oct 22; this report will be refreshed after Q3. The live price updates itself; the financials do not.
Quarterly revenue, year-over-year growth. The business
This is a conviction call, and it is labelled as one. The Bullish rating rests on a conviction case worth $205 a share. That case needs the bull operating path (13 GW of AI compute by 2030, about $340 billion of revenue that year) and a 9% discount rate with a 20x exit multiple. My base case is worth $37 a share, well below today’s $143.48, and the probability-weighted value of my four scenarios is $54, so my disciplined rule reads Bearish. For a reader, that means higher risk and a lopsided outcome: if the beliefs in the conviction case hold, the stock can compound at roughly 17% a year; if they do not, most of today’s price is not supported by cash flows I can model.
Thirty-eight analysts tracked by S&P Global average a $223 target (median $225, range $118 to $450); most of the initiating banks were IPO underwriters. Morningstar, which did not underwrite, puts fair value at $62. My model sits near Morningstar for three reasons.
Renting GPUs earns high prices while chips are scarce. Prices per watt fall as newer chips arrive, and the hardware has to be replaced every five to six years. My base case lets revenue per watt-year fall from $34 in 2027 to $18 by 2035 and charges full replacement capex.
SpaceX’s adjusted EBITDA adds back $2.7 billion of stock compensation over the last four quarters. I deduct it. I also count AI depreciation as a real cost, because it is the cash needed to keep the fleet the same size.
Peers in AI infrastructure (Nvidia, Oracle, CoreWeave) carry two-year betas near 2; satellite and platform peers near 1.2. Weighted by where SpaceX’s value sits, that gives a beta of 1.6 and a 13.2% cost of equity, well above the 9 to 10% that a $300 target typically uses.
SpaceX reports three segments. Space sells Falcon 9, Falcon Heavy and Dragon missions to commercial customers, NASA and the Space Force, and develops Starship. Connectivity is Starlink: broadband from more than 10,200 satellites to 12.0 million subscribers in 167 markets, plus enterprise, aviation, maritime and government service (including the secure Starshield network) and Starlink Mobile, which connects ordinary phones through about 30 carriers. AI came with the February 2026 merger with xAI, which already owned X: it sells compute from the Colossus data centers to other AI labs, the Grok models and subscriptions, and X advertising. SpaceX bought the AI coding company Cursor for about $60 billion in stock in August.
Reported results, USD billions (subscribers in millions, ARPU in dollars a month, compute in gigawatts)
| Line item | FY2023 | FY2024 | FY2025 | Q2 2025 | Q1 2026 | Q2 2026 |
|---|---|---|---|---|---|---|
| Space revenue | 3.56 | 3.80 | 4.09 | 0.75 | 0.62 | 0.96 |
| Connectivity revenue | 3.87 | 7.60 | 11.39 | 2.59 | 3.26 | 4.29 |
| AI revenue | 2.96 | 2.62 | 3.20 | 0.74 | 0.82 | 2.56 |
| Total revenue | 10.39 | 14.02 | 18.67 | 4.07 | 4.69 | 7.81 |
| Adjusted EBITDA | 3.82 | 5.35 | 6.58 | 1.21 | 1.13 | 3.54 |
| Income (loss) from operations | (3.51) | 0.47 | (2.59) | (0.97) | (1.94) | (0.14) |
| Net income (loss) | (4.63) | 0.79 | (4.94) | (1.01) | (4.28) | (0.54) |
| Capital expenditures | 4.42 | 11.16 | 20.74 | 2.83 | 10.11 | 18.37 |
| Starlink subscribers | 2.3 | 4.4 | 8.9 | 6.0 | 10.3 | 12.0 |
| Starlink ARPU | 99 | 91 | 81 | 85 | 66 | 66 |
| AI compute, nameplate | 0.0 | 0.3 | 0.8 | 0.4 | 1.0 | 1.4 |
Sources: SpaceX final prospectus (424B4, June 12, 2026) for FY2023 to FY2025; Q2 2026 earnings release and 10-Q (August 4, 2026) for the quarters. History is restated to include xAI and X for all periods (common-control mergers). Adjusted EBITDA excludes depreciation, stock compensation, restructuring and impairments. Q1 2026’s net loss includes a $1.5 billion loss on refinancing xAI and X debt.
The bull thesis is vertical integration: SpaceX launches its own satellites at cost, the satellites fund the next build, and the same rockets and factories are meant to carry AI compute into orbit later. Each link is at a different stage of proof.
170 launches in 2025, 2,213 tonnes to orbit, mostly its own satellites. Proven.
About 10,200 satellites; V3 satellites (1 Tbps each, up to 60 per Starship) started deploying in July. Proven, scaling.
Connectivity earned $4.7B of adjusted EBITDA in H1 2026 on $2.7B of capex. Self-funding today.
Pays for steps 1, 2 and part of 41.4 GW on the ground at June 30; orbital compute satellites planned from 2027 to 2028. Unproven economics.
| Flight | Date | Result |
|---|---|---|
| 12 | May 22, 2026 | First V3 vehicle and new pad. Upper stage splashed down on target; booster lost on its landing burn. |
| 13 | Jul 24, 2026 | Deployed 20 production V3 Starlink satellites and relit a Raptor in space; the booster was lost on landing. |
| 14 | Sep 28, 2026 | First orbital flight, about 275 km. Deployed 26 V3 satellites. One of six upper-stage engines shut down early and the flight was shortened. |
| 15 | No earlier than Nov 8, 2026 (provisional) | Possible first attempt to catch the upper stage. Musk put the odds of a first successful catch at “at least 50 or 60 percent.” |
Sources: Q2 2026 earnings release; Space.com and NPR coverage of Flight 14 (Sep 28, 2026); NASASpaceflight launch schedule via Wikipedia’s Starship flight list (provisional dates).
Only about 2.9 billion of 13.6 billion shares trade freely today. The prospectus sets a long release calendar, and two large steps fall in the next two months.
| Earliest date | Shares (up to) | Who |
|---|---|---|
| Aug 6 to Oct 9, 2026 | 2.27B | Released in steps after the first earnings report; already trading |
| Oct 22, 2026 | 1.3B | 28% of the 180-day lockup, two trading days after Q3 results |
| Oct 26, 2026 | 328M | 135th-day step (falls on a Saturday, Oct 24) |
| Dec 8, 2026 | 798M | End of the 180-day lockup |
| Feb to May 2027 | 1.06B | Extended lockup for insiders other than Musk, in four steps |
| Jun 12, 2027 | 6.4B + 352M | All of Musk’s shares become eligible, plus the next insider step |
| After Q2 2027 results | 352M | The last of the extended lockup |
Source: 424B4, “Shares Eligible for Future Sale.” Dates tied to earnings are my mapping from the Oct 20 release date. The conditional early release of 455.8M shares did not trigger, because the stock never closed 30% above the IPO price before the first report. The float of 2,904.3M shown by Robinhood on Oct 16 matches the prospectus schedule exactly.
Anthropic pays $1.25 billion a month for about 300 megawatts, roughly $50 per watt-year. If SpaceX can keep renting at anything close to that as it scales to 10 gigawatts, the business is worth far more than my base case. If rental prices follow older chip generations down, it is worth less. This one variable moves the value more than everything else together.
Subscribers grow about 1.7 to 2 million a quarter. ARPU fell 22% in a year as cheaper plans and new countries came in, and management expects it to keep falling. Enterprise and government revenue grew 108%, and management expects it to match or exceed consumer revenue.
A reusable Starship cuts the cost of launching V3 Starlink satellites and, later, orbital compute. Management says flights could reach one a day about a year from now; the program has not yet caught or reflown an upper stage. Today an AI watt costs about $35 to $45 of capex on the ground.
Capex of about $65 billion in 2026 against $9.9 billion of operating cash flow over the last four reported quarters means the cash pile shrinks fast. A reported $40 billion debt raise would double debt; equity issuance would dilute. Musk controls the board, and some related-party dealings (Valor leases, Tesla Megapacks) need watching.
The model builds each segment from its drivers: Starlink subscribers times ARPU plus kits, enterprise and government, and Starlink Mobile; Space revenue less Starship R&D; AI compute as average gigawatts times the share rented times revenue per watt-year, plus Grok, X and Cursor. Segment margins follow the company’s adjusted EBITDA definition; stock compensation is then deducted as a cost. AI capex is new capacity plus replacement of a fifth to a sixth of the fleet each year. Net operating losses shelter tax until about 2028.
Base case, USD billions
Scenario stories and every driver are in the Excel model (Scenarios sheet). The bull case is close to Morgan Stanley’s published path (about $319 billion of revenue in 2030); the right tail is close to Goldman Sachs’s reported $474 billion projection for 2030 (New York Times, June 11, 2026).
SpaceX lost money and burned $25 billion of cash in the first half of 2026. The split by segment shows what kind of spending that is.
RevenueCapital spending
My reading: Starlink is investing from strength; AI is a bet on rental prices. Connectivity earned $4.7 billion of adjusted EBITDA in the half and spent $2.7 billion: it funds itself. AI spent $23.6 billion against $3.4 billion of revenue. That spending is backed by $14.1 billion of contracted sales and management claims a payback under one year, but it buys hardware that ages in five to six years under contracts customers can leave on 90 days’ notice. Whether it is investment or burn depends on prices per watt in 2028 to 2030, which no one can observe today.
Source: 10-Q for the quarter ended June 30, 2026 (cash flow statement and segment note) and the Q2 2026 earnings release. Operating cash flow $3.466B; capex AI $23.551B, Connectivity $2.699B, Space $2.226B.
SpaceX has operating losses, heavy capital spending and three businesses at different stages. A multiple of today’s earnings would divide by a negative number, and a sales multiple would treat a 61%-margin subscription business and a capital-hungry compute rental business as if they were the same. A cash-flow model built from each segment’s drivers is the only method that charges the capex when it is spent and asks whether it earns its cost.
Subscribers × ARPU; launches and Starship R&D; gigawatts × share rented × price per watt.
Why: each is a number SpaceX reports every quarter, so the model can be checked.
EBITDA after stock comp, less cash tax, capex by segment and spectrum payments.
Why: the risk is funding, and replacement capex is the hidden cost of AI.
Beta 1.6 from segment peers (the stock has only 88 sessions), cost of equity 13.2%, WACC 13.1%.
Why: 88 sessions give betas from 1.6 to 3.2 depending on the index. Peers are steadier.
13x 2035 EBITDA after stock comp, cross-checked against 3% perpetual growth.
Why: SpaceX still grows in 2035; I publish the much lower cross-check anyway.
Bear 25%, base 45%, bull 22%, right tail 8%.
Why: outcomes range from a Starlink company with stranded GPUs to the management plan. Bull and tail together stay at 30%.
Bullish if the 5-year expected return beats the cost of equity by 2 points a year. The conviction case is shown beside it, not mixed in.
Why: the reader can see where judgment overrides the rule.
Equity bridge: net cash of $67.8 billion is June 30 cash and securities ($100.0B) plus bitcoin ($1.1B) less debt and leases ($39.5B, including $13.3B of related-party Valor equipment leases), plus $6.2B of option exercise proceeds. The 14,500 million diluted shares include options, RSUs, Cursor’s assumed awards, the 261.8 million shares owed to EchoStar at the spectrum closing (target November 2027), and Musk’s 1,302 million unvested restricted shares. I count those restricted shares in the primary value because they are outstanding, they vote, the market cap includes them, and a board Musk controls can change their terms. Excluding them (13,198 million shares) lifts the model value from $54.21 to $59.56 and the conviction case from $205.35 to about $226.
| Method | How I used it | Reason |
|---|---|---|
| Segment DCF, exit multiple, four scenarios | Primary | Ties value to the drivers each quarterly report updates, and charges AI capex when it is spent. |
| DCF with perpetual growth | Cross-check | Gives $7.69 for the base case: replacement capex keeps normalized free cash flow small relative to EBITDA. |
| Reverse DCF | Cross-check | Solves for what $143.48 assumes: AI revenue per watt about 3.3x my base path, or a 48x exit multiple. |
| EV / sales and EV / EBITDA against peers | Context | Established peers trade at 2.8x to 8.3x trailing sales; SpaceX at 82x cannot be reached from peers. |
| Sum of the parts | Not used separately | The segment build already is one, inside a single cash-flow model; separate multiples would double count shared launch and capital. |
| Option value for Mars and orbital compute | Not priced | No cash flows can be estimated. The right-tail scenario carries the orbital-compute upside instead. |
I discount unlevered free cash flow for H2 2026 and 2027 to 2035 at a 13.1% WACC with mid-year discounting, and capitalize 2035 EBITDA after stock compensation at 13x as a value at the end of 2035. The cost of equity is 13.2%: a 5.2% risk-free rate plus a peer-based beta of 1.60 times a 5.0% equity risk premium. The beta blends two-year betas against IVV: AI infrastructure 1.90 (Nvidia 1.90, Oracle 1.72, CoreWeave 2.71 daily), connectivity 1.20 (Iridium 1.05, Alphabet 1.15, Amazon 1.40) and space 2.00 (Rocket Lab, Tesla), weighted 55/40/5 by where the base-case value sits. Debt is only 2% of capital at market values, so WACC is close to the cost of equity. Thirteen times sits between telecom and satellite operators (T-Mobile 7x, Viasat 8.5x, Iridium 12.7x adjusted EBITDA) and AI infrastructure (Oracle 14x, CoreWeave 17x on trailing EBITDA). Even so, 13x implies 10.5% perpetual growth for the base case; a 3% perpetuity gives $7.69.
| Company | EV / sales | EV / EBITDA |
|---|---|---|
| T-Mobile | 2.8x | 7.1x |
| Viasat | 2.8x | 8.5x |
| Iridium (being acquired) | 7.0x | 12.7x |
| Oracle | 6.8x | 14.3x |
| CoreWeave | 8.3x | 16.8x |
| Rocket Lab | 36.3x | n/m |
| SpaceX at $143.48 | 81.8x | 227x |
Trailing four quarters. Sources: SEC 10-Q filings and company releases; prices from Robinhood, Oct 16, 2026. T-Mobile, Viasat and Iridium on company-adjusted EBITDA; Oracle and CoreWeave on operating income plus depreciation. Multiples are context, not inputs.
Holding everything else in my base case fixed, $143.48 requires AI revenue per watt-year about 3.3 times my path: about $111 in 2027 and still $59 in 2035, against roughly $50 on today’s best contract. Alternatively it requires a 48x exit multiple on 2035 EBITDA. Street-style inputs do not close the gap on their own: at a 9% WACC and 20x, my four scenarios weigh to $116. The price needs both a lower discount rate and an operating path near the bull case. That combination is the conviction case.
The conviction case is not a fifth scenario and carries no probability weight. It is my bull operating path run through the same model, with the valuation assumptions a long-term holder of SpaceX would use: a 9% discount rate and 20x 2035 EBITDA. It gives $205.35 a share ($188.77 at 10% and $223.54 at 8%). Held over five years at a 9% hurdle, that is about 17% a year from today’s price, which is why the headline reads Bullish. My disciplined rule, applied to the probability-weighted $54.21 at a 13.2% cost of equity, reads Bearish. Both are shown so you can judge the bet.
| Belief | Conviction case | My base case | Evidence today |
|---|
Values for 2030 from the model (public/spacex-model.mjs, CONVICTION) and the Excel Conviction sheet, which recomputes the case with live formulas. Morgan Stanley’s initiation (Jul 7, 2026, via Yahoo Finance) projects about $319 billion of 2030 revenue and a $300 target.
Each slider applies to all four scenarios. The scenario values, the model value and the rule update instantly; the conviction case updates for the three operating sliders. The published call does not change.
The same three people have run SpaceX for more than a decade. The difference since June is that public shareholders now own a minority of the economics and almost none of the control.
Founder (2002), Chief Executive Officer, Chief Technical Officer and Chairman
What it means: the company’s ambition is his, and so is the key-person risk. Holders get his execution record and no say over capital allocation.
President and Chief Operating Officer since 2008; director
What it means: the operating continuity behind Falcon’s cadence. Her succession is a quiet risk.
Chief Financial Officer since 2011
What it means: the $100 billion run-rate target is the first public promise to check, on the Q4 report.
Nine directors; Class B elects 51%
What it means: experienced, but close to Musk. Botha, added in June as an independent audit-committee member, is the check to watch.
| Holder | Shares | Stake | Note |
|---|---|---|---|
| Elon Musk | 6,418.5M beneficial | ~84% of votes | Includes 1,302M unvested restricted Class B and 350M Class B options (Schedule 13G, Aug 2026). |
| Alphabet (Google) | 551.2M Class A | 7.2% of Class A | A Google executive, Donald Harrison, has been a director since February 2015. |
| Valor Equity Partners (director Antonio Gracias) | 460.6M Class A | 6.5% of Class A at the Aug filing (503.4M) | Distributed 42.8M shares to its investors on Sep 11, 2026; also SpaceX’s related-party lessor of $20.2B of AI hardware. |
| Founders Fund (Peter Thiel) | about 423M Class A | 5.5% of Class A | Schedule 13G, as of Jun 30, 2026. |
| Tesla | 19.0M Class A | under 1% | Also a supplier (Megapacks) and partner in the Terafab chip project. |
Sources: 424B4 (management, beneficial ownership, related-party transactions); Schedule 13G filings by Musk, Alphabet, Gracias and Founders Fund (Aug 11 to 14, 2026); Form 4 filings (Sep 15 and Sep 24, 2026); 10-Q Item 5 (10b5-1 plans). Economic and voting percentages are my calculation on 13,571M shares.
| When | Event | What I am watching |
|---|---|---|
| Oct 20, 2026 | Q3 2026 results, after the close | AI revenue as the Anthropic and Google contracts ramp; Q3 capex; net cash; Starlink adds and ARPU; any update on the $100 billion run-rate target. |
| Oct 22 and 26, 2026 | About 1.6B shares unlock | How the stock absorbs the largest supply step so far. |
| No earlier than Nov 8, 2026 | Starship Flight 15 | First attempt to catch an upper stage; a clean flight supports routine reuse in 2027. |
| Dec 1, 2026 | $1.11B-a-month compute contract starts | First revenue from the fourth large AI customer. |
| Dec 8, 2026 | 180-day lockup ends (about 0.8B shares) | Selling by early investors; registration rights become usable. |
| Late Jan or Feb 2027 | Q4 results and first 10-K | Did the December run-rate reach $100 billion? 2027 capex and funding plan. |
| 2027 | Propellant transfer; Artemis III docking test; Starlink Mobile next-generation satellites; about $40B debt raise (reported) | The milestones the conviction case needs. |
| Jun 12, 2027 | Musk’s lockup ends | Up to 6.4B shares become eligible; any pledging or sale would matter. |
| Nov 30, 2027 (target) | EchoStar spectrum closing | 261.8M new shares and up to $8.5B of cash; Starlink Mobile’s US service plan. |
This report uses data through Q2 2026. Here are my estimates for Q3, so the call can be checked against the result. They are my estimates, not consensus.
The single most important line will be AI infrastructure revenue: it shows whether the rental prices in today’s contracts carry through as capacity grows.
Download the Excel model. It contains live formulas across inputs, four scenario blocks, a Conviction sheet that recomputes the conviction case, valuation, a sensitivity grid, the peer-beta build and reported history. Every hardcoded input carries a source note. It reproduces this page’s numbers exactly.
This is personal research by Phat Ngo, prepared as a work sample. It is not investment advice or a recommendation to buy or sell any security. The headline rating is the author’s conviction call: it rests on the conviction case described above, while the probability-weighted model value is below the current price and the author’s own rule reads Bearish. The author’s portfolio holds SPCX at a 10% target weight, added at the October 16, 2026 close. Forecasts, scenario probabilities and the rating are the author’s judgments and may be wrong. Figures are as reported in public filings and news coverage on the dates cited.