SpaceX (SPCX) — IC Memo: What To Do at $135 | DogBone Capital Research
memo
SpaceX (SPCX) — IC Memo: What To Do at $135
Aug 9, 202623 views
IC memo distributed privately June 11, 2026, hours before SPCX priced at $135. Intrinsic per-share $59.99 / $112.48 / $169.05 (bear/base/bull); flow-supported incl. L3 $70.45 / $141.61 / $245.72. Accumulation band $95–115, full conviction below $100. Conviction dials calibrated by a 711,397-comment sentiment engine — 14 pre-registered hypotheses, 4 supported, 2 refuted. Post-IPO scorecard added at publication: $225.64 high June 16; $104.83 low Aug 3, inside the band; $133.11 close Aug 7. Every number traces to a cell.
SPCX | Nasdaq | Prices June 11, 2026 | Lists June 12, 2026 | Fixed offer $135.00 ($1.77T; $75B primary)
Confidential | SpaceX (SPCX) | June 10, 2026 — For internal IC use only. Not investment advice.
Originally prepared June 10, 2026, and distributed privately by email on June 11, 2026, hours before IPO pricing. Private until first public publication on this platform, August 9, 2026 — memo body unchanged from the distributed version; the post-IPO scorecard was added at publication. Original email trail and model file available on request.
INVESTMENT THESIS
The $135 offer prices SpaceX at 1.20x our Base-case intrinsic value of $112.48 and 1.38x Damodaran's post-prospectus $97.83. Intrinsic value does not clear the offer; flow does — the $75B book is oversubscribed and index fast-entry rules add passive demand in staggered waves roughly 5/10/15 trading days post-listing. Our framework therefore separates two decisions: the allocation (a flow-and-scarcity trade, justified only by Layer-3 dynamics) and the position (an intrinsic decision, with an accumulation band of $95–115 and full conviction below $100).
What is new in this memo: a 711,397-comment social-sentiment engine — built, benchmarked and validated in-house — now calibrates the model's conviction and risk-weighting dials, and a full audit of the valuation engine fixed two defects that had been flattering every scenario. Sentiment moved risk weights, never revenue.
Bear
Base
Bull
Mgmt Ambition
Street Chatter
Intrinsic per-share (L1+L2)
$59.99
$112.48
$169.05
$171.55
$185.30
Offer $135 / intrinsic
2.25x
1.20x
0.80x
0.79x
0.73x
Market-supported (incl. L3)
$70.45
$141.61
$245.72
$239.06
$262.66
Enterprise value ($T)
0.72
1.42
2.17
2.20
2.39
2036 revenue ($B)
296
400
510
523
1,252
WACC / terminal g
9.00 / 4.5%
8.37 / 4.5%
8.25 / 4.5%
8.50 / 5.0%
8.50 / 4.5%
Terminal-value reliance
70%
75%
76%
78%
77%
Model: SpaceX_SPCX_Valuation_Model_v10.2 (10-Jun-2026). Intrinsic basis = L1 triangulation + L2 optionality; L3 scarcity OFF by default on all scenarios. Per-share = (EV + net cash $1.85B + $75B raise) / 13.3B fully-diluted shares.
A note on the seller's intent (a reading, not a claim). The structure of this deal is consistent with a company raising capital, not insiders selling out: the full $75B is primary, proceeds are retained for the buildout, and no secondary component rides along. One coherent reading — and it is only a reading — is that SpaceX views the IPO as a financing transaction and expects public shareholders to be rewarded later, as its private holders have been: Tesla's 2010 IPO investors and SpaceX's tender participants (Baron Capital reports ~54% annualized since 2017 across 27 rounds) were both paid for patience, and the same arithmetic that prices this deal at $1.77T plausibly makes its founder the first trillionaire. We present this lens on alignment because it shapes how the lock-up and post-IPO supply may behave; Sections 6 and 9 carry the counterweights, and nothing in our entry bands depends on it being true.
1. The Decision: What To Do at $135
This memo answers one question: what do we do on June 12. Our May initiation framed SpaceX as an extraordinary business at an extraordinary price; the work since — the S-1 reconciliation, a Damodaran cross-check, a ground-up audit of the model, and a year of social discourse scored comment-by-comment — sharpens that into bands:
Price zone
Framework reading
Below $100
Aggressive accumulation. Below Damodaran's independent intrinsic ($97.83) and our Bear-Base midzone; the market is paying us for AI-overreach fear.
$100 – 115
Core accumulation band. Brackets our Base case ($112.48); Anthropic-cancel stress ($103.21) sits inside it.
$115 – 135
Patience. Above Base on fundamentals; payment for flow we do not own.
$135 allocation
Tactical participation only. Justified by Layer-3 dynamics (oversubscribed book, FTSE fast-entry ~day 5, no listed substitute) — a trade with an exit discipline, not a thesis. With-L3 Base of $141.61 explains the price; it is not intrinsic value. Polymarket consensus: 71% odds of a day-one close above $2T (≈$150+/sh) — the pop is already priced, which makes it the trade, not the thesis.
Above $170
Trim zone. Price enters Bull / Management-Ambition territory ($169–172) that our own sentiment work does not corroborate.
The honest tension, stated up front: the institutional order book (~$250B of demand) disagrees with the social discourse (net sentiment −0.21). Both are data. Sentiment measures the crowd's conviction; the book measures allocations. Our architecture keeps them in separate layers precisely so this disagreement is visible rather than averaged away.
2. What Changed Since the May Initiation
Hard numbers replaced estimates. FY2025 actuals tie to the S-1 exactly (revenue $18.67B; segments 4.09 / 11.39 / 3.20; EBIT −$2.59B; net cash +$1.85B; 13.3B fully-diluted shares incl. the raise).
Contracted compute widened. Google joined Anthropic as an LLM-platform tenant: $920M/month (~110k GPUs) Oct-2026 through Jun-2029, alongside Anthropic's $1.25B/month through May-2029. Both carry cancellation features; both are modeled as toggles, not assumptions.
Independent cross-checks arrived. Damodaran's post-prospectus intrinsic: $97.83/sh ('offer too rich'). Morningstar: ~$780B equity (≈ $64/sh post-raise — almost exactly our Bear). Goldman's roadshow deck: $474B revenue by 2030 — our Street Chatter case, quarantined as the AI-overreach scenario.
The named-investor spectrum went on the record. Baron Capital — a SpaceX holder through 27 tender rounds since 2017 — placed a $1B order at the offer and is publicly framing SpaceX at $10–30T within 10–15 years, with Starlink alone at ~$14T. Coatue's framework reads the market's rising value-per-launch as business-quality migration (constellation → platform), sizing Starlink against a $200–400B global telco profit pool. Damodaran's post-prospectus update lands at ≈$100/sh including the raise — and he is publicly not buying at $135.
The book is now oversubscribed. Demand reached 2x ($150B) within a day of the roadshow and stands at roughly $250B against the $75B offering as of this morning (Barron's, June 10), with pricing June 11 and a fixed $135 — demand confirmation for the allocation decision, not for intrinsic value.
3. The Sentiment Engine: 711,397 Scored Comments
We scraped a trailing year of Reddit and Hacker News discourse (balanced 50/50, 711,617 gated comments), passed every comment through an evaluative gate, and scored the 293,270 evaluative comments (58.8% gated out as non-evaluative noise) on a 1–5 investment-conviction scale, each tagged by segment, risk driver and audience type. Scoring was done by an LLM selected through a staged bake-off: 18 models attempted across five providers on a stratified sample, 11 producing clean comparable output, ranked head-to-head against Claude Opus 4.8 as an external frontier judge and confirmed non-inferior by a paired McNemar test — the winner agrees with the judge on 85% of gate decisions and 96% of scores (±1) at roughly 2% of the judge's cost ($90 billed vs ~$31,000 like-for-like at observed token volume) — the winner (gemini-3.1-flash-lite) was confirmed statistically non-inferior by a paired McNemar test. Full-corpus billed inference: ~$90 (vs a $22 metered estimate — variance disclosed in the methodology document). Scoring validity: 99.97%.
The headline — the crowd is mildly bearish (weighted mean 2.88, net −0.21) — is the least useful number. The signal is in the concentration:
Segment
n (eval.)
Mean
Net
Read
Connectivity / Starlink
27,084
3.28
−0.01
Most constructive; least controversial
Space / Launch
73,186
3.23
−0.02
Constructive; Starship event-driven, bimodal
AI / xAI
159,171
2.74
−0.29
Clearly bearish; largest, best-evidenced slice
Cross-company (merger)
29,507
2.43
−0.45
Most bearish; skeptical of combined entity
Risk driver
Net
Read
Governance / Musk
−0.75
The single most decision-relevant finding: on 44,287 evaluative comments, the 'Elon premium' reads as a liability, at 15.1% salience.
Regulatory
−0.66
Bearish but only 4.6% of discourse — an under-priced tail the crowd is not watching.
Financial / Valuation
−0.38
Skeptical of exactly the props (multiples, scarcity) our old model leaned on.
Tech / Product
+0.02
Engineering credibility intact across 135,692 comments — supports the base case.
Hype / Momentum
+0.10
The only net-positive driver, at just 7.7% share: no euphoria to fade, no panic to buy.
Three structural findings matter as much as the levels. First, retail is not the exit liquidity: informed and retail audiences score statistically equal (2.96 vs 2.93), refuting the most common bear framing of this IPO. Second, there is no FOMO momentum: weekly sentiment is flat across all 53 weeks into pricing. Third, the discourse is silent on the single largest idiosyncratic risk — Anthropic at ~47% of 2027E AI revenue draws just 1.85% salience — so that risk must be carried analytically (it is; Section 5). Platform mix does not drive any of this: the Hacker News vs Reddit gap is 0.04 points.
In the crowd's own words:
"The valuation Elon chose for SpaceX is pretty simple. He got out a calculator and punched in the numbers to see what the valuation would need to be for him to become the world's first trillionaire on IPO day." — Reddit, Feb 2026 (Governance)
"SpaceX is too big to fail. It's important for national security. I wonder if Elon wants to tangle all his businesses into SpaceX so they are all kept afloat by SpaceX's importance." — Hacker News, Feb 2026 (Cross-company)
"They had nothing but great things to say about how the Starlink team listened... and the changes Starlink made to keep observing clutter-free. Other constellations, not so much." — Reddit, Aug 2025 (Starlink)
4. How Sentiment Fed the Model
Discipline first: sentiment calibrates conviction and risk-weighting. It is never allowed to move a revenue line, a margin, or a contract schedule. Each change below is tagged to the hypothesis that justifies it (14 pre-registered hypotheses, scored against the corpus before any model edit was made).
Finding (hypothesis)
Model dial
Change
Base impact
Grok uniformly skeptical, AI capital intensity (H11, H2)
* L3 is OFF on every scenario; these dials shape only the with-L3 memo value ($141.61 Base). The governance widening cut the Base L3 stack from 40% to 31%.
Equally important is what sentiment refused to support. It killed the exit-liquidity bear argument (H3 refuted — worth stating against our own caution). It withheld support for any positive scarcity premium, which is why L3 stays off. And it does not corroborate the house-view revenue raise in Section 6 — a dissent we put on the record rather than bury.
5. The Model: Three Layers, One Audited Engine
The architecture separates what the business earns (L1), what it might become (L2), and what the market will pay (L3). All five scenarios run one identical, self-contained engine — a 15-year FCFF forecast to 2040 with a Gordon terminal value.
The May initiation weighted DCF at only 40% because the DCF then ran at $326B — a quarter of the multiples-based answer — and we papered over the gap with comps and scarcity. That defect is gone. The rebuilt Base DCF carries $1.42T on cash flows alone, so DCF now anchors at 85%. Mechanics: tax ramps 10% → 25% as NOLs burn off; a fading execution-risk premium (100bp yrs 1–3, 50bp yrs 4–6) replaces any capex-linked WACC kludge; depreciation is true 5-year vintage (each year = trailing-5-year average capex, pre-2026 base rolling off) — the audit found the old cumulative method never retired vintages, inflating the terminal tax shield in every scenario by ~$7/sh; and the LLM-platform contracts (Anthropic + Google) run on explicit monthly schedules with cancellation toggles that now propagate into the terminal value (a second audit fix — previously the toggle physically could not reach the TV).
Layer 2 — Optionality ($170B Base, probability-weighted, 12% discount to each option's own target year)
Embedded option
Pot. ($B)
Target yr
Prob.
Multiple
PV ($B)
National security / Golden Dome
10
2030
0.90
10x
57.2
Enterprise agentic-AI platform
25
2032
0.25
12x
38.0
Orbital compute (incremental)
50
2035
0.23
8x
33.2
Starship commercial transport
20
2035
0.45
10x
32.5
Human spaceflight / tourism
5
2035
0.45
8x
6.5
Deep-space infrastructure
30
2040
0.08
5x
2.5
The NatSec probability (0.90) is the only one near certainty — it is anchored by an awarded $2.29B Space Force SDN Backbone contract, not a story. The agentic-AI line is where sentiment cut (Section 4). Orbital compute (0.23 probability, $33B PV) deserves its own note: management's published roadmap is far more aggressive than our dial — Starlink-V3-derived AI satellites at GB300-rack power (150 kW peak / 120 kW sustained, ~3ms from LEO), an aspirational 1 GW/yr annualized run-rate exiting 2027 scaling 10x per year, and a 100M-sq-ft 'Terafab' for in-house silicon. Gavin Baker's point prediction for orbital-compute reality is 2H-2028 to 1H-2030; Google is already partnered to fly TPUs. Our 0.23 prices a credible engineering path on an aggressive timeline — if conviction changes, this is the dial to move, not the revenue lines.
Layer 3 — Scarcity (a toggle, OFF everywhere)
L3 prices what the market pays beyond value: scarcity of a listed substitute (+14%), IPO scarcity (+6%), strategic premium (+7%), AI-infrastructure scarcity (+8%), liquidity/sentiment (+4% after the corpus cut it in half), governance discount (−8% after the corpus widened it). Net +31% in Base — shown as a memo ($141.61), never added to intrinsic value. The May memo blended this in; the CIO should know we have stopped doing that. Why 31% and not a round guess: the stack is calibrated to observables, then decomposed so each line can be struck independently. Secondary marks ($130 Forge/Sacra) and the Hyperliquid synthetic ($203) bracket the with-L3 Base of $141.61; the float is thin (555.6M new shares, ~4% of shares out) against an oversubscribed book; FTSE fast-entry forces passive buying ~day 5; and the only two corpus-calibrated components both moved AGAINST the premium (liquidity halved, governance widened). It is a reconstruction of an observed premium, not an aspiration — which is exactly why it stays out of intrinsic value.
6. The Five Cases: One Worldview per Scenario
Each scenario in the model is a coherent worldview, not a sensitivity run. The five cases below narrate what each case believes, what would have to be true, and what it pays — figures from SpaceX_SPCX_Valuation_Model_v10.2 and our May 2026 initiation. They run in the order the committee should debate them: what the upside is (Bull), what we underwrite (Base), what management promises (Management Ambition), what we must survive (Bear), and what the Street is selling (Street Chatter). Each footer also reports a with-L3 value — what the market may pay while scarcity dynamics hold (no listed substitute, IPO scarcity, index fast-entry, a locked float). Read L1+L2 as the long-term fundamental anchor and the with-L3 line as short-term pricing context: that premium should compress as float grows and substitutes list, which is why it stays out of every headline number.
BULL — $169.05 per share: First-Mover Compounding at Unprecedented Scale
The worldview: no company in history has assembled this combination of first-mover advantages across multiple trillion-dollar markets simultaneously. The S-1's own track record reads like a syllabus in industrial disruption:
First private company to develop and launch a liquid-fuel rocket to reach orbit (2008)
First private company to successfully dock a spacecraft with the International Space Station (2012)
First to propulsively land (2015) and refly orbital-class rocket boosters (2017)
First to deploy a large-scale LEO broadband satellite constellation (2019)
First private company to transport astronauts to orbit, restoring America's crewed launch capability (2020)
First to manufacture consumer-grade phased-array user terminals at scale (2022)
First to deploy a large-scale LEO satellite-to-mobile constellation (2025)
First to build a gigawatt-scale AI training cluster and largest coherent supercomputer (2026)
First gigawatt-scale Megapack battery installation (2026)
Uniquely positioned to build orbital AI compute at scale
Each milestone is a capability moat. Reusable rocketry alone keeps launch costs roughly 10x below any competitor, which is what makes Starlink's economics possible: 10.3M subscribers at a 63% EBITDA margin against the ~20% typical of legacy satellite operators. The moat translates directly into unit economics.
Starlink: the profit engine. Q1 2026 Connectivity revenue was $3.26B with $1.19B of operating income, and segment EBITDA margin has climbed from 41% in 2023 to 63% in 2025 — a trajectory more consistent with software than telecom. Penetration sits below 1% of the S-1's $1.61T broadband-and-mobile TAM; the runway, not the run-rate, is the asset.
Contracted compute: $15B annualized at high margin. The Anthropic Cloud Services Agreement prices COLOSSUS and COLOSSUS II capacity at $1.25B per month through May 2029 — $15B annualized at ~60% EBITDA margin, largely monetizing infrastructure already built. Mutual termination rights make this high-quality rather than bond-like revenue; in the Bull, frontier-model scale keeps compute demand growing through and beyond the 2029 renewals.
Convergence optionality. Golden Dome national-security architecture, Starship point-to-point transport, orbital AI compute, and the Macrohard agentic-AI platform live in the L2 layer — probability-weighted and discounted, $116B to $295B across the five scenarios and $247B in this one. None of it sits in the revenue forecast; it is upside from capabilities no other entity can replicate.
Everything compounds together. In the Bull, the moats reinforce: 2036 revenue reaches $510B, AI terminal margins 36%, the optionality layer's probabilities rise (Starship transport 0.65, orbital compute 0.40), and the WACC eases to 8.25% as execution de-risks. The offer is 0.80x this case — the market is paying Bull prices for a Base business, which is precisely the gap our entry bands arbitrate.
2036 revenue $510B · WACC 8.25% / g 4.5% · EV $2.17T · Offer = 0.80x · with-L3 $245.72
BASE — $112.48 per share: Disciplined Optimism
The worldview: extraordinary asset quality, honestly costed. The title is carried over from our May initiation because nothing about the philosophy changed — only the engine underneath it got audited.
Starlink is the engine. Sixty-one percent of revenue, margins climbing 63% toward 65%, and the mobility wins (American's 500+ aircraft, United's 1,000+) compounding enterprise value on top of residential broadband. The sentiment corpus makes it the least-controversial business in the company — which is why it leads the story.
Contracted compute pays for patience. FCFF turns positive in 2026 (+$1.2B): $11.5B of contracted LLM-platform revenue at ~60% margin does in one year what the May model needed until 2027 to achieve.
($B)
2025A
2026E
2028E
2030E
2033E
2036E
Space / Launch
4.1
5.1
8.1
12.9
25.6
50.9
Connectivity (incl. mobile)
11.4
14.3
22.8
36.3
73.8
152.8
AI (ex-LLM platforms)
3.2
4.6
9.4
19.4
57.1
168.0
LLM platforms (contracted)
—
11.5
26.0
19.2
24.0
27.8
Total revenue
18.7
35.6
66.3
87.7
180.5
399.5
EBITDA
6.6
15.7
31.8
41.5
90.9
189.1
EBITDA margin
35.3%
44.2%
47.9%
47.3%
50.3%
47.3%
Capex
20.7
13.5
16.2
21.7
33.3
59.8
FCFF
(14.2)
1.2
13.2
16.3
43.2
92.5
AI reaches breakeven, not glory. AI ex-platforms reaches EBITDA breakeven by 2028 and a 30% terminal margin — deliberately adjacent to Damodaran's 25% EBIT-equivalent ceiling rather than above it.
The cliff is respected; renewal at trend. LLM revenue dips in 2030 as the contracts roll off as written, then renews at trend — a visible scar in the revenue line instead of a smoothed one, defended by revealed demand (Colossus builds in 122, 91, 66 days; Google queuing behind Anthropic). Capacity this scarce tends to re-let; the cancellation toggle exists for the world where it does not.
What it pays. The offer is 1.20x this case. Our accumulation band ($95–115) brackets it, and the Anthropic-cancel stress ($103.21) lands inside the band — the Base does not need rescuing by the overlay.
Near-term pressure, long-term conviction. We expect early downward pressure: AI cash consumption will dominate headline risk, and a $75B raise is its own supply. That is what the accumulation band is for. The ceiling is not the constraint — management's $28.4T TAM is context rather than forecast, but with penetration below 0.1% in every segment, even modest gains over 15 years sit above this path.
2036 revenue $400B · WACC 8.37% / g 4.5% · EV $1.42T · Offer = 1.20x · with-L3 $141.61
MANAGEMENT AMBITION — $171.55 per share: Management Executes Its Own Roadmap
The worldview: take management at its word, then cost the words honestly.
The roadmap is public, and it is enormous. Orbital AI compute at a 1 GW/yr annualized run-rate exiting 2027, scaling an aspirational 10x per year; a 100M-sq-ft Terafab for in-house silicon; AI satellites that are, in management's framing, GB300 racks with 70-meter wings. This is the company's stated plan, not our extrapolation.
Vertical integration is the margin story. Successive Colossus data centers stood up in 122, 91, then 66 days; in-house chips target a fraction of merchant silicon cost; and the company positions itself as the natural co-design partner for next-generation data-center architecture. If the integration thesis lands, the cost curve bends for every business at once.
Where we hold the line. This case carries a 40% terminal AI margin — well above Damodaran's 25% enterprise ceiling — and 5.0% terminal growth. Both are flagged in the Sensitivity sheet as the levers that matter; the AI-margin row alone moves ~$5/share per 5 points. It also runs the highest terminal-value reliance of the five (78%): the most faith-dependent case on the page.
Read it as management's case, costed. Even granting the roadmap with on-book capex, the answer ($171.55) lands within a few dollars of our Bull — a useful sanity check that our Bull already embeds most of what execution can deliver inside a decade.
2036 revenue $523B · WACC 8.50% / g 5.0% · EV $2.20T · Offer = 0.79x · with-L3 $239.06
BEAR — $59.99 per share: Contract Cliff and Compression
The worldview: the contracts end as written, AI economics never mature, and the multiple does the rest. Nothing in this case requires a disaster — only the absence of grace.
The cliff is modeled, not feared. Anthropic runs to its May-2029 term and is held flat at $15B with no renewal and no growth; Google's lease ends June-2029 at zero. Both sit behind live cancellation toggles — flip Anthropic off entirely and the Base loses $9.27/share, so the Bear is not even the worst contracted world, just the most plausible cold one.
AI never finds its margin. Terminal AI EBITDA margin is 22% against 30% in Base — the corpus's most-evidenced read (net −0.29 on 159,171 comments) corroborates exactly this caution: heavy capex, cut-throat pricing, and a sector that competes its margins away.
Capital intensity is real. FY2025 AI capex was $12.73B — 397% of the segment's own revenue — against a $6.35B operating loss, and Q1 2026 AI capex alone ran $7.72B. PitchBook called the financials 'reckless.' The Bear simply assumes the intensity never earns its keep.
Governance concentration. Musk's span across SpaceX, Tesla, xAI, X, Neuralink, and The Boring Company is key-person risk no discount fully absorbs, and the dual-class structure concentrates control over capital allocation — including the xAI merger itself. The corpus prices governance harder than any other driver (net −0.75). One refinement on the May framing: sentiment refuted the retail-exit version of this argument (H3), but the funding-pressure version survives — an IPO of necessity can misprice without retail being the mark.
Compression does the rest. At 94x trailing revenue, deceleration re-rates the multiple long before cash flows matter. The Bear carries a 9.00% WACC and the lowest terminal-value reliance of the five cases (70%) — it is the scenario least dependent on faith.
What survives: Starlink. As our May initiation put it, even this scenario describes "still a transformational business." Starlink keeps compounding through the cliff, and the floor lands almost exactly on Morningstar's independent ~$64. Bears here lose patience, not capital.
2036 revenue $296B · WACC 9.00% / g 4.5% · EV $0.72T · Offer $135 = 2.25x this case · with-L3 $70.45
STREET CHATTER — $185.30 per share: The Roadshow Case, Quarantined
The worldview: the underwriters' arithmetic, run through our engine, kept deliberately separate from our own conviction (carried in the model workbook as the Street Upside sheet).
Goldman's numbers on our engine. AI revenue from $3.2B (2025) to $322B by 2030 — a ~100x ramp — with total revenue of $474B by 2030 and $1.25T by 2036. We re-pathed the segment split to Goldman's own (Launch $8.3B / Starlink $144B / AI $322B in 2030) so the case is faithfully theirs, not a caricature.
Why it lives in quarantine. Its capital efficiency (revenue added per dollar of cumulative capex) is 0.78 — the worst of the five cases — and terminal-value reliance runs 77%. It is the only scenario where the top line outruns capital discipline, which is the signature of a story priced on momentum.
The honest label: AI overreach. Damodaran's post-prospectus critique names the prospectus's $28T TAM as fantasy (his AI TAM: $3–4T) and calls AI overreach the central risk. Our May initiation quarantined the same case for the same reason; v10.2 keeps the label.
Its use to the committee. It prices the whisper. If post-IPO trading carries the stock beyond ~$185, the market is paying more than the sell-side's own arithmetic supports — a sell signal generated by the bulls' math rather than ours.
2036 revenue $1,252B · WACC 8.50% / g 4.5% · EV $2.39T · Offer = 0.73x · with-L3 $262.66
7. Stress Testing
Per-share, Base engine, recomputed after every v10 change (replica matches the live sheet to the cent):
WACC \ g
3.5%
4.0%
4.5%
5.0%
5.5%
7.50%
120.00
130.93
145.50
165.90
196.49
8.00%
106.26
114.22
124.46
138.12
157.23
8.37%
97.98
104.40
112.48
122.96
137.09
9.00%
86.58
91.16
96.76
103.77
112.77
9.50%
79.34
82.92
87.22
92.47
99.04
Anthropic full cancellation (toggle ON): Base $112.48 → $103.21. The model survives its biggest customer leaving — and lands inside our accumulation band.
AI terminal EBITDA margin ±5pp: ±$5.2/sh ($107.30 at 25%, $117.67 at 35%). The single biggest fundamental lever — consistent with Damodaran's 'AI overreach is the key risk'.
AI 2036 revenue $120B → $220B: $108.66 → $116.65 — deliberately damped, because AI value rides on margin and capital intensity, not the top line alone.
Dilution to 14.5B shares: $103.18 — comp pool creep is a real, quantified risk, not a footnote.
Terminal-value reliance runs 70–78% of DCF EV across scenarios. That is the honest cost of valuing a 2040 story in 2026, and the reason the WACC x g grid above deserves more committee time than any other exhibit.
8. Key Risks
Anthropic concentration: ~47% of 2027E AI-segment revenue from one counterparty with a 90-day mutual cancellation right — and the discourse never mentions it (1.85% salience). Quantified: −$9.27/sh on full cancellation; Bear additionally holds Anthropic flat at $15B with zero growth.
Governance: dual-class control, key-person dependency, related-party history (the xAI merger itself). The corpus prices this harder than any other driver (net −0.75); our L3 governance discount widened accordingly, and it is the first risk we would expect to bite in a drawdown.
Regulatory tail: spectrum, ITAR, FAA cadence, national-security review — net −0.66 in discourse but only 4.6% salience. Low crowd attention + high real exposure = the classic under-priced tail; carried as an explicit analyst overlay.
AI capital intensity: FY2025 AI capex ran 397% of AI revenue. Our base assumes intensity fades to 45% by 2030 and 15% terminal; every point of stickiness is roughly a dollar of per-share value. Damodaran names this the memo's central risk in his own words — 'overreach in the AI space': capex and R&D spent as if the prospectus's $28T TAM were real (his AI TAM: $3–4T), under an 85%-voting-control structure that lets one person make that bet unilaterally.
Valuation compression: 94x trailing and ~50x 2026E revenue at the offer. Any growth deceleration re-rates the multiple before it re-rates the cash flows.
9. What Would Make Us Wrong
Too low: Goldman's $474B-by-2030 is the underwriters' case, not ours — but if AI revenue tracks even half its ramp by 2028, our Bull becomes the floor and the Street case stops being a quarantine ward. Starship reaching commercial cadence early would move $32B of L2 into L1. And mechanically, switching the model's L3 overlay ON maps the offer to a very strong 3–5-year bull case the framework presents without endorsing: Base $141.61 and Bull $245.72 — roughly 2x the $135 offer at the Bull stack — with even the modest-return path from the offer requiring both the flow story to hold and execution to land. Too high: Anthropic walking at the 6-month mark (the toggle exists because Musk himself framed it that way), AI terminal margins capping at Damodaran's 25% enterprise ceiling, or the house-view anchor raise (Section 6) proving to be exactly the overreach the corpus warned about. The model's dissent ledger is symmetrical on purpose: Morningstar ≈ our Bear, Damodaran ≈ our Bear-Base gap, Goldman ≈ our Street.
10. Conclusion
SpaceX at $135 is a great business at a price that already pays for the next three years of execution. Our base-case intrinsic value is $112.48; the offer asks 1.20x that, and everything above intrinsic is being paid for scarcity and flow — real forces, with real expiry dates (lock-up, index inclusion, the first AI-capex headline miss). The framework's answer is patience with teeth: participate in the allocation only as a flow trade with an exit discipline, set the accumulation band at $95–115 where our Base, the Anthropic-stress case and Damodaran's independent number all cluster, and treat any post-IPO air pocket — which our own May memo predicted — as the entry the offer refuses to give us.
Every number in this memo traces to a cell, every sentiment claim to a scored comment, and every judgment call to a version note that says who made it and why. That is the standard this committee should hold us to.
Gating: evaluative-content gate excludes non-evaluative comments from all means (58.8% gated out). Every reported mean is evaluative-only; blended platform numbers are never reported without the per-platform split (HN is structurally hedged — ~65% of its comments gate out as non-evaluative — so evaluative volume skews Reddit 58/42).
Scorer selection: ~18 models across five providers (Anthropic, OpenAI, Google, Groq, xAI) benchmarked on a topic-stratified ~300-row sample after ~100-row screens; initial consensus ranking discarded as circular; all models re-ranked against Claude Opus 4.8 as external frontier judge (a 120-row hand-labeled gold set built as fallback); cost gate <$30; winner confirmed non-inferior by paired McNemar (85% gate agreement, 96% scores ±1, at ~2% of the judge's estimated cost). Winner: gemini-3.1-flash-lite via Google's Gemini Batch API; full corpus ~$90 billed vs a $22.47 metered estimate — variance attributed to retries and debug-window interactive calls, disclosed rather than smoothed.
Design: 14 hypotheses pre-registered before scoring; each maps to a specific model lever. Verdicts: 4 supported, 2 refuted (incl. the exit-liquidity thesis), the rest partial/unresolved — reported as such.
Known limitations, disclosed: HN topical bleed, quantified: only ~8.7% of AI-segment comments are explicitly SpaceX/xAI-specific — but that subset is MORE bearish (net −0.39 vs −0.28 sector-general), so the bleed dilutes rather than inflates the read; HN lacks engagement weighting (means are robust; 'high-engagement' quotes are Reddit-only); sentiment measures discourse, not order flow — the 2x book is the proof.
Appendix B — Model Exhibits
Live screenshots from SpaceX_SPCX_Valuation_Model_v10.2. The control panel (bottom of the first exhibit) drives the L3 overlay and both contract-cancellation stresses across all five scenario sheets from a single cell each.
Exhibit B1 — Valuation Summary: five scenarios, one engine; intrinsic basis with the with-L3 memo line; central control panel.
Exhibit B2 — Base Case engine (drivers and forecast, 2025–2030 shown): margin ramps, capex intensity, tax ramp, contracted LLM-platform revenue.
Exhibit B3 — Sensitivity: WACC x terminal-growth grid centered on the live Base ($112.48), plus single-variable stresses.
Exhibit B4 — Key sentiment figure: net sentiment by risk driver across 293,270 evaluative comments; the governance read (−0.75) sized the L3 governance discount.
[Exhibit images are not reproduced in this web version; the distributed PDF and the model file are available on request.]
DISCLOSURES
This document is for internal investment-committee use, and for informational and educational purposes only. It does not constitute investment advice, a recommendation, or a solicitation to buy or sell any security. The analysis is based on SpaceX's Form S-1 (SEC Accession No. 0001628280-26-036936), publicly available data, the author's proprietary financial model (SpaceX_SPCX_Valuation_Model_v10), and a proprietary social-sentiment corpus scored with benchmarked large-language models; AI-assisted scoring is subject to classification error, which the validation protocol bounds but does not eliminate. All forward-looking projections are estimates subject to material uncertainty. The author may hold positions in securities mentioned. Past performance is not indicative of future results.
Sources: SpaceX Form S-1 (May 20, 2026); Damodaran post-prospectus valuation (Jun 1, 2026); CNBC; FT/Yahoo Finance (Goldman roadshow forecasts, Jun 4, 2026); CNBC/TechCrunch/Tom's Hardware (Google compute lease, Jun 5, 2026); SpaceNews ($2.29B Space Force SDN award); Bloomberg (American/United Starlink); LSEG/FTSE Russell (IPO fast-entry); Morningstar via CNBC (Jun 3, 2026); proprietary Reddit/Hacker News sentiment corpus (711,397 scored comments, Jun 2025–Jun 2026); Baron Capital SPCX investor call (R. Baron, Jun-2026); All-In Podcast E274 (G. Baker); Coatue All-In Summit presentation (P. Laffont); SpaceX company update on orbital compute (E. Musk, Jun-2026); A. Damodaran post-prospectus valuation update; Polymarket.
Methodology
The sentiment engine has its own 19-page companion document: "SPCX Sentiment Engine — Methodology & Results" (June 10, 2026) — corpus construction (896,418 comments collected → 711,617 gated → 711,397 scored; 293,270 evaluative), the staged bake-off (~18 models across five providers, judged against a frontier reference, paired-McNemar non-inferiority), the 14 pre-registered hypotheses and their verdicts, pipeline architecture, and known limitations. Full-corpus inference cost ran under 1% of the frontier-only alternative. Appendix A above is the summary; the full document is available on request.
Post-IPO scorecard (added at publication)
Everything below this line was written on August 9, 2026. Nothing above it changed, save one disclosed correction noted at the end of this section. Verified against exchange data (daily bars, June 12 – August 7).
Date
Tape
Framework line it tested
Jun 12 (day one)
Opened $150.00, closed $160.95
Polymarket's priced-in pop ($150+) delivered; allocation was the trade
Jun 16
Intraday high $225.64
Deep in the trim zone (>$170); short of both the with-L3 Mgmt Ambition ($239.06) and Bull ($245.72) marks — flow, not intrinsic
Late June
Consolidation, closes ~$153–171
Between the patience band and the trim zone
July
Steady decay through the $135 offer
The predicted air pocket: AI-capex headline risk plus post-raise supply
Aug 3
Intraday low $104.83
Inside the accumulation band ($95–115); brackets Base ($112.48) and the Anthropic-cancel stress ($103.21)
Aug 4–5
Q2 print: revenue +92% YoY to $7.8B, a beat; stock sold off on $18.4B quarterly capex ($15.8B AI)
The memo's named risk — capital intensity dominating headlines — arrived on schedule
Aug 7
+15.8% close at $133.11
Lock-up expiry (911.5M shares) absorbed without the feared supply break; Terafab co-investment announced. Price back at the offer
Score, honestly kept: the accumulation band caught the low almost exactly ($104.83 vs $95–115), the trim zone marked the top within days, and the with-L3 ceiling ($245.72) was never breached. The band was bought; the framework's next test is whether Base intrinsic ($112.48) holds as float grows. One correction to the distributed version, made here rather than silently: the June text characterized the order book with a specific oversubscription multiple; the final book size was never verified, so this publication says only "oversubscribed."
Prepared June 10, 2026. Distributed privately June 11, 2026. First published August 9, 2026.
