SpaceX (SPCX) — The Quarter That Doubled the Bill
SpaceX's third quarter tripled its compute revenue, turned a $4.4 billion profit and beat the published revenue estimates in every segment. Then the chief financial officer guided the current quarter's capital spending to $35–40 billion, double what he had promised in August. Our fair value moves from $78.51 to $123.95. The base case barely moved; the upside case got much bigger. At $147.95 the stock sits well above the first and below the second.
The company that reported on 20 October is not the company we valued in June. Then, SpaceX was a launch monopoly funding a satellite network and an artificial-intelligence side bet. In the quarter just reported the side bet, renamed Super Intelligence, was the largest business by revenue, the most profitable by adjusted earnings and the destination of 84 cents of every capital dollar. The question for the share price is no longer whether Starlink can pay for Starship. It is whether a compute fleet that earns a dollar back on every dollar inside a year can keep doing so while it grows eightfold, and who lends it the money in the meantime. Everything else in this note — the lock-up, the Tesla committees, the tripwires — is read through that question.
What this note answers
- What did the quarter earn, what did it cost, and what did the beat consist of?
- What is SpaceX worth after it, with every growth stage argued, and why did our number move from $78.51?
- What does the share price require the compute build-out to be, against what management says it will be?
- What arrives next: the lock-up releases of 22 and 26 October, the Tesla committees, the capital the plan needs.
- What would change our view, in observable terms.
Terms used throughout, defined once. Super Intelligence (SI): the segment SpaceX previously reported as AI — the Grok models, the X platform's advertising, Cursor (the coding-assistant company acquired on 14 August) and the compute infrastructure rented to other companies under cloud services agreements (CSAs). Nameplate compute: the electrical capacity of the data centres, in gigawatts (GW); SpaceX prices rentals in dollars per watt per year. Adjusted EBITDA: operating profit before depreciation, share-based compensation and acquisition charges — the company's measure, used here only beside the cash figures. Free cash flow: operating cash flow less capital spending. Lock-up: the contractual restriction on pre-IPO holders selling; its releases are set by the June prospectus. Float: shares free to trade. Growth stages: the explicit forecast is split into periods, each lasting as long as the thing that drives its growth; every length is argued in the stage table. Fair value: the probability-weighted sum of three scenario values at a 10% cost of capital, the same rate as our June and August notes.
The case for the price, in the company's terms
The quarter is the strongest argument yet that the second act is real. Super Intelligence revenue went from $2.56B to $7.56B in one quarter, up 171% excluding Cursor (its infrastructure line +192%), as cloud services agreements signed earlier in the year began paying at full rate. The segment earned $4.95B of adjusted EBITDA on 1.7 GW of nameplate capacity, and the chief financial officer said new deployments are priced "slightly above the high end of the $30 to $50 per watt range" with "a payback of less than a year." Backlog is $63 billion, $42 billion of it to be recognized within twelve months. The chief executive put the end-2027 fleet at 13 GW "highly confident" and the end-2027 annual revenue run-rate "somewhere between $500 billion and a trillion." Capacity is sold before it is built: by his account the company is "hounded relentlessly by every hyperscaler" and could rent its compute "several times over."
Starlink, meanwhile, did what the skeptics said it could not: revenue reached $5.25B, up 69% on the year, with subscribers at 13.6M and the monthly price per consumer rising to $69 from $66 after two years of decline. Enterprise and government revenue grew 91%. Starship reached orbit with 26 operational V3 satellites, each launch carrying roughly ten Falcon 9 launches of bandwidth, and the next flight will attempt to recover both stages. The balance sheet holds $89.9B of cash and securities against $38.8B of debt, the company is rated investment grade, and the first quarter of 2027 was guided to $33–37 billion of revenue — more than the whole of 2025. The proponent's conclusion: a company compounding revenue at this rate, with returns on new capital measured in months, deserves a multiple of the base case, not a discount to it. The four pre-print targets this note cites run from $200 (Mizuho) to $275 (Evercore ISI); TipRanks' 35-analyst average was $217.50, in a $75–$450 range.
Of these claims this note tests the quarter's cash economics, the stated price per watt against what the fleet actually earned, the capital the guided path requires, and the lock-up and merger arithmetic. It cannot test, within a ten-year model, whether compute pricing per watt rises as models improve (management's view) or falls as every hyperscaler adds power (the house's); that is the single input the value turns on, and it is carried as a scenario and a tripwire, not a verdict.
What the quarter earned, and what it cost
Revenue was $14.57B, +161% on the year and 87% on the quarter before. Every segment beat the published revenue estimates — Visible Alpha's compilation had $12.86 billion in total against $14.57B delivered, $6.47 billion for Super Intelligence against $7.56B, $4.92 billion for Starlink against $5.25B. Operating income was $2.79B after −$143M in the second quarter. Net income was $4.4B, or $0.35 a diluted share on 12,475M shares.
Two earnings figures circulated. The company reports no adjusted earnings per share; the $0.45 some services carried is net income with the $1.2B of Cursor retention awards expensed in the quarter added back. The GAAP figure of $0.35 itself rests on $1.61B of net income below the operating line: $873M of interest income on the cash pile less $675M of interest expense, $1,025M of other income (of which $561M came from a related party, a mark on an investment the 10-Q will identify), and a $390M tax benefit. On operating income alone, after a 21% tax, the quarter earned about 18 cents a share. The beat is real; most of it is not operating.
Exhibit 1 — The quarter by segment: revenue and the capital spent to earn it
How to read it. Three groups of bars, one per segment: quarterly revenue in billions of dollars for Q3 2025 (light grey), Q2 2026 (grey) and Q3 2026 (gold), on the left scale. The black tick across each bar is that segment's capital spending in the same quarter, same scale. Compare the gold bar with its tick: inside the bar for Starlink, near its top for Launch, more than twice the bar's height for Super Intelligence. The one thing it shows: the fastest-growing business spends more than twice its revenue on capital each quarter.
The cash statement is where the quarter is best read. Operating cash flow was $9.63B, of which $2.75B was customers paying in advance (deferred revenue rose to $17.5B from $14.3B, now $17.5 billion of the $63 billion backlog). Capital spending was $19.6B, $16.4B of it for compute. Free cash flow was −$9.96B. Net cash fell from $60.6B at the end of June to $51.0B: $89.9B of cash and marketable securities less $38.8B of debt and finance leases, of which $12.9B is still owed to Valor Equity Partners, the firm of director Antonio Gracias, under the equipment leases the company accounts for as borrowings. Share-based compensation ran $2.36B in the quarter including the Cursor awards, 16% of revenue; the house expenses it.
The two businesses the June note worried about both improved. Starlink's price per consumer stopped falling and rose 5% on the quarter on what the CFO called a favourable mix; its gross margin reached 55%, about three points of which was a one-time tariff refund that will not recur; the company said total gross margin in the fourth quarter would look more like the second quarter's. The Space segment, which carries Starship's development cost, earned $357M of adjusted EBITDA after two negative quarters and an operating result of −$7M, against a loss of −$542M in June — twelve customer launches, including the Roman Space Telescope, and the first revenue-bearing Starship flight. Starship research spending rose again, to $1.23B from $1.08B.
The evening session, one data point. In the after-hours session that followed the release and the call, the stock traded at $158.97 at 19:45 ET, +7.4% on the $147.95 close and +28% against our fair value; Tesla traded +1.4% on the committee news. Options had priced a move of about 7–8% either way into the print. The house records the reaction as one observation: the thirty sessions that follow, through two lock-up releases and the 8 December remainder, will say more about where the stock is absorbed than the first evening did, and the tripwires are written to read them. One evening can price a beat; it cannot settle how the bill the call presented will be funded and earned back.
The bill: a quarter of capital spending that doubled on the call
On 4 August the chief financial officer said the next two quarters would be "very similar to the current quarter from a CapEx level perspective," then $18.4B. The third quarter came in at $19.6B, on the guide. On 20 October he guided the fourth quarter to "$35 to $40 billion," because capacity is arriving faster than planned: about 2.2 GW of nameplate compute at the end of October and 2.5–3 GW by year-end, from 1.7 GW at 30 September.
The arithmetic of the plan is the arithmetic of the stock. Turning on one gigawatt of this generation of compute costs roughly $40 billion of chips, power, cooling and buildings — the third and fourth quarters together spend about $50 billion for 1.1–1.6 GW of new nameplate, which is how the house derives the figure. Thirteen gigawatts by the end of 2027, the chief executive's "highly confident" target, therefore implies something like $400 billion of capital spending in 2027, against $89.9B of cash and securities today and a business that generated $9.63B of operating cash in its best quarter. The chief financial officer did not dispute the scale; he described the funding: operating cash flow first, "multiple pools of capital available in the debt market" on the investment-grade rating, and partnerships "with the whole value chain" of vendors and customers. The $40 billion Apollo-led chip financing the Financial Times reported on 6 October, which the house carries as a single-source watch item, was neither confirmed, sized nor denied, and appears nowhere in the release or the 8-K; the 10-Q's debt and subsequent-events notes are the next place it can.
Three relationships bear on how that capital is raised and from whom. The two cloud agreements the company has disclosed — Anthropic's, $1.25 billion a month through May 2029 for about 325,000 GPUs, and Google's, $920 million a month for 110,000 GPUs from October 2026 to June 2029 — together come to $26B a year, about the whole of the third quarter's annualized infrastructure revenue ($25.7B); both can be ended by either party on 90 days' notice once their initial periods pass, which is why the chief executive could say the company rents "at effectively the spot market rate." Alphabet is also a shareholder: 551.2M shares on its August 13G, 4.1% of all shares (7.2% of the Class A, the basis the 13G states). NVIDIA, the sole chip supplier, held 122.8M shares at 30 June. A third of the company's debt is owed to a director's firm. None of it is improper and all of it is disclosed. It means the capital that funds the build-out and the customers that justify it overlap with the shareholder register, and that the revenue book is priced to 2029 rather than locked to it. That is the book the valuation has to rest on.
What SpaceX is worth after this, and why the number moved
Our June and August valuations summed three segments at point values. The quarter makes that method obsolete for the business that now matters most: Super Intelligence is no longer a mark on a private company but a fleet with a stated price per watt, cost per gigawatt, payback and build plan. We therefore value four businesses as staged cash-flow models — Starlink, the Falcon and Dragon launch franchise, Super Intelligence including Cursor, and the Starship programme as a development cost with a scenario payoff — at the same 10% cost of capital as June and August, so the change in value can be decomposed. Every stage's length is set by what drives that stage's growth and stated below so a reader can dispute it. Share-based compensation is expensed; terminal growth is capped at 4%.
How long each business grows faster than normal. Each length is an argument, not a convention; the model workbook's Stages sheet carries the year-by-year paths.
| Business | What drives the faster growth | How long, and why | Then | Growth or path while it lasts (bear / base / bull) |
|---|---|---|---|---|
| Starlink | Constellation capacity and coverage: V3 satellites on Starship (each launch about ten Falcon 9 launches of bandwidth), Starlink Mobile from early 2028 on S-band and the 800 MHz spectrum agreed on 8 October, aviation across roughly 12,000 contracted aircraft | 4 years: the build-out management has dated — weekly Starship flights in the first half of 2027, twice weekly in the second, "a few hundred" V3 satellites by mid-2027, mobile service early 2028, spectrum closing in the first half of 2027. By 2030 the capacity step is in place or it has failed | Five years converging to 3 / 4 / 4% terminal growth; capital intensity falls from 28% of revenue to 18% as the constellation is in place | 22% / 35% / 50% a year from a $21.0B run-rate (69% today, 89% subscriber growth, price per user up 5%) |
| Launch (Falcon, Dragon) | The contracted manifest: 18 new National Security Space Launch missions, three Crew missions through 2028, Starfall and a lunar payload on Starship | 2 years: the visible manifest runs through 2028; beyond it customer demand depends on Starship's price, which is the Starship line's question | Five years converging to 3 / 4 / 4%; SpaceX's ~90% share of mass to orbit cannot rise | 5% / 12% / 25% from $4.2B of trailing revenue at a 50% franchise margin before Starship costs |
| Super Intelligence | Gigawatts turned on times dollars per rented watt: cloud agreements priced through 2029 (Anthropic to May 2029, Google to June) but terminable on 90 days' notice after their initial periods, about 200,000 GB300 chips a month now, Vera Rubin next year, priced "slightly above the high end of the $30 to $50 per watt range" with "a payback of less than a year" | 3 years: the horizon the agreements are priced to (a pricing horizon, not a lock-in), the $42 billion of backlog inside a year, and the five-year revenue life of a GPU vintage. After 2029 the first fleet is being replaced and the house's price per watt has faded toward the cost of capital | Five years in which additions taper to replacement plus growth and the price per watt converges to a utility return (−8% a year in the base case; management argues it rises, the bull's −5% is the closest the house carries) | Gigawatts added 2027–29: 3 / 1.5 / 0 (bear, then run-off) · 3.75 / 4.5 / 4.5 (base: the guided Q4 capex pace, continued) · 7.5 / 9 / 9 (bull: twice the guided pace). Price $42 / $50 / $55 per rented watt; 50 / 55 / 60% of the fleet rented (management: "split it 50/50, more or less … 60/40 one way or the other") |
| Starship | Full and rapid reuse (Flight 15 catch attempt next month), orbital propellant transfer (early 2027), Starbase Louisiana at scale by 2029 | 3 years of development spend (~$8 billion a year, tax-effected), to management's own dates; the bear runs it a fourth year with no payoff | A scenario payoff in 2030: $0 / $150 / $450 billion for the external launch market Starship creates — Starlink's own capacity gain is already in the Starlink line | — |
Why these lengths. Once its driver stops, faster-than-average growth rarely survives long: across US companies from 1951 to 1998, sales growth stayed above the median five years running for 6.3% of firms and ten years running for 0.9% (Chan, Karceski and Lakonishok, 2003). Hence five-year convergence wherever a stage ends. Super Intelligence's three years rest on agreements priced to 2029 — terminable on 90 days' notice, which is why the stage is three years and not their full term — and a chip life that can be looked up; a longer stage is the assumption that most favours the price, and the sensitivities price it. Starlink's four years rest on a capacity step the company has dated and begun.
The scenarios. The base case follows the capital pace the company guided — $35–40 billion a quarter buys about 0.9 GW — into 2027, accelerating modestly to 4.5 GW a year in 2028 and 2029, and reaches 15.5 GW by the end of 2029 earning $50 per rented watt in 2027, fading 8% a year. The bull doubles the pace to 28 GW, prices at $55 fading 5%, rents 60% of the fleet and depreciates chips over six years. The bear funds the chips already committed, then stops: 6 GW at 2029, $42 per watt falling 15% a year, a four-year chip life; when a replacement gigawatt cannot earn its cost back, the fleet runs off. Management's own 2027 pace — 10 GW added, which the house extends at 12 and 12 for 2028 and 2029 because no later figure was given, 37 GW by 2029 — is run on the base case's prices and margins as the proponent's path (the dotted line in Exhibit 4) and is not weighted. The 25 / 50 / 25 weights are the house's standing weights, kept so the number is comparable with June's and August's; the 4% terminal ceiling applies to every business, because none outgrows the economy it sells into forever; the proponent's run keeps both.
| Bear (25%) | Base (50%) | Bull (25%) | |
|---|---|---|---|
| Starlink | $128B | $372B | $870B |
| Launch franchise | $14B | $23B | $39B |
| Super Intelligence incl. Cursor | −$4B | $399B | $3,543B |
| Starship programme (spend less payoff) | −$25B | $91B | $311B |
| Net cash $51.0B less the Q4 stub (−$29.5B) | $21.9B | $21.9B | $21.9B |
| Equity value | $135B | $907B | $4,784B |
| Per share (13.579B shares) | $9.91 | $66.78 | $352.32 |
Weighted 25 / 50 / 25, fair value is $123.95 a share, $1,683B of equity, against $2,009B at the $147.95 close: the stock trades 19% above it. The bear is positive only because the model stops buying chips once a replacement gigawatt cannot earn its cost back over its life; a bear that kept buying into falling prices would destroy more than the equity is worth, and scenario equity is floored at zero because that is what a shareholder actually faces.
Exhibit 2 — Equity value by case, stacked by business, against the market
How to read it. Three horizontal bars, one per scenario with its weight, in billions of dollars on the top scale. Each stacks the four businesses — Starlink (light gold), Launch (light grey), Starship (grey), Super Intelligence (dark gold) — where positive; negative values, net cash and the Q4 stub are netted in the small figure at left. The value a share is printed at each bar's end. The dashed black line is the market value at the close; the solid gold line is the weighted value. Compare each bar's end with the two lines: the base case ends well short of the market, the bull well past it, and the weighted value sits between them. The one thing it shows: how much of the weighted value one scenario contributes.
What moved the number, and what did not. The table below is the attribution the house now attaches to every change in fair value, each row sourced to a filed figure, a stated figure or a house input.
| Factor | Kind | $ a share | Share of move |
|---|---|---|---|
| Share count 13.176B -> 13.579B | balance-sheet | -2.33 | -5% |
| Net cash $60.6B -> $51.0B | balance-sheet | -0.71 | -2% |
| Q4-26: the guided $35-40B capex quarter | this-year | -2.15 | -5% |
| Starlink: run-rate $21.0B, staged (4y build-out at 35% base), capex 28% -> 18% | long-run | +2.04 | +4% |
| Launch franchise + Starship programme replace the $170B Space point value | long-run | -5.67 | -12% |
| Super Intelligence, base case: $330B point value -> fleet model at the guided capex pace | long-run | +2.54 | +6% |
| Super Intelligence, bull case: $620B -> fleet model at twice the guided pace | long-run | +53.82 | +118% |
| Super Intelligence, bear case: $110B -> run-off (the zero floor on scenario equity does not bind) | long-run | -2.11 | -5% |
| Fair value 78.51 → 123.95 | +45.44 | 100% |
Each row is one factor, sourced in the model workbook's Attribution sheet; the rows reconcile to the move. Kinds: balance-sheet (filed), this-year (the guided quarter), long-run (the staged paths).
Exhibit 3 — From $78.51 to today's fair value, one factor at a time
How to read it. A waterfall in dollars a share on the left scale. The first black bar is the August fair value; each following bar adds (gold) or subtracts (grey) one factor, its size printed above; the final gold bar is the new fair value. Left to right: share count, net cash, the guided fourth quarter, Starlink, the launch and Starship lines, then Super Intelligence split into its base, bull and bear cases. Compare bar heights: the one thing it shows is that nearly the whole rise is one bar, the bull case of the compute business.
Read it plainly. The filed changes — the share count and the net cash — took $3.04 off. The guided fourth quarter took $2.15 off. Starlink's new model is worth within a few dollars of August's point value. The launch and Starship lines together are worth $5.67 less than the old Space value. Super Intelligence's base case adds $2.54 a share. And the bull case adds $53.82 — 118% of the entire move. The fair value rose because the upside scenario became much larger, not because the central estimate did: on the capital pace management guided, SpaceX is worth about what we said it was in August. The weighted figure is published because that is the house's method and the bull case is a real possibility with stated inputs; 71% of it is one scenario, and the reader should hold it that way.
What the price requires. Hold every other input and ask what the market is paying for. At $147.95 the weighted value reaches the price when the build-out in every case is 1.36× the house's — a base-case fleet of 20.1 GW by 2029 instead of 15.5 GW — or when the 2027 price per rented watt is about $56 instead of $50 (at $40 the weighted value is $87; at $60, $161). On that extended management path applied to every case the weighted value is $164; on the proponent's inputs throughout — the extended path, bull pricing, bull Starlink, bull Starship — it is $424, with the same caps. The price, in other words, sits between the plan the company has funded and the plan it has described.
Two inputs a reader will want to move. The 10% cost of capital is kept from June and August for comparability and because the compute leg's cash flows are negative for three years in every case; it is an input, not a cap: at 9% the weighted value is $153 and the stock is -3% against it; at 11% it is $103. The payback the CFO cited, under a year, is the rented watt's — $50 a year on $40 of capital, before operating cost. On the house's base inputs a rented watt pays back in about 1.3 years after a 62% margin, and a fleet of which 55% is rented in about 2.3; the gap between the stated figure and the model's is the half of the fleet that earns no rent. The smaller levers: Starlink's stage at three or six years, $117 or $146; half a point on every terminal growth rate, $116 to $133; Starship's payoff at zero or doubled, $115 or $133; the compute stage at two or five years, $121 or $124.
Exhibit 4 — The compute business's cash by year: the race between earnings and capital, with management's path
How to read it. Annual free cash flow of the Super Intelligence business, billions of dollars on the left scale, 2027 to 2036 along the bottom: bear (grey), base (dark gold, heavier) and bull (gold), each labelled with its 2029 fleet, plus a dotted black line for management's stated 2027 pace (+10 GW), extended by the house at +12 and +12, on the base case's prices, margins and chip life. Values are clipped at −200 and +400. The shaded band is the three-year build-out; after it additions taper and the price per watt fades. The hollow circle is each line's first cash-positive year. Compare where each line crosses zero and how deep it runs first. The one thing it shows: in every case, management's included, the fleet spends far more than it earns through 2029; the paths differ in when that reverses, not whether.
On the $100 billion December run-rate. In August management promised a $100 billion annualized revenue run-rate by December, against $58.3B of annualized third-quarter revenue today; on the call it was restated as December's revenue times twelve and described as on track. The first-quarter 2027 revenue range of $33–37 billion annualizes to $132B–$148B. Neither is a model input: the Q1 range is a test of the base case's stage-1 growth, which it clears, and the December figure is scored in February. The model's horizon is years; the next three sessions are about supply.
Sixteen hundred million shares, three sessions
The lock-up ladder in the June prospectus ties its largest release to this report: "on or after the second full trading day on Nasdaq immediately following the public release of our quarterly financial results … for the quarter ended September 30, 2026," up to 28% of the shares under the 180-day lock-up may be sold. The results came out Tuesday after the close; Wednesday is the first full trading day; Thursday 22 October is the second. The house carries the tranche at 1,276.1M — 28% of the pool the prospectus's own 20% and 7% rows imply — against a stated "up to 1.3 billion." The 135th-day 7% tranche, 328.4M, is dated Saturday 24 October and so trades from Monday 26 October. Together: 1,604.5M shares in three sessions, 55% of the 2,904M float that exists today, 20 days of average volume, about $189B of stock at the current price on the first day alone.
Exhibit 5 — Shares free to trade, from today to May 2027
How to read it. Step bars of cumulative shares free to trade, millions on the left scale, from today's float (black) through each release on the prospectus ladder; the two October releases are gold, later ones outlined. The bracket gives the shares added over the three sessions, their share of today's float and the days of average volume they represent; the top-right figure is total shares outstanding. The chief executive's shares, released June 2027, are not drawn. Compare each step with the one before: the one thing it shows is that the 22 October step is the largest before June 2027, and that the float still ends the year at about two-fifths of the company.
The house has a baseline. On 6 August the first tranche, 911.5 million shares into a 646 million float, produced 255.2M shares of volume — 3.4× the 30-day average, 28% of the released shares — and the stock closed +6.1%, then +15.8% the next day; the de-rating had happened the day before, on the first session after the results. The four 7% tranches and the 59 million affiliate release since passed on 73–121 million shares and closes between −4.0% and +1.2%. This release is 1.4 times the August tranche in shares into a float 4.5 times larger, so the relative supply is smaller; it lands two days after a strong print and the day after Tesla reports. The house does not forecast the reaction; it sets the reading rule and scores it next time.
After these two rungs the float is about 33% of shares outstanding. The remainder of the 180-day pool, 797.6M shares, releases on 8 December, taking it to 39%. The extended lock-up — roughly 1.76 billion shares held by early investors and officers other than the chief executive — begins releasing 20% two trading days after the fourth-quarter results, which is why the timing of that report now matters: the second-quarter results came 35 days after quarter-end, the third-quarter results 20. The chief executive's 6.4B shares carry no early-release provision before 12 June 2027; like every locked-up share, they could move earlier only with the written consent of Goldman Sachs on behalf of the underwriters, which the prospectus reserves.
Was the fast report a rush to free shares while the multiple is high? The question earns a paragraph, not a section. For it: the report came 20 days after quarter-end on four days' notice, the company used 391 million new shares as currency for Cursor in August, and a beat landed two days before the largest release on the ladder. Against it: the company has sold no stock since the IPO (the only registration statements since are the August S-8s for employee plans); the two insider transactions on record ran on plans adopted in June — President Gwynne Shotwell exercised and sold 342,170 shares on 22 September at a $153.57 weighted average under a plan dated 23 June, her direct stake unchanged at 2.47 million, and Valor distributed 42.8 million shares in kind to its investors under a plan dated 12 June; and a company that just tripled a segment has an ordinary reason to report quickly. The evidence does not carry the question. Its one testable piece becomes a tripwire: the lag of the fourth-quarter report, which sets the date of the next release.
The merger committees, and the odds
Asked about a combination with Tesla, the chief executive said: "Both SpaceX and Tesla have formed special committees to discuss this idea at the request of shareholders. These special committees consist of independent directors, and because I'm a conflicted party, I can't actually be in the discussions." He called the idea good, said the committees "have not reached agreement on the terms of any transaction," and set out the hurdle: on the Tesla side "we'll need a 50% vote of all shares — not just those who vote." None of this is in the 8-K; it was reported the same evening by Bloomberg, CNBC, CNN and Axios. Tesla's own proxy, filed the day before for a 3 December meeting, carries no merger item; it does disclose that one Tesla director, Ira Ehrenpreis, has sat on SpaceX's board since February.
Exhibit 6 — What Tesla's holders would own in an all-stock combination
How to read it. Two rows of dots on one scale: the share of a combined company Tesla's holders would receive. The upper row is at today's closes — the solid dot a deal at market, the hollow dots at 15% and 30% premiums to Tesla's price, each labelled with the exchange ratio in SpaceX shares per Tesla share. The lower grey dot is the at-market figure from the house's 3 August note. Compare the solid dot with the grey one: the one thing it shows is that SpaceX's rise since August moved about four points of a combined company from Tesla's holders to SpaceX's.
The exchange-ratio arithmetic has moved since August, in SpaceX's favour. At $147.95 and $373.78, an all-stock combination at market would be 2.53 SpaceX shares per Tesla share, 10.0B new shares, +74% to the count, and Tesla's holders would own 42% of the combined company (against 46% in our August arithmetic, when SpaceX traded at $112). A 30% premium takes that to 49%. The combined market value is $3,486B. The vote is the harder part: the chief executive owns 17.7% of Tesla's shares outstanding (the proxy states 19.8% on a denominator that excludes 423.7 million voting-agreement shares); on a test of all shares, where an abstention counts as a no, the rest of the register must deliver 39% of its shares in favour if his shares vote for. Tesla's SolarCity acquisition in 2016 added a second test, a majority of the disinterested shares cast at the meeting, and a committee here could do the same.
In August, before any committee existed, the house put the probability of an announced combination within twelve months at about 15%. It now puts a definitive agreement within twelve months at about 45% and a completed combination by the end of 2027 at about 30%. The prediction markets are higher — Polymarket prices an announcement by 30 June 2027 at 57% and by year-end 2027 at 66%, and Kalshi's definitive-agreement contract for before May 2027 last traded at 66% (61% bid, 66% offered) — and the house discounts them for four reasons: no banker, adviser or term has been disclosed in any filing; the Tesla vote counts all shares; the process is a conflicted-controller transaction of the kind Tesla has litigated for a decade; and SpaceX's committee has to defend giving away 42% of the company at a ratio that improves for Tesla every time SpaceX's shares fall. Tesla and SpaceX stay separate notes; a definitive agreement or Tesla's proxy for a special meeting, not committee news, triggers a merger note.
Starship and the space economy, briefly
Our August note on the space economy set three undemonstrated milestones for Starship's second step: propellant transfer in orbit, reuse of a ship without heavy refurbishment, and a measured payload to a stable orbit. Flight 14 supplied the third — 26 V3 satellites deployed to the operational constellation — and the company has dated the other two: the catch of both stages on Flight 15 next month and a refuelling attempt early in 2027. The Space segment's numbers moved that note's two triggers in opposite directions: the operating result narrowed from −$542M to −$7M — against +$259M a year ago it widened, and no peak in Starship spending was dated — while external launch revenue fell to 12.1% of the company's total from 12.3%, because Super Intelligence grew faster. The launch economics gained a data point; the segment's share lost one.
The receipt: eight tests the house had armed, scored
The house carried eight open triggers on SpaceX into this print — six from the August scorecard note, two from the space-economy note. All eight are graded here.
Q3 printed 20 Oct (announced 16 Oct, 20 days after quarter-end vs 35 for Q2). Revenue $14.57B (+161% y/y), op income $2.79B, GAAP diluted EPS $0.35. The 28% tranche (1,276.1M) is free Thu 22 Oct; the 7% 135th-day tranche trades from Mon 26 Oct. Re-scored in this note.
Lowest close since 5 Aug: $108.27 (5 Aug). 19 Oct close $147.96. The trigger re-arms at the fair value this note sets.
Q3 capex $19.6B (SI $16.4B). On 4 Aug the CFO guided the next two quarters 'very similar' to Q2's $18.4B; on 20 Oct he guided Q4 to $35-40B — roughly double — and the CEO targets 13 GW of compute by end-2027 against 1.7 GW today. No peak was dated; the trajectory steepened.
Neither the release nor the 8-K mentions a financing. On the call the CFO said the company is talking to 'a number of available pools of debt capital' and intends to stay investment grade; no size, lender or instrument. The FT figure stays single-source. The 10-Q's debt and subsequent-events notes are the next read.
Musk: both boards have formed special committees of independent directors 'at the request of shareholders'; no terms agreed; a Tesla-side vote needs 50% of all shares. Committees are not an announced combination (the trigger reads 'announces a merger or combination'). Not in the 8-K; reported by Bloomberg, CNBC, CNN, Axios.
CFO: 'on track', now defined as December revenue x 12. Q3 revenue annualized $58.3B; the Q1-27 revenue range of $33-37B annualizes to $132-148B. Scored on the Q4 release.
Space operating result -$7M vs -$542M in Q2 (narrowed 99%) and +$259M a year ago (a swing to loss). Segment adjusted EBITDA +$357M after two negative quarters. Starship R&D $1,225M, up from $1,076M; no peak dated. Read sequentially the loss narrowed to break-even; read year-over-year it widened. The house scores the trigger on the sequential read and records the other.
Space revenue $1.77B of $14.57B = 12.1% vs the 12.3% Q2 base: down 0.2 points as SI grew faster. First of two consecutive rises not met; window runs to Feb 2028.
What carries the story to the next update
Each tripwire carries its basis, its tolerance and what the model does with the answer, so that the next update applies rules written now rather than judgments formed then.
- The unlock, 22–26 October. Basis: regular-session volume and official closes for the three sessions, brokerage consolidated feed, against the 20 October close. Absorbed: three-session volume of at least 1,604.5M shares (one turn of the released stock) with the 26 October close within 5% of the 20 October close. Not absorbed: the 26 October close more than 10% below it on less than one turn of volume. Either answer leaves fair value unchanged; a not-absorbed read raises the odds the 8 December and January rungs trade as supply events, which the house treats as price-path information.
- The fourth-quarter report lag. Basis: days from 31 December to the release. At 20 days or fewer the extended lock-up's first 351.9 million shares release about 22–25 January; at 35, in the second week of February. Scored on the release date; the house notes a repeat of this quarter's pace as a fact about the company's reporting rhythm, not a motive.
- Fourth-quarter capital spending, and a 2027 number. Basis: the Q4 release's capex line against the $35–40 billion guide, and whether the call gives a 2027 capex figure or a dated peak. Inside the range: the base case's 0.9 GW a quarter holds. Above $40 billion: the base case's 2027 additions move to the pace actually spent — more gigawatts, which raise value only if tripwire 4 shows them earning the stated price, and raise the funding question either way — and the bull's weight does not change for it. A dated peak at any level fires the capex trigger the house has carried since August and moves the base case's compute stage from three years to the dated length.
- What the fleet actually earns. Basis: Super Intelligence infrastructure revenue excluding Cursor, divided by average nameplate for the quarter. Third quarter: about $16.6 per average nameplate watt per year — about $33 per rented watt if half the fleet is internal, against the stated $50-plus — because half the fleet trains Grok and new agreements ramp. The base case needs about $27 by the first quarter of 2027. Two consecutive quarters at $30 or more move the base rented share to 60% and the base price drift to −5%; two at $20 or less move them to 50% and −12%.
- December's $100 billion and the first quarter's $33–37 billion. Basis: the Q4 release (December revenue times twelve, as management defined it) and the Q1-27 release. Each is a test of the stage-1 growth rates, not an input: a Q1 print at or above $33 billion confirms the base; below $30 billion moves the base compute additions to the bear's.
- The merger. Fires only on a definitive agreement announced by either company (an 8-K, item 1.01) or on Tesla filing a proxy for a special meeting. Committee statements, leaks and prediction-market moves do not fire it; dissolution of either committee resets the house probability to 15%.
- The chip financing. Basis: the 10-Q's debt and subsequent-events notes, any 8-K under items 1.01 or 2.03, or a company statement. Confirmed at $30 billion or more: the net-cash line in every case absorbs the interest cost and the related-party share of debt is re-read. Denied or lapsed by the Q4 release: closed.
- The price. Fair value $123.95: a close at or below it resolves the call and the house re-underwrites; a close at or below the base case, $66.78, would be the first time since the IPO that the stock traded at the house's central estimate.
What would change our view
The house would move toward the price's case on any two of: the fleet earning $30 or more per average watt for two consecutive quarters (tripwire 4) while capital spending stays inside its guide; a 2027 capital-spending figure with a dated peak, which converts the build-out from open-ended to finite; Starship recovering both stages and demonstrating propellant transfer by mid-2027, which brings the Starship payoff's base case toward its bull; or Starlink's revenue growth holding above 50% through 2027 with capital intensity below 25% of revenue. Any two of those move the base case toward the bull's inputs, and the fair value with it. The house would move toward the bear on any two of the mirror conditions: realized revenue per watt at or below $20 for two consecutive quarters; a second quarter of capital spending above guide with no 2027 figure or dated peak; a lock-up rung that is not absorbed, followed by a second; or Starlink's growth falling below 30% with capital intensity above 30% of revenue. A completed Tesla combination changes the subject of the note, not the value of either business, and would be written as its own note.
The call
AVOID maintained. Fair value $123.95 a share (bear $9.91 / base $66.78 / bull $352.32 at 25 / 50 / 25), up from $78.51 on 5 August. At the $147.95 close the expected return to fair value is -16%, and the price is 2.2× the base case. The recommendation is unchanged because the gap is still negative and because the part of the fair value that closed it is the part with the least filed support. The gap is narrower than in August: the quarter earned the company a larger upside case, and the house has given it one. What the house will not do is move the central estimate on a capital plan that has been described but not yet funded or spent. The fourth quarter will show the spending; the first quarter of 2027 will show whether the fleet earns what the company says it charges. The Fund holds no position in SPCX or TSLA.
Sources & Method
Company filings. Third-quarter results: Form 8-K and Exhibit 99.1, SEC accession 0001628280-26-066482, furnished 20 October 2026 (period ended 30 September 2026): every revenue, operating-income, adjusted-EBITDA, capex, segment, balance-sheet and nine-month cash-flow figure; the Cursor purchase accounting ($42.5 billion GAAP consideration, $23.4 billion of post-acquisition awards, $608 million of revenue from 14 August); the 800 MHz agreement of 8 October. Third-quarter stand-alone cash flows are the nine-month figures less the six-month figures in the Form 10-Q for the quarter ended 30 June 2026 (accession 0001628280-26-052535, filed 4 August 2026), which also supplies the June balance sheet, the Valor lease disclosure and the June 10b5-1 plans. Cursor closing: Form 8-K, accession 0001628280-26-056945, 14 August 2026 (389,289,254 plus 1,752,426 Class A shares issued; 29.1 million RSUs and 44.4 million options assumed). Lock-up terms: Form 424B4, accession 0001628280-26-042639, 12 June 2026, "Shares Eligible for Future Sale" and the lock-up agreements; the 28% tranche is stated as "up to 1.3 billion" and the house carries 28% of the pool the 20% and 7% ex-affiliate rows imply (911.5 ÷ 0.20 = 4,557.5 million), 1,276.1 million. Insider filings: Form 4, accession 0001628280-26-063378 (Shotwell, 24 September 2026: five sales totalling 342,170 shares at weighted averages of $151.15–$154.72, $153.57 overall, under a plan adopted 23 June 2026; options exercised the same day); Form 144, accession 0001950047-26-009740 (22 September); Forms 4 and 4/A, accessions 0001213900-26-100186 and 0001213900-26-105393 (Gracias/Valor, in-kind distributions of 42,790,223 shares under a plan adopted 12 June 2026). Schedules 13G of 11–14 August 2026: Elon Musk 6,418,547,515 shares; Antonio Gracias 503,414,530; Google LLC/Alphabet 551,189,500; Founders Fund entities 175,152,700. NVIDIA's holding of 122,764,805 shares at 30 June 2026 from its Form 13F-HR, accession 0001045810-26-000065. Tesla: Definitive Proxy Statement, accession 0001104659-26-116700, 19 October 2026 (3,952,730,153 shares outstanding at the record date; Elon Musk 699,580,882 shares, 19.8%; meeting 3 December 2026; Ehrenpreis's SpaceX directorship). No third-quarter Form 10-Q had been filed when this note was prepared. All filings retrieved from SEC EDGAR on 20 October 2026.
Earnings call. Q3-2026 webcast, 20 October 2026, 16:30 ET. The release states that a transcript will be posted at ir.spacex.com; none was available when this note was prepared. Quotations attributed to CEO Elon Musk, President Gwynne Shotwell and CFO Bret Johnsen are taken from Benzinga's automated transcript (posted 17:42 ET) and used only where a second independent transcription or contemporaneous report carries the same wording (teslanorth.com, shacknews.com, TechCrunch, Fortune, Yahoo Finance, Newsquawk); single-source wording is paraphrased. The $500 billion–$1 trillion 2027 run-rate remark is quoted from the three outlets that reported it; the automated transcript renders the lower bound as "$100 billion," which the house treats as a mishearing and records. The provenance table is archived with the note's data. Call-only figures — the Q4 capex range, the Q1-27 revenue range, the December run-rate definition, backlog, nameplate path, pricing per watt, payback, the Tesla committees — appear in no filing and are identified as spoken in the text.
Market data. SPCX $147.95 and TSLA $373.78: the 20 October 2026 closes, taken as the 16:00 ET regular-session prints from the brokerage consolidated feed (the exchange's settled close had not posted when the note was built; the 19 October settled closes were $147.96 and $368.06); the after-hours prints of $158.97 (SPCX, 19:45 ET) and $379.00 (TSLA) are cited as the evening session only. Daily regular-session bars and volumes from the same feed for 20 July–20 October 2026 (the 6 August session: 255.2 million shares, close $114.92 against $108.27; 7 August close $133.11); 30-day average volume 81.7 million and float 2,904.3 million as of 20 October; the four 7%-tranche sessions and the affiliate-release session as listed. Consensus estimates: Visible Alpha compilation as reported by Fortune (revenue $12.86 billion, SI $6.47 billion, Connectivity $4.92 billion) and Bloomberg as reported by Yahoo Finance (revenue $12.7 billion, EPS $0.15, adjusted EBITDA $6.59 billion, capex $19.6 billion); sell-side targets before the print as reported by TipRanks (Evercore ISI $275, Pivotal $230, RBC $225; 35-analyst average $217.50, range $75–$450) and Stocktwits (Mizuho $200). Prediction markets: Polymarket "Tesla and SpaceX merger officially announced by …" contracts, gamma API, pulled 20 October 2026 at 22:30 UTC (31 December 2026: 23.5%; 30 June 2027: 57%; 31 December 2027: 66%; the contracts move intraday); Kalshi KXCOMPANYACTIONMERGER-27 ("definitive, binding agreement"), contract data for the before-1-May-2027 rung, 20 October 2026 (last 66¢, 61 bid / 66 offered). Cloud services agreements: Anthropic ($1.25 billion a month through May 2029, ~325,000 GPUs, terminable by either party on 90 days' notice after the first three months) from the 424B4; Google ($920 million a month, ~110,000 GPUs, October 2026 to June 2029, terminable on 90 days' notice after 31 December 2026) from the free-writing prospectus, accession 0001628280-26-041150, 5 June 2026. The 800 MHz price of about $8 billion is the Wall Street Journal's reporting, not a company figure. The Apollo-led $40 billion financing is the Financial Times' report of 6 October 2026, single-source.
Model. Four staged free-cash-flow models at a 10% cost of capital (unchanged from the 12 June and 5 August notes), mid-year discounting, terminal value from the end of 2036 at 3% (bear) or 4% (base, bull), share-based compensation expensed, a 21% tax on operating profit. Starlink: run-rate revenue of Q3 × 4 ($21.0B); stage-1 growth 22 / 35 / 50% for four years, five-year convergence; EBITDA margin 58→48 / 62→52 / 66→56%; capex 32→22 / 28→18 / 25→15% of revenue. Launch: trailing-twelve-month Space revenue ($4.2B); growth 5 / 12 / 25% for two years; franchise EBITDA margin 45→38 / 50→42 / 55→45% before Starship costs; capex 12→10%. Super Intelligence: a fleet model — nameplate 2.5 / 2.75 / 3.0 GW at end-2026, additions as in the stage table, $42 / $50 / $55 per rented watt in 2027 drifting −15 / −8 / −5% a year through stage 2, rented share 50 / 55 / 60%, capex $42 / $40 / $38 billion per GW bought in 2027 falling 5% a year, chip life 4 / 5 / 6 years with straight-line depreciation by vintage and replacement only where a replacement gigawatt earns its cost over its life, EBITDA margin on infrastructure revenue 55→45 / 62→52 / 68→58%, software and advertising run-rate $7.0B growing 20 / 35 / 55% for three years, SBC 12→5 / 10→5 / 9→4% of revenue. Starship: ~$2.1 billion a quarter of development spend (Space R&D less an estimated $250 million of franchise R&D, plus ~$1.1 billion of the segment's capex), tax-effected, for three years (four in the bear); payoff $0 / $150 / $450 billion in 2030. Q4-2026 stub: revenue $18.5 billion, EBITDA $11.0 billion, SBC $2.4 billion, tax $0.6 billion, capex $37.5 billion (the guide's midpoint) — a house estimate. Scenario equity is floored at zero. The reverse solves scale every case's 2027–29 additions, or every case's 2027 price per watt, until the weighted value equals the price. Every input, path, sensitivity and the attribution are in the downloadable workbook; every release figure the model consumes is asserted against the archived release text before the model runs.
Prior TON618 research relied upon. The First Print, Scored — SpaceX (SPCX) (5 August 2026), which set the $78.51 fair value, the segment values decomposed here, the August unlock baseline and the triggers scored in §7; The First Print: What $111 Still Has to Believe (3 August 2026), which set the merger arithmetic and the 15% probability re-set here; SpaceX (SPCX) Tech-Equity Valuation — Initiating Coverage (12 June 2026); and Twenty Times Cheaper (22 August 2026), which set the Starship milestones and the two Space-segment triggers.
Limitations. The capex-per-gigawatt figure is derived from two quarters of spending against stated nameplate and is the model's most consequential house estimate after the price per watt; both are shown on grids. The rented share of the fleet is the chief executive's description, not a disclosed metric. The Starship payoffs are house scenario values with no filed source and are sensitized to zero. The model and the merger arithmetic are run on the 20 October close, the last regular-session print before the call; the after-hours reaction and the first day's trading are inputs to the tripwires, not to the value — a day's trading says little next to the thirty that follow. Call-sourced figures carry the provenance caveats above.
Disclosures
Information only. TON618 Capital. This report is for information purposes only. Nothing here is an offer to sell or a solicitation of an offer to buy any security, fund interest, or digital asset, and nothing here is personalized investment advice or a recommendation regarding any instrument.
Publisher's exclusion. All research is published solely as general, impersonal information of regular circulation. It is not tailored to the objectives or circumstances of any individual and is not issued in connection with compensation from any client. The Fund has no clients and distributes all research free of charge. On that basis it publishes in reliance on the publisher's exclusion from the definition of "investment adviser" under the Investment Advisers Act of 1940 (§202(a)(11)(D); cf. Lowe v. SEC, 472 U.S. 181 (1985)).
Registration & conflicts. TON618 Capital is not registered as an investment adviser or broker-dealer in any capacity. The Fund is a Bitcoin fund and may hold or transact in the securities or digital assets it discusses; material conflicts are disclosed where they exist. Ownership: the Fund holds no position in SPCX or TSLA as of the report date, and neither is a Fund holding or a BTC-correlated instrument; the note also names Alphabet (GOOGL) and NVIDIA (NVDA) as SpaceX shareholders and counterparties and makes no recommendation on any of them. Compensation: the Fund received no compensation from any party in connection with this report and charges nothing for it.
Use of AI. Artificial intelligence is used in the creation of this research. All methodology and data integrity are reviewed and approved before publication by TON618 Capital's Chief Investment Officer, Keyth Beck; errors may nonetheless occur, and readers should verify independently.
CFA. This report was prepared to align with CFA Institute analytical standards (methodology only). CFA® and Chartered Financial Analyst® are registered trademarks owned by CFA Institute. That reference describes the analytical framework applied; it does not imply the report was prepared, reviewed, or authored by a CFA charterholder, and the report is not issued, reviewed, endorsed, certified, or approved by — nor affiliated with — CFA Institute.
Risk & feedback. Past performance is not indicative of future results. Digital assets and equities are volatile and may result in total loss of capital. Corrections and feedback are welcome — please direct them to CIO Keyth Beck at keyth@ton618capital.com. Version 1.0.