Dialectic Group / Perpetual Fund / November 2026 / Research Update
SpaceX

Thesis & Valuation
Memorandum

Physical-driver sum-of-the-parts Model v11 Post-Q3 2026 Update

$288.19/shareBase Fair Value (was $352.75 v10 / $285.83 Aug)
$211-$339/shareBear-Bull Range
$3.98 TBase Equity Value
Page 02

Executive Summary

· Valuation date

In August we put fair value at $285.83 a share and said, in effect, that the market had handed the opportunity back to buyers. Then SpaceX reported its second public quarter, and the AI business did something AI businesses rarely do: it made buckets of money. Super Intelligence revenue nearly tripled quarter on quarter to $7.56B at a mind-bending 65.5% adjusted EBITDA margin, total revenue reached $14.575B (+161% year on year) and SpaceX guided Q1 2027 revenue to $33-37B (Q3 8-K; Q3 call). Maybe they should in fact change the ticker to SXSI because this is absolutely the sexiest asset on the planet, or off. At the current valuation against guided YE27 ARR, you owe it to your bloodline to own some SPCX. However, we have tempered short-term forward looking compute predictions and some revenue lines as there is a real risk that the Democratic-led investigations and regulatory holdbacks could cause short-term pain and some delays to planning and execution.

We’ve rebuilt the model from the ground up around physical units: gigawatts of compute energised, rent and build cost per gigawatt, Starship phasing out Falcon 9 flights, Starlink satellites and subscribers, and orbital data centres from 2030. v11 (6 Nov) cuts YE27 terrestrial compute to bear 6 / base 8 / bull 11 GW (was 10/13/16), shifts to ~80% internal compute by 2029 with Cursor revenues inflecting off the old flat revenue upward as Grok Bot is a product that we love and we think the world is about to wake up to. We re-anchor the AI-spend demand-cap share to 8 GW, and reiterate our outlier prediction that cumulative capex stays near ~$1.0T with a $200B cellular/M&A provision inside that envelope. The Street is missing the mark on the planned CAPEX spend, especially when one considers such expenditures as the particle collider for Terafab, terrestrial cell networks, a full-scope gas turbine manufacturing facility and probably something akin to Terafab for solar panel production. We trim 2027–28 Starship flights for the conservative sake of regulatory mud. Guardrails unchanged: WACC 10%, g 4%, 13,811M shares, 65/30/5 blend, Q3 actual margins held flat, no segment EV negative, every Starship flight has real work, debt funding charged interest. Base fair value is $288.19; bear $211.12; bull $338.91 (soft landing zones ~$200/~$280/~$350 — physicals drive).

SPCX closed at $123.62 on 5 November (StockAnalysis; −9.05% on the day, W-074), below the $135 IPO price and roughly 2.3x below our base. The selloff tied to post-midterm AI/data-centre regulatory fear (W-070), a broad Nasdaq AI rout, and the $40B GPU financing overhang (W-073). This market is still offering the equity at a discount to even our bear case.

Our base assumes customers pay $60B per gigawatt-year by the end of 2027, rising to about $74B by 2030 before fading 2% a year from 2032. That is above Goldman's spot rate card ($40-50B) and well above term pricing ($20-30B). If rent stays flat at $60B, fair value falls to $267.75. If the SI boom politely declines to keep booming (which we see absolutely no evidence of whatsoever), see the bear case. Either way, the upside is literally out of this world; the dispersion rewards those willing to cut through the noise and act with conviction.

Page 03

Q3 2026: The Quarter the AI Segment Ate

$14.575BRevenue, +86.5% q/q, +161% y/y; Bloomberg consensus $12.89B
$8.809BAdj. EBITDA, 60.4% margin (45.3% in Q2)
1.7 GWNameplate compute at 30 Sep; "surprised if less than 13 GW" by YE27
$33-37BQ1 2027 revenue guide (CFO, Q3 call)

Table 1. Q3 2026 reported, $B (SEC 8-K EX-99.1)

Segment Q3-26 Q2-26 q/q Adj. EBITDA Margin Capex
Super Intelligence (AI) 7.560 2.561 +195% 4.953 65.5% 16.444
Connectivity (Starlink) 5.248 4.291 +22.3% 3.499 66.7% 1.543
Space (launch) 1.767 0.962 +83.7% 0.357 20.2% 1.606
Total 14.575 7.814 +86.5% 8.809 60.4% 19.593

Source: SpaceX Q3 2026 8-K EX-99.1 (acc. 0001628280-26-066482). Cash and marketable securities $89.86B, debt $38.84B, net cash $51.02B at 30 Sep. Cursor contributed $608M from the 14 Aug closing; advertising $538M. Starlink consumer subscribers 13.6M (consensus 14.29M) at $69 ARPU (consensus $66.74). 35 launches and 476 t to orbit in the quarter.

What we learned this quarter

  • SpaceX is an SI compute business in the short term as this is the most profitable business on the planet. Super Intelligence revenue ($7.56B) exceeded Space plus Connectivity combined ($7.02B). Annualised, SI earned about $19.5B per average gigawatt (Dialectic arithmetic), comfortably above the $9-14B we used in August. Revenue per GW is the single number that moved this model most.
  • The margins are real. SI at 65.5% and Connectivity at 66.7% adjusted EBITDA. Payback on compute build at modelled 2027-30 rent is about 8-12 months (about 3 years at Q3's realised $19.5B per GW), but payback continues to trend down. This reminds us of the early days of Bitcoin mining.
  • Starlink missed slightly on subscribers and beat on price. 13.6M consumer subscribers versus 14.29M expected, but ARPU of $69 beat. We model 15.2M by year end and keep $69 flat; the dynamic-pricing pilot in Kansas and Missouri (list price x0.8 to x1.2) suggests SpaceX would rather price for capacity than give it away (PCMag).
  • Earnings quality. Other income and a tax benefit added $1.415B, about 32% of $4.404B net income (Dialectic arithmetic on the 8-K). Operating income of $2.79B and the EBITDA beat (+33% vs consensus) are the cleaner reads.
  • Launch counts fell while tonnage per launch rose. Starship will replace Falcon 9 faster than we had previously projected. 35 launches (12 customer, 23 internal) versus 44 a year ago, as Falcon 9 makes room for Starship. Flight 14 on 28 September was the first revenue Starship mission, deploying 26 Starlink V3 satellites. Starship Flight 15 is now NET 12 November and will attempt the first ship catch (NASASpaceflight). One website reported it had already flown, which it hadn't; we model rockets, not wishes.
Page 04

What Changed Since $286

Our August target price of $285.83 came from our v5 model, which valued each business on revenue and margin paths. v10 is a different machine: each business starts from its physical drivers and the segments add up, to the cent, to a consolidated DCF. That makes a like-for-like bridge from $285.83 impossible, so we show two things instead: how the inputs moved, and the documented step changes inside the new build.

A note on discipline. Early drafts of the physical model let every gigawatt rent at full market price forever, and counted xAI's own use of its own data centres as rent it paid itself. Two fixes tamed it: xAI's own capacity is now valued at what Grok, Cursor and X actually earn, and total compute is capped at SpaceX's share of global SI infrastructure spend. Much cleaner and still super SXSI (read, ‘sexy’).

Table 2. Inputs: August memo (v5) vs this update (v11 base)

Item Post-Q2 memo (Aug 2026) v11 base (Nov 2026)
Fair value per share $285.83 $288.19
Discount rate (WACC) / terminal growth 9% / 4% 10% / 4%
Diluted shares / net cash 13,740M / $60.5B 13,811M / $51.0B (30 Sep)
Revenue 2027E / 2028E $204.2B / $318.1B (economic basis) $238.6B / $426.0B (reported basis)
Compute online YE26 / YE27 / YE28 2.1 / 7.9 / 11.2 GW 2.75 / 8.0 / 12.0 GW
Revenue per GW ~$9-14B per GW (memo-implied) External rent $51B avg 2027, $73.5B 2030, $67.8B 2035
Build cost per GW ~$50B (memo note) $40B 2027, $32B floor to 2031, $29.5B by 2035
Starship flights 2027 24 (Q2 Drive model) 63 flown, 65 path (210 in 2028)
Orbital data centres none 0.44 GW 2030, 3.6 GW 2035

Sources: Analyst Comparison and Summary tabs of the v11 model; August figures from the post-Q2 memo Annex A, C and G.

Table 3. Documented step changes inside the physical-driver build, base case, $ per share

Step Fair value Change
Physical model with demand cap, Q2-era inputs, WACC 9% $316.43
Q1-27 rent ramp ($42B to $60B per GW-yr) so Q1-27 lands inside the $33-37B guide $315.31 -1.12
Q3 actual margins held flat (SI 65.5%, Connectivity 66.7%) $338.48 +23.17
Year-end 2027 compute 13 GW (management) instead of 10 GW (Street) $425.33 +86.85
Discount rate 9% to 10% $346.24 -79.09
Case-ordering fixes, every Starship flight given a use and a cost, AI spend +6.2%/yr after 2030 $355.73 +9.49
Smooth Starship ramp, build cost -2%/yr and rent fade from 2032 $352.62 -3.11
Starfall return-capsule revenue from 2027 $352.75 +0.13
Other pre-v11 changes not separable in this file (2028-35 GW path, Starlink flight allocation, orbital headroom, calibration) $342.30 -10.45
Year-end 2027 compute 13 GW to 8 GW; demand cap binds from 2031 $318.80 -23.50
~80% internal compute plus Cursor $308.59 -10.21
$200B cellular/M&A provision $298.80 -9.79
Starship 2027-28 trim $298.87 +0.07
Rent growth 2028-30 from 10% to 7% a year (v11 base) $288.19 -10.68 (-64.56 vs v10)

Source: Summary tab sections L0, L, M and N (static offline recalculations, cumulative). The v10 to v11 steps are an approximation: they are recalculated live on the v11 file, the prior external compute share is taken as 50% from 2028, and the first step holds pre-v11 inputs this file does not separate. Measured on v11, the 9% to 10% step is -$69.27 a share (bear -$51.76, bull -$79.35).

Page 05

The Physical Inputs

Spreadsheets are cheap; gigawatts, rockets and satellites are not. Every revenue line starts with something you could, in principle, photograph from a helicopter or satellite right now. Here are the six charts that drive the model, all in the base case with bear and bull shown where they differ.

Physical inputs through 2035 from model v11. Base fair value is $288.19 per share. Open full size
Figure 1. Physical inputs at a glance, 2026–2035. Data from model v11, with bear and bull cases.

1. Compute online: from 1.7 GW to 24 GW

Terrestrial compute goes from 1.7 GW at 30 September to 2.75 GW at year end and 8.0 GW by the end of 2027, below the 13 GW management said it would be "surprised" to miss. The build then reaches 12 GW in 2028 and 20 GW in 2030, and the demand cap holds terrestrial capacity at 20 GW to 2035, with orbital data centres taking the total to 23.6 GW. The bear reaches 6 GW by end-2027 and the bull 11 GW, against the ~10 GW JPMorgan, Deutsche Bank and Raymond James cluster. This is absolutely enabled by SpaceX’s ability to produce gas turbines. Right now they are focused on vanes and blades; we expect them to produce gas turbines end-to-end by the end of the decade while adding an unimaginable amount of solar to the Texas landscape and then to the stars.

The cap is SpaceX's share of global SI infrastructure spend, one path for every case: about $1.0T in 2026, $3.0T in 2030 (the low end of Nvidia's $3-4T outlook), then +6.2% a year to $4.05T in 2035 (Dell'Oro and Morgan Stanley data-centre capex paths; ABI's 8.0% brackets it). Global SI spend 20% lower from 2029 changes value by about $2.60 a share in each direction, because the demand cap binds only from 2031.

One quirk worth admitting: the model adds about 5.25 GW in 2027 (to 8 GW YE) and then continues a smoothed catch-up through 2030 so cumulative consolidated capex 2026–30 stays near Ryan's ~$1.0T differentiator (base ~$975B), including a $200B cellular/M&A provision. Real life will be smoother. Value is not very sensitive to the timing, but the funding need is (see Financing).

Base-case total AI compute rises to 8.0 GW in 2027, 20.4 GW in 2030 and 23.6 GW in 2035. Open full size
Figure 2. AI compute online, GW at year end. Source: Colossus_DC- UPDATED and Orbital_DCs tabs.

2. Rent and build cost per gigawatt

External rent ramps from $42B per GW-year in Q1 2027 to $60B by Q4 2027 (2027 average $51B), then grows 7% a year to 2030 (bear 5%, bull 8%), is flat in 2031 and fades 2% a year from 2032 (bear 3%, bull 1%). Base rent ends 2035 at $67.8B. Orbital capacity rents at the same price as terrestrial; a 20% orbital premium would add only $4.03 a share, so we don't count on it.

Build cost per new gigawatt is $40B in 2027 (Vera Rubin era, near Milk Road's ~$45B), $36B in 2028, then a $32B floor to 2031 (JPMorgan cites management "mid-$30Bs", Deutsche Bank $35-36B, ARK $32.6B for the latest build) before falling 2% a year to $29.5B in 2035. SpaceX's own Q3 filing implies at least $36.4B per GW added over nine months (Dialectic arithmetic), and Musk says capex efficiency is "severely underestimated" (Teslarati). As analysts, it is our job to remember that every entrepreneur must necessarily be optimistic about their prospects in order to face the challenges of each day and continue to pull rabbits from hats. We adjust and discount accordingly. Every $10B per GW of extra build cost is worth -$35.33 a share.

About 80% of SpaceX's compute serves SpaceXSI itself by 2029. That share is valued at what Grok, Cursor and X earn per gigawatt (about $9B per GW-year in Q3, rising to $31B by 2030), not at market rent, so blended revenue per GW is $39.5B in 2030 rather than $73.5B. We think this could change because Grok Bot is probably the best product we’ve used, maybe ever. Entire companies, including Dialectic, are beginning to run on a substrate of Grok Bots. The product is a revelation. This is where the competitive dynamic really gets interesting. As Grok Bots hit the enterprise flywheel, competitors who rent their compute from SpaceXSI today face a dual threat – losing on lack of compute supply and product features. As Elon alludes, ‘live by the sword, die by the sword’ – this is true in the Sport of Kings (foundation models).

Base-case external rent reaches $67.8 billion per GW-year in 2035 as build cost falls to $29.5 billion per new GW. Open full size
Figure 3. Rent per GW-year vs build cost per new GW. Rent path is a Ryan Zurrer / Dialectic assumption; build-cost anchors from JPMorgan, Deutsche Bank, ARK and the Q3 filing.

3. Launches: Falcon 9 headed for early retirement, Starship gets the starting job next season

We have Starship flying 63 times in 2027 (65 on the path) and 210 in 2028 (slight trim vs v10 for FAA/political mud); Falcon 9 drops to 60 flights in 2027 and 24 in 2028 and is gone by 2029 (one Starship carries the Starlink capacity of about ten Falcon 9s). From there the base rises every year without much of a dip: 235 flights in 2029, 273 in 2030 and 549 in 2035. Each flight has a primary customer: Starlink growth and five-year replacement (164 a year), commercial and government satellites and lunar cargo (225 external missions by 2035), NASA HLS (6 a year), Mars windows (65 in 2035, cost only, no revenue), and then the granddaddy opportunity of them all – orbital data centres (53 in 2035), Starfall and a 5% test and spare allowance for other opportunities and maybe some Space Art.

For scale: SpaceX told the FAA it wants to launch 10,000 times a year, and Elon added that this is "just from Starbase Louisiana" (X, 28 Oct). Our 549 makes us, by a factor of about 18, the conservative ones in this conversation. We fly 549 of the 870 flights on the 2035 Starship path; we could not name a paying use for the other 321. If Elon brings thousands of flights home within the decade, it is almost certainly for ODCs and the valuation numbers begin to extrapolate to a level that butt-hurt bears cannot comprehend and thusly we will avoid that debate for the time being and hold this celestial upside in the back pocket for stargazing nights.

Base-case Starship flies 63 times in 2027, 210 in 2028 and 549 in 2035. Falcon 9 retires by 2029. Open full size
Figure 4. Launches per year by vehicle and use. Source: Launch_Services-UPDATED tab. Flights rise every year in every case (Summary check: ok).

Consumer subscribers grow from 15.2M at the end of 2026 to 23.2M in 2027, 56.2M in 2030 and 110.2M in 2035 (bear 68.2M, bull 158.4M), keeping steady at ARPU $69 a month, the Q3 actual, as that is a level that should wash the big three telecoms out over time. Direct-to-cell starts in 2028 at 15M subscribers and reaches 142M in 2035. The fleet grows from about 13,500 satellites at the end of 2026 to 29,800 in 2030 and 42,800 in 2035, with a fifth of it replaced each year. Connectivity revenue reaches $86.7B in 2030 and $183.0B in 2035, which is actually extremely conservative when you consider executives’ argument that connected devices will explode upward as robots come online en masse.

Not modelled, on purpose: a Boost Mobile purchase (talks only), the Grain 800 MHz spectrum price (undisclosed) and anything that requires US carriers to be friendly. T-Mobile's CEO says there is "no room" for a SpaceX MVNO. They can pound sand and will get consumed by the sandworm in the end.

Base-case fleet reaches 29,800 satellites in 2030 and 42,800 in 2035. Broadband subscribers reach 110 million and direct-to-cell subscribers reach 142 million. Open full size

5. Orbital data centres from 2030

Musk says orbital compute added per month "probably exceeds terrestrial around the end of 2028" and that future superintelligence compute "will round up to 100%" in space. We start later and dramatically smaller, as otherwise the valuation numbers become stratospheric: orbital data centres come online in 2030 at 0.44 GW and reach 3.6 GW by 2035 (about 15% of total compute), at the same rent as terrestrial (that may not be the correct way of looking at it), inside the same demand cap. They add $165B of EV in the base ($11.97 a share). Switching them off entirely reduces fair value by $13.01 a share.

Orbital data centres start at about 0.44 GW in 2030 and reach 3.6 GW in 2035. Open full size
Figure 6. Orbital data centres installed, GW. Source: Orbital_DCs tab.
Page 06

Five Years of Income Statements

Here is where the physical inputs end up: one income statement a year from 2027 to 2031, base case, revenue on the left and net profit on the right. Super Intelligence is the river; Starlink and Space are tributaries that would each be a large public company on their own. Net interest turns into net interest income from 2031 as the debt is repaid. But remember one major thing! The capex build will be double the $1T the Street is projecting by the end of 2030. Terafab will have a particle collider. Ask the Swiss how much one of those costs.

Table 4. Base case P&L, $B

$B 2027E 2028E 2029E 2030E 2031E
Super Intelligence (incl. xAI products) 199.9 370.8 483.5 711.0 838.0
Orbital data centres - - - 8.8 29.5
Connectivity (Starlink) 31.5 46.7 65.9 86.7 107.6
Space (external launch, incl. Starfall) 7.1 8.5 7.6 9.0 11.6
Revenue 238.6 426.0 557.1 815.4 986.6
Adj. EBITDA 153.6 276.4 362.5 531.6 643.4
Depreciation (tax basis) (62.6) (105.5) (146.0) (193.3) (207.5)
Net interest (expense) / income (5.8) (12.1) (10.8) (4.8) 5.8
Net profit 67.3 125.4 162.5 263.5 349.0

Source: DCF tab, rows 181-204 (live). Tax at 21% of pre-tax profit. Segment rows may not add to revenue due to rounding.

FY2027 revenue is 238.6 billion dollars and net profit is 67.3 billion dollars in the base case. Open full size
Figure 7. FY2027E. Revenue $238.6B, net profit $67.3B. The year of the 5.25 GW build: heavy depreciation and $5.8B of net interest.
FY2028 revenue is 426.0 billion dollars and net profit is 125.4 billion dollars in the base case. Open full size
Figure 8. FY2028E. Revenue $426.0B, net profit $125.4B. The first full year of 8 GW at higher rent.
FY2029 revenue is 557.1 billion dollars and net profit is 162.5 billion dollars in the base case. Open full size
Figure 9. FY2029E. Revenue $557.1B, net profit $162.5B. Falcon 9 is retired; Starship flies 235 times.
FY2030 revenue is 815.4 billion dollars and net profit is 263.5 billion dollars in the base case. Open full size
Figure 10. FY2030E. Revenue $815.4B, net profit $263.5B. Orbital data centres appear ($8.8B) and net interest expense falls to $4.8B.
FY2031 revenue is 986.6 billion dollars and net profit is 349.0 billion dollars in the base case. Open full size
Figure 11. FY2031E. Revenue $986.6B, net profit $349.0B. Net interest turns positive.
Page 07

Valuation: Sum of the Parts

Each business is valued by DCF over 5-, 10- and 20-year horizons, blended 65/30/5, with terminal values on normalised steady-state cash flow. The segments, less a cellular/M&A provision of about $155B booked at consolidated level only, sum to the consolidated EV. No segment is negative in any case or horizon; the smallest is the bear's orbital data centres at $5.5B on the 5-year horizon. We scarcely value the deep-space optionality in this model, but love to discuss the possibilities offline – yes, gold bugs, we’re going to go get lots of metals in the asteroid belt, so deal with it.

Table 5. Segment EV, blended 65/30/5, $B (and $ per share)

Segment Bear Base Bull Base $/sh
Colossus terrestrial data centres (rent × GW) 2,704 3,472 3,971 251.40
Orbital data centres 40 165 179 11.97
Starlink (incl. spectrum and direct-to-cell) 255 411 588 29.75
Space (external launch, Starfall) 21 36 45 2.58
Cellular/M&A provision (consolidated only) (155) (155) (155) (11.21)
Enterprise value 2,865 3,929 4,630 284.50
Net cash (30 Sep 2026) 51 51 51 3.69
Equity value / fair value per share 2,916 / $211.12 3,980 / $288.19 4,681 / $338.91 288.19

Source: Sum of Parts and Summary tabs (live). Per-share figures divide by 13,811M diluted shares; rows may not add due to rounding. Horizon values (base): 5-yr $259.88, 10-yr $338.77, 20-yr $352.68. If the 2027-28 funding gap were raised as equity at market instead of debt, base fair value would be $278.42.

Roughly 88% of enterprise value is the Colossus cluster. That is not a statement that rockets don't matter; it is a statement that rockets are how you get to own the data centres, the satellites and, eventually, the orbits. xAI's own compute is valued at product revenue, so Grok, Cursor and X sit inside the Colossus line rather than as separate fantasies. One risk we must admit is if the Dems/DSA sweep into political power and persecute Elon and SpaceX. That would be an unimaginable crime against humanity’s future but well within the realm of possibilities.

Table 6. Sensitivities, base case fair value $ per share (v11, live recalculation)

Change $/sh vs base
WACC 9% / 11% 357.46 / 239.02 +69.27 / -49.17
Terminal growth 3% / 5% 254.69 / 334.98 -33.50 / +46.79
Rent flat at $60B per GW-yr from Q4-27 267.75 -20.44
Rent path = bear (+5% 2028-30) / bull (+8%) 280.08 / 293.14 -8.11 / +4.95
Global AI spend 2029-35 -20% / +20% 285.58 / 290.81 -2.61 / +2.62
Year-end 2027 compute 10 GW (Street) instead of 8 292.89 +4.70
Build cost +$10B per GW, all years 252.86 -35.33
Imputed market rent on xAI's own use (the old way) n/a in v11 n/a
Orbital data centres off 275.18 -13.01
Starlink net adds -30% 283.78 -4.41
Diluted shares 14,143M 281.42 -6.77

Source: v11 model, live recalculation with each input restored after its run. Bull case with zero share persistence: $338.65 (vs $338.91). The imputed-rent switch has no effect in v11.

Starfall: a small capsule with a big idea

SpaceX has put up a product page for Starfall, a return capsule that carries up to 1,000 kg for weeks-long microgravity research and manufacturing, with 1 kW of power, Ethernet and a Starlink link (spacex.com/starfall). There is no price. The Q3 release says three commercial missions are signed; Reuters names the first customer, Space Cargo Unlimited, flying on the first commercial Starship mission in 2028 (Reuters, 15 Sep).

We model it modestly: 4 capsules in 2027, 12 in 2028, 45 in 2030 and 200 in 2035, at $15M falling to $10M each (Varda, the nearest comparable, is around $2.5-12M per mission), at the Q3 Space margin of 20.2%. That is $0.06B of revenue in 2027, $0.59B in 2030 and $2.0B in 2035, worth $0.13 a share (bull $4.5B in 2035). Volumes and prices are Dialectic assumptions [FLAG]. Doubling the volumes adds $0.14. Starfall is a call option on in-space manufacturing and point-to-point cargo; we have priced the option, not the dream.

Financing: the $275B year

Building 5.25 GW in a year is expensive. Base capex is $275B in 2027, unlevered free cash flow is -$149B, and the cumulative trough is -$178B. After the $51B of net cash, the funding need is $127B, of which the ~$40B Apollo-managed GPU finance lease covers the first $40B (CNBC; 8-K 26 Oct). Peak new debt is $133B. In the P&L, existing debt costs 5.86% (the June investment-grade notes), new debt 8.0% (the October GPU financing priced ~7.75-8.3%) and cash earns 4.0% [FLAG]. Fitch affirmed BBB+ after the lease (Fitch).

Once new debt is repaid (over three years), the model pays out half of cash above a $20B minimum: $329B in 2031, $363B in 2032 and $468B in 2035 in the base. This is a policy assumption [FLAG], not a forecast of what SpaceX will announce; it exists so that the cash pile doesn't earn absurd interest. The valuation is an unlevered DCF at 10%, so neither the interest nor the payout changes the fair value.

Page 08

Street View: How We Compare

We are above most of the Street on price and below it on 2027 physical capacity. The difference is mostly rent per gigawatt and how long it lasts. The smartest published 2028 numbers come from Sequoia's Shaun Maguire, a SpaceX shareholder since about the same time as us: $731B of revenue and 21 GW of terrestrial compute by the end of 2028, against our $426B and 12 GW. Run through our framework, his inputs give $247.82 a share, below our base, because his rent per gigawatt is lower over time. The Street is mostly arguing about the same multiplication while no one really allows themselves to price in all of the upside of space – ODCs, off-planet fabrication, lunar/Mars colonisation. Call it the X trolls that hold everyone back from recognising the real opportunities here.

Table 7. Dialectic vs selected analysts (each cell as published; n/a = not published)

Metric Dialectic v11 base SemiAnalysis Goldman Sachs Milk Road (Reidhead) Maguire (Sequoia)
Price target / fair value $288.19 (bear $211, bull $339) n/a (no PT) $245 (21 Oct) $244 "fundamental"; $342-427 with "Elon premium" (12 Aug) n/a (~50th-percentile case, 30 Oct)
Revenue 2027E $238.6B path to $300B ARR end-27 if 50% of capacity monetised $192.5B (post-Q3) $314B ~$274B
Revenue 2028E $426.0B n/a $205.6B (pre-Q3) n/a ~$731B
Compute YE27 8.0 GW ~10 GW 7.0 GW (pre-Q3) 13 GW 21 GW terrestrial + 2.8 orbital by YE28
Revenue per GW-yr $51B (2027 avg) to $73.5B (2030) $30-50B rental $40-50B spot, $20-30B term n/a $19.5B to $40B
Capex per GW $40B (2027), $32B floor to 2031 ~$50B n/a ~$45B (Vera Rubin) $55B to $30B
Starship flights 2027 / 2028 63 / 210 n/a n/a n/a 38 / 250

Source: Analyst Comparison tab of the v11 model, which cites each cell's publication and date. Other published targets: JPMorgan $500, Raymond James $800, Cantor $345, BofA $275, Deutsche Bank $250, Morningstar $70 fair value. Goldman's rate card is as cited by Deutsche Bank tables (Oct 2026).

Morningstar's $70 (which is completely mindless) and Raymond James's $800 bracket the Street by a factor of more than eleven.

Page 09

News Since the Print, and What Could Go Wrong

The news (none of it changes the base case yet)

  • Anthropic is the big tenant. Anthropic's public S-1 shows SpaceX compute commitments of up to $120.5B through 2029, rising to about $3.1B a month once all contracted capacity is delivered (expected December). Initial periods end in January and March 2027; after that, either side can exit on 90 days' notice (Yahoo Finance). Bloomberg identifies Anthropic as "Customer B", 34.5% of Q3 revenue (Bloomberg Law). A further tenant pays $1.11B a month from 1 December, per the CFO (name undisclosed).
  • Boost Mobile talks. The Washington Post reports SpaceX is in talks with EchoStar to buy Boost Mobile (7.38M subscribers at end-Q2) or use its network, after the big three carriers declined. No price (Reuters via Investing.com). SPCX touched $142.05 intraday on the report and closed at $123.62 on 5 November. [NOT MODELLED]
  • Tesla merger. Piper Sandler calls an all-stock SpaceX-Tesla deal "more likely than not" and says many believable scenarios imply an initial re-pricing of less than ±15% (Investing.com). Both boards have special committees; there are no terms. Our Post-Q2 edition covers the governance maths. [NOT MODELLED]
  • Starship Flight 15 is NET 12 November, pending FAA approval of the booster and ship catches; Flight 16, likely the first from LC-39A, is aimed for December. Falcon 9 is paused in Florida while 39A is readied for Starship.
  • SpaceX is leaving the old launch business on purpose. Crew Dragon stops by 2030, no commercial Falcon 9 sales for missions after 2028, and it "will no longer sell rides to the Moon" (Ars Technica). Our Falcon 9 exit by 2029 matches.
  • Grok Bot ate the Halloween candy. Musk says serving costs were "unsustainably high relative to revenue" after paying users hit weekly limits days early, and limits were tightened (TechCrunch). Demand is not the problem; unit economics on the consumer side are a watch item.
  • Generosity, Musk-style. Musk has pledged one of his own SpaceX shares to every US child in 2027, about 72M shares (CNBC). These are existing shares, so there is no dilution, only a generation of very motivated Starship fans.

Risks

  • Rent normalisation. Our base rent ($60B per GW-year at end-2027, ~$74B in 2030) is above Goldman's spot card ($40-50B) and SemiAnalysis's $30-50B, and far above term pricing ($20-30B). This is the single largest judgement in the model. Flat $60B costs $20.44 a share; the bear's +5% path costs $8.11.
  • Post-midterm Democratic control (W-070). Democrats won full control of Congress and the Texas governorship (Gina Hinojosa over Abbott, AP). Hinojosa has backed a moratorium on new AI data centres; Texas is ~27% of planned US data-centre GW per Wells Fargo, but Colossus 1 and 2 sit in Tennessee/Mississippi, not Texas, so the direct GW hit is modest. Separately she asked Texas Parks and Wildlife to pause any sale of ~1,300 acres around Boca Chica State Park next to Starbase (CNN); no sale had been approved. Federal oversight risk now extends to Starshield/contracts and to any Tesla–SpaceX merger; Cruz and DeSantis have warned of Democratic 'persecution' of Musk (Benzinga) — no investigation has been announced. Axios (4 Nov) and CNBC coverage frame the same theme. No moratorium is enacted. v11 already cuts YE27 base terrestrial compute from 13 GW to 8 GW partly to reflect permitting and political friction — do not double-count that cut as a further haircut on top of this risk. FLAG: Dialectic judgment, not a company guide change.
  • Customer concentration. One customer was 34.5% of Q3 revenue and can leave on 90 days' notice after early 2027. Anthropic itself reported Q3 revenue of $16.8B and positive cash flow (Axios), which helps, but concentration is concentration.
  • Funding and execution of the smoothed 2027–30 build. Base cumulative consolidated capex 2026–30 is ~$975B (SI + ODC + non-SI + spectrum + a $200B cellular/M&A provision for Starlink Mobile ground RAN and optional Boost/MNO, inside Ryan's ~$1T envelope — W-072 Qualcomm multiyear 5G from 2027, W-065 Boost talks). 2027 capex alone is ~$275B (was $425B in v10). Debt-funded at ~8% with interest on the gap; equity-funding the gap at market would cut base value to about $278. Segment-sum check of ~$155B is the PV of cellular booked consolidated-only so no segment EV goes negative.
  • Starship cadence. Five Starship flights in 2026, 63 in 2027 and 210 in 2028 (v11 trim for FAA/political mud; still rising every year, every flight with demand-backed work). Each step needs permits (the FAA is scoping up to 456 launches a year at Starbase) and catches that have not yet happened.
  • Legal and local. A federal judge let the NAACP's Clean Air Act suit over Colossus 2's gas turbines proceed (CNBC); Louisiana residents are suing over the Starbase tax deal. We expect further litigation and permitting challenges; adverse rulings could delay projects or add cost.
  • Sovereign competition. We were in India earlier this week and there is zero chance they ever let Starlink into the country. The Ambanis dominate it. Same story for China, potentially Canada (which is laughable) and much of western Europe. A range of countries won't rely on Starlink for government communications. None of this changes 2027, but it caps the long tail of government revenue outside the US.
  • Earnings quality and dilution. 32% of Q3 net income came from other income and a tax benefit; Cursor equity awards and lock-up expiries add supply. Shares at 14,143M instead of 13,811M cost $6.77.
Page 10

Conclusion

SpaceX's second public quarter turned our biggest assumption, that SI compute would be a real business at GOAT margins, into reality. In order to keep the numbers where the compute is for the time being – on Earth – we raised the discount rate, capped demand at what the world can spend, gave every rocket a job and every gigawatt a tenant, and still arrived at $288.19 a share in the base case — still about 2.3x the $123.62 close — after cutting YE27 compute to 8 GW and booking ~$200B of cellular/M&A inside a ~$1T cumulative-capex envelope through 2030.

The bear ($211.12) is well above today's $123.62. The bull ($338.91) needs YE27 at 11 GW, ~80% internal compute with Cursor inflecting, and rent growing ~8% a year to 2030 — a tighter range than v10's march toward $500. Soft landing zones were ~$200/~$280/~$350; physicals drove the outcome.

Page 11

Sources and Method Notes

All model figures come from the Dialectic SpaceX model v11 (6 November 2026), base case unless stated; "Dialectic arithmetic" marks our own calculations on published figures, and [FLAG] marks Dialectic assumptions without a public source. Soft FV landing zones ~$200/~$280/~$350 guided calibration but did not force-fit cells. Prices are official closes (SPCX $123.62 on 5 Nov). Nothing here is investment advice; Dialectic holds SPCX.