The Body and the Option
For a $1.75T IPO, the first question is not whether it is expensive. It is: which part are you actually buying. Split it into two pieces -- the knowable, priceable body, and the option held aloft by narrative -- and the fragile point of the whole valuation comes into view.
IPO fact layer (hardened post-S-1, re-verified 2026-06-09)
| Item | Figure | Source / quality |
|---|---|---|
| Latest cash tender valuation | $800B ($421/share pre-split; = $84.2 after the 5:1 split) | Fortune 2025-12-13 (hard, tender-cleared) |
| xAI merged into SpaceX | $1.25T (SpaceX ~$1T + xAI $250B) | CNBC/Bloomberg 2026-02 (hard) |
| IPO pricing | Fixed $135/share, no range, no book-build; implied valuation ~$1.77T (7th largest US-listed company, above Tesla's ~$1.6T) | CNBC/Fortune/Reuters 2026-06-03 (hard) |
| Structure | whole-company (Starship + Starlink + xAI + X), all primary, no existing holders selling | Bloomberg 5/21 + SpaceNews 06-04 (hard) |
| Raise size | $75B (555.6M Class A shares), largest IPO in history (above Aramco's $25.6B) | Fortune/Reuters 06-03 (hard) |
| Timeline | Books close 6/10, pricing 6/11, Nasdaq listing 6/12 | CNBC 06-03 (hard) |
| Ticker | SPCX | CNBC 06-03 (hard) |
| Use of proceeds | First listed item: "expand AI compute infrastructure"; repay the $20B bridge loan within 6 months | S-1 / NCSU 06-04 (hard) |
| Gray-market comp | Forge $128.90 (06-09) / Nasdaq Private Market $125 (05-26), both below the offer price | Forge/NPM (hard) |
Price-anchor discipline (updated 06-10): at the first draft (05-28), $1.75T / $75B / 6/12 were all pre-S-1 soft numbers; all have since landed hard with the S-1: the $135 fixed price x total share count implies ~$1.77T. Note the direction of two new hard facts: (1) gray-market prices sit below the offer price -- the offer price is not a market-clearing price, it is a seller-set price; (2) the cash tender six months ago was $84.2 (split-adjusted), so $135 = +60% in half a year, and that 60% has no corresponding fundamental increment (Q1'26 net loss $4.27B, with losses accelerating) -- what it corresponds to is the IPO narrative premium itself. The valuation positioning below still argues from the "body = real / upside = option" structure.
The body -- knowable, priceable, real
Starlink -- the profit engine (61% of revenue, the only profitable segment)
10.3M subscribers (2026-03-31). 2025 revenue $11.4B (+50% YoY), operating profit $4.42B, 61% of SpaceX's total revenue; Q1 2026 operating profit roughly $1.2B. This is the only self-funding cash-flow engine inside SpaceX -- the hard floor under the body's valuation.
Launch -- cash cow + near-monopoly
SpaceX total company revenue in 2025 was $18.67B (+33% YoY), of which Starlink was 61%, with the remainder mostly launch services; Q1 2026 revenue $4.69B, adj EBITDA $1.13B. 165 Falcon launches in 2025, roughly 82% of all global commercial launches. This is an already-monetized, already-profitable installed franchise that leans on no narrative whatsoever.
Starship -- the cost curve (known path, not yet topped out)
12 test flights (7 successes, 5 failures). V3 / Block 3 first flew on 2026-05-21 with a "mixed success" result, adding Raptor 3 and a LEO docking / quick-disconnect interface. The booster already achieves "chopstick" catch recovery; the upper stage has not yet attempted a catch; the orbital propellant-transfer demo is expected within 2026 and remains incomplete. Full reuse is an engineering problem on a known path, not a physics no-go zone -- book it as a solvable cost curve.
xAI -- vertically integrated compute (and simultaneously the cash burner)
Merged in 2026-02 at $250B. Roughly $500M ARR, roughly $1B/month of cash burn, and it owns X. Its place in the narrative is the junction of "shared compute + Starlink + orbital data centers." Worth noting: xAI is the only part of the body that is burning cash while going public -- and it is precisely the thread that stitches the body's valuation to the "space compute" narrative. In other words, the business most dependent on narrative sits exactly on the seam of the narrative premium.
The proposition
Body = Starlink cash flow + launch near-monopoly + Starship cost curve + xAI compute. The first two have real cash flows; the latter two are priced on growth -- together they land roughly at the December $800B tender plus xAI's $250B, i.e. the merger-implied $1.25T order of magnitude.
The slice stacked from $1.25T up to $1.75T (roughly $500B) rides on one narrative vehicle: space AI data centers + Kardashev II. The entire IPO valuation question therefore reduces to an option-pricing problem: what is that orbital-data-center option actually worth. To answer it, you must take apart its physical feasibility -- which is exactly what the next several sections do.
The Problem It Actually Solves: the Energy Wall
Space AI data centers solve neither cooling nor latency -- they solve the energy wall. This has to be said first, because it sets the ceiling on what the option's underlying is worth.
The real bottleneck of terrestrial hyperscale AI compute is not GPUs. It is electricity, plus land and permitting. GW-scale AI campuses have already slammed into the physical limits of grid capacity, substations, and transmission lines; and building a new power plant (especially nuclear) and connecting it to the grid runs a 5-10 year permitting cycle -- a hard constraint AI expansion cannot route around. Land, water, community permits, carbon policy: all of it is "Earth governance friction" that ground-based builds cannot dodge.
Space bypasses exactly that layer: in Sun-synchronous orbit, solar power is near-24/7, near-unlimited, permit-free, interconnection-free, land-free, and free of carbon controversy. It liberates AI compute expansion from "constrained by Earth's grid + land + politics" into "constrained only by launch cost + heat rejection + attrition."
A conservative correction
What should be anchored is not "free cooling" but energy and space unconstrained by Earth governance. That is the kernel of why "Kardashev II / harness the Sun" is not entirely hype -- it points at the energy ceiling of civilization-scale AI compute: on Earth, the ceiling is politics and the grid; in space, the only ceiling left is launch. Supporting evidence: Musk's exact words at Davos (2026-01) were "lowest cost to generate AI compute will be in space... within two to three years, three at the latest"; SpaceX has filed with the FCC for roughly 1 million satellites to serve as orbital data centers, explicitly framed as the first step toward "Kardashev Type II."
So the option's underlying is real -- the energy wall is a hard constraint on terrestrial AI, and space genuinely routes around it in principle; this is not a fantasy with no underlying. But say the whole sentence: a real underlying does not mean the option is in-the-money. The outcome rests entirely on the delivery side.
Heat Rejection and Attrition -- One and the Same Temperature Constraint
The energy wall supplies the demand-side motive (ground power keeps getting more expensive); whether the supply side can deliver hangs on two gates -- and worse, those two gates are coupled into a single constraint through operating temperature.
Heat rejection (the bottom layer)
In vacuum, radiation is the only path left, and it is locked down by the Stefan-Boltzmann law: radiated power is proportional to area x emissivity x T^4. Area is mass and mass is cost; emissivity is already pressing the blackbody limit -- the only nonlinear lever is T^4: every increment of operating temperature scales heat rejection by the fourth power. So the real breakthrough in cooling is not "build bigger radiators" but "make the chips and loops run stably at higher temperatures."
Attrition (coupled to heat rejection)
A 3-year refresh cycle, no card swaps on orbit, and radiation steadily shortening hardware life. The crux: running hot for cooling's sake -> accelerates electromigration, depresses MTBF -> drives up attrition; radiation -> bit-flips plus cumulative dose -> drives attrition the same direction. In one sentence: cooling wants high temperature, lifetime wants low temperature, and the two collide head-on at the single knob called "temperature."
What the coupling means
These are not two independent engineering problems but opposing pulls on the same variable (operating temperature T): any move that "raises T to improve cooling" directly worsens attrition economics; any move that "lowers T to extend life" directly weakens cooling. Feasibility thus collapses into one question: does a temperature sweet spot exist at all, where both sides hold simultaneously.
The Three-Way Mutex -- a Physics No-Go Candidate
The temperature heat rejection wants, the temperature frontier process nodes can survive, and the process node radiation hardness demands -- each pair conflicts, fencing off a sweet spot that may not exist at all.
| You want | It forces you to accept | Nature of the conflict |
|---|---|---|
| High-temperature heat rejection (T^4 efficiency) | GPUs running at 200-400 C | Silicon CMOS tops out at ~100 C; leakage is already runaway at 250 C |
| Frontier compute density | 3nm leading-edge process | Leading-edge nodes are the least heat-tolerant and least radiation-hard |
| Long life and radiation hardness | Fall back to older nodes + redundancy | Older nodes lack compute density, defeating the point of going to orbit |
This is not an engineering matter of tuning parameters. It is a structural mutex of CMOS physics stacked on process-node trade-offs: for high temperature you must leave silicon; for compute density you must use the least heat-tolerant leading-edge nodes; for lifetime you must fall back to older nodes that lack the density. Whether the sweet spot exists is a materials-physics question -- not a capital question.
The Feasibility Criteria Chain -- Difficulty Is Not Evenly Distributed
Twist the preceding sections into one multiplicative criteria chain: three conditions must hold simultaneously at the same temperature sweet spot; miss any one and the whole thing fails.
Difficulty ranking (1-10, 10 = near a physics no-go zone)
| Gate | Difficulty | Nature | Timeline | Who controls it |
|---|---|---|---|---|
| Launch cost (full Starship reuse) | 5 | Engineering + capital, known path | 5-8 years | SpaceX [x] |
| High-temperature two-phase cooling (two-phase flow in microgravity) | 6 | Engineering; prototype validated on ISS, needs a million-fold scale-up | 10-15 years | Solvable by SpaceX |
| On-orbit attrition / whole-satellite turnover economics | 6 | Manufacturing scale-up, tied to the Starship cost curve | 10 years | SpaceX [x] |
| Radiation-environment lifetime | 7 | Materials; rad-hardening is mature but trades off against frontier nodes | Long-term | Industry |
| High temperature x frontier node x rad-hard three-way mutex | 9 | Materials-physics no-go candidate, needs a paradigm jump | 15-20+ years / possibly unsolvable | Outside SpaceX's circle |
Capped by a single point, not averaged
Difficulty is not evenly distributed; it is single-point capped. Overall feasibility = the highest gate = 9. A multiplicative criteria chain is only as feasible as its weakest link -- even if launch cost went to zero, that cannot rescue a sweet spot stuck in materials physics. And this capping gate (maturation of wide-bandgap compute chips) lies entirely outside SpaceX's circle of control.
Wide-Bandgap Computing -- Still Early-Stage Research
Wide-bandgap semiconductors (GaN / SiC) are the only direction that could conceivably punch through the "high temperature x compute" wall. But for AI compute logic, the 2026 state of the art stops at the "single flip-flop" level -- nowhere near engineering readiness.
| Dimension | Measured state, 2026 | Implication |
|---|---|---|
| Temperature ceiling | WBG devices top out at ~400 C absolute; rising leakage current already compresses the operating limit further at ~250 C | Tighter than expected |
| Demonstrated logic level | Best reported GaN logic: a single D flip-flop operating at 160 C | Not an AI accelerator -- several orders of magnitude away |
| Materials gap | No high-performance p-type GaN -> no complementary CMOS logic | The foundation of complex digital compute is missing; it can only be mixed with silicon MOS, which drags the whole system back to silicon's limits |
| Actual role today | GaN / SiC data-center penetration is roughly 17% in 2026, all of it power / power-conversion devices; Intel's GaN-integrated logic remains a research-stage prototype | Not compute logic |
The grounded conclusion
Wide-bandgap AI compute logic currently sits in early-stage research -- "a single flip-flop, missing the p-type transistor" -- not engineering. This is not something SpaceX can accelerate with money -- it is a paradigm jump in materials physics. Capital can compress engineering timelines; it cannot move a materials system that has not yet even assembled its complementary logic devices.
Musk Is Attacking the Wrong Wall
Starship (difficulty 5) solves launch cost -- and launch was never the capping gate. The real gate is the high-temperature, radiation-hard compute chip (difficulty 9), outside SpaceX's circle of competence, dependent on a materials breakthrough from the entire semiconductor industry.
Even with Starship perfectly reusable and cost-to-orbit at zero, space compute could still be locked out entirely for want of "a frontier AI chip that runs stably at 300 C." SpaceX has thrown all of its engineering firepower at the wall it is best at, and the one most solvable for it -- entirely rational as engineering; but on the question "when does space compute go in-the-money," the wall it is attacking is not the binding constraint.
The CEO narrative test
"Cheapest compute in orbit within three years" plus "Kardashev II" is classic top-of-cycle narrative-speak, isomorphic to Ellison's "$300B Stargate" and Altman's "AI is bigger than the internet" -- a far-horizon vision in service of this year's IPO valuation. Physically possible does not mean feasible within three years.
The sequencing is itself the signal: the Davos pitch (2026-01) + the million-satellite FCC filing + the whole-company IPO all advanced in the same window, with the valuation stacked up to $1.75T along the way. A CEO in a hurry to shout + a simultaneous IPO = the classic distribution-prep signal. The narrative's function is to discount a 15-20-year, possibly unsolvable option into this year's offer price.
The Sell Side's Math -- Who Wrote the Model, Who Collects the Fees
On the eve of pricing (2026-06-10), the sell-side camp packaged $1.77T as "still cheap." Lay their numbers on the table and contest them line by line. Start with one structural fact: the people producing the models and the people waiting to collect the fees are the same people.
The interest structure (before any argument)
| Fact | Figure | Source |
|---|---|---|
| GS = lead-left underwriter (23-bank syndicate; controls pricing and allocation) | Fee pool $800M-$1B, GS takes the largest share | Yahoo Finance 2026-06 |
| GS leads the $20B bridge loan (BofA/Citi/JPM/MS participating), refinancing X/xAI's 12.5% high-coupon debt | IPO proceeds must repay it within 6 months; matures 2027-09 | Octus / IFR 2026-05 |
| MS runs the retail allocation channel; its subsidiary EquityZen simultaneously operates SpaceX pre-IPO secondary trading | -- | Yahoo 2026-06 |
This is not "sell-side optimism" -- it is creditor pricing
The bank handing out the "100x AI revenue" model is simultaneously lead creditor on a $20B loan whose repayment source is this IPO's proceeds. The author of the model is the beneficiary of the model's conclusion -- this is not GS valuing the company for investors, it is GS pricing its own balance sheet. During the pre-pricing research quiet period, no named GS analyst signed off on any of it; every number reached the public only as roadshow material relayed through media (Investing.com in June called out the "underwriter + analyst dual role" credibility problem by name).
Line-by-line debunk
| Sell-side claim (source) | The math, unwound | The counter |
|---|---|---|
| GS roadshow: xAI AI revenue 2025 $3.2B -> 2030 $322B (100x), "bigger than AWS / NVDA"; total company revenue $474B by 2030; EBITDA $6.6B -> $352B (Yahoo/247wallst 06-04) | 5-year CAGR ~151%; first +388% in 2026, then four more consecutive years of +83% | The industry-wide 2030 LLM revenue SOM base case is $300-500B (EIP AI hype cycle endgame report, bottom-up). GS is asking you to believe that a company with $3.2B of current revenue, a $6.4B operating loss, and trailing GPUs will, within 5 years, swallow 64-100% of the entire industry's revenue pool. That is not a forecast; that is selling the denominator as the numerator. |
| MS roadshow: 2040 revenue $3.4T (182x), adjusted EBITDA $2.7T+ (Stocktwits/BigGo 06-05) | 15 consecutive years of +41%/yr; $3.4T = ~8.5 times 2025 Apple; EBITDA margin 79% | No company in history has delivered a 79% EBITDA margin (Apple ~34%, NVDA peak ~65%). A 14-year extrapolation is unfalsifiable -- its function is not forecasting, it is handing $1.77T an anchor that says "still 2x upside." The MS that produced the number happens to run the retail channel: "182x" is all a retail investor needs to hear. |
| Passive index buying + float of only 3-4%: "index funds will eat 30% of the float within 15 days of listing"; Cramer calling a double to $4T (247wallst 06-04) | Engineered scarcity: the price is set by the smallest marginal dollar | Tiny float + forced passive buying = a borrowed price; the historical prototype is Palm in 2000-03 (carve-out listing with a tiny float, pricing that implied negative value for parent 3Com, -90% within a year). And the lockup ladder is queued up behind it: 20% unlocks 2 days after the first earnings report, ~45% cumulative within 135 days (CNBC 05-21). A price built by passive inflows gets dismantled by unlock supply. Cramer publicly calling a double is itself a sentiment-top reading. |
| "2x oversubscribed, ~$150B in orders, single $10B+ institutional tickets" (Seeking Alpha 06-09) | A fixed price with no book-build: oversubscription discovers no price | Meanwhile the gray market sits below the offer: ~$125-129 split-adjusted (NPM 05-26 / Forge 06-09). $135 is not a market-clearing price, it is a seller-set price -- the real marginal buyer off-exchange will only pay 95% of it. The order book's bustle and the gray market's discount coexist; the latter is the price signal. |
| ARK: 2030 base $2.5T / bull $3.1T; Starlink 2030 revenue $200B at 80%+ gross margin (Fortune 2025-06-11) | Starlink actual 2025: $11.4B revenue, 39% operating margin -> requires 17.5x in 5 years plus a doubling of margins | A year-old pre-IPO model; the methodological track record is the TSLA $7,000 price target. Satellite bandwidth is a contested commodity (Kuiper has begun deployment); the 80% gross-margin assumption has no mechanism behind it. |
Dissecting the 2x subscription -- the biggest order book in history, the coldest multiple among hot IPOs
The sell side markets "$150B in orders, biggest ever" as evidence of heat. But an order book's absolute size is a numerator game -- what determines allocation tightness and the post-listing bid ladder is the multiple. Put SpaceX's 2x into a hot-IPO comparison set (multi-source verified 2026-06-10):
| IPO | Raise | Oversubscription | Day 1 | 3-12 months |
|---|---|---|---|---|
| Alibaba 2014 | $25B | 14-17x | +38% | Doubled within a year |
| ARM 2023 | $4.9B | 10-15x | +25% | Choppy, then +176% (6 months) |
| Facebook 2012 | $16B | Institutional >5x (reportedly 15-20x) | +0.6% (stabilized) | -50% in 4 months |
| Reddit 2024 | $0.5B | 4-5x | +48% | As much as +562% within a year |
| Aramco 2019 (most comparable) | $25.6B | Institutional book 2.95x (4.65x overall incl. retail) | +10% | Broke issue within 3 months (-16%) |
| SpaceX 2026 | $75B | ~2x | ? | ? |
Rivian 2021 ($12B, multiple undisclosed, +29% day one, -80% to date) and CoreWeave 2025 (downsized offering, then +250%) show the multiple has no stable predictive power for the 3-12 month path (Cornelli & Goldreich, JF 2003 likewise only confirm a positive correlation between the multiple and day-one returns) -- the medium term is decided by fundamentals and supply, not by book heat.
What 2x means -- a three-layer reading
Layer one, lateral position: 2x is the lowest multiple in this comparison set; the only comparable is Aramco's 2.95x institutional book -- and Aramco is one of the two weakest aftermarkets in the set (broke issue within 3 months). Genuinely hot IPOs (ARM / Alibaba) run at 10-17x. "Biggest order book in history" is true because the $75B denominator is the biggest in history, not because demand is extraordinary.
Layer two, squeeze out the padding: institutions routinely pad orders in deals like this to hedge allocation haircuts -- in the SpaceX coverage a strategist said it verbatim: "investors routinely inflate orders in deals like this" (Yahoo 06-09); on the academic side, Cornelli & Goldreich (JF 2003) confirm that rationing expectations are endogenous to book-building. In a 10x book the padding is harmless; squeeze the customary inflation out of a 2x book and real demand at $135 most likely sits at 1x or below -- corroborated by the gray market at $125-129 (below the offer price): two readings of the same fact.
Layer three, what it implies for the stock (two forces on a time lag): short term (day one to month one), a ~4% float + forced index buying + allocation rationing means the scarcity machine can still manufacture a pop -- that has nothing to do with whether demand is real; Aramco also opened +10%. Medium term (months 1-6), price is handed over to two different variables: the real demand floor (around 1x, with no unfilled demand ladder waiting underneath) and unlock supply (+20% two days after earnings, ~45% by day 135, supply expanding ~11x). Low multiple + gray-market discount + lockup ladder: the historical rhyme is Aramco -- a respectable day one, below issue within three months. If day one does deliver a big scarcity-machine pop, that is not evidence against the thesis; it is the gift of a distribution window.
Reconciling the sell-side story with this report's framework
GS's $322B AI-revenue story, placed inside this report's structure, amounts to: discounting Section V's difficulty-9, outside-SpaceX's-circle, 15-20-year-or-never materials gate as if it were already open. And the timing is worse -- the first listed use of proceeds is "expand AI compute infrastructure," meaning retail's $75B is picking up exactly the hyperscaler-grade AI capex torch, running straight into the 2026 H2-2027 capex clearing window (see the EIP AI hype cycle endgame report). Every line of the sell-side model requires both "the extortion economy extends indefinitely" and "materials physics surrenders on schedule" to hold; Sections II-VII of this report have argued the latter fails, and endgame report 3.3.1/3.3.2 argued the former fails.
Count the Body; Time Value Only for the Option
Translate the physics conclusions of the preceding sections back into valuation: the orbital data center is a long-dated call option (pure time value left, nothing more) and does not enter the body's valuation. The body = Starship + Starlink + xAI's vertical integration.
Valuation structure
Body (counted): Starlink cash flow ($11.4B revenue / $4.42B operating profit, the only profitable segment) + launch near-monopoly ($18.67B total company revenue, 82% of global commercial launches) + Starship cost curve (known path) + xAI compute vertical integration -- roughly corresponding to the merger-implied $1.25T order of magnitude.
Option (time value only): the orbital AI data center. The underlying is real (it hedges the terrestrial power wall), but realization hangs on a single materials make-or-break (difficulty 9, outside SpaceX's circle, 15-20 years or possibly never). Give it pure time value only, not body status -- it is the narrative vehicle for the slice from $1.25T up to $1.75T, not cash flow in hand.
The single forward indicator to watch
Shift down from "Starship reuse" to "wide-bandgap compute chip maturity"
The true make-or-break for space compute is not launch but the maturation of wide-bandgap (GaN / SiC) logic / AI compute chips -- the lowest, slowest, and least SpaceX-controllable link in the entire chain. The specific point to watch: complementary GaN logic (i.e. the p-type GaN breakthrough) progressing from "a single D flip-flop" toward "complex digital blocks."
The key may not be in SpaceX's hands
Whoever breaks through first on wide-bandgap compute chips is the party that actually holds the key to space compute -- and that may be some semiconductor-materials company, not SpaceX. In option-pricing terms: even if the upside does pay off, most of the value capture may not land on SpaceX at all.
No threat to the edge-inference side
Space data centers map to training / centralization (latency-insensitive); edge / CDN maps to inference / edge (latency-sensitive); the two are not fighting over the same ground. The space data center is the energy-liberated version of "the center," not a substitute for edge inference -- which means it is no headwind to any edge-compute position.
Fair value -- three anchors, one range
Fully diluted share count ~13.1B ($1.77T / $135). Three independent anchors converted to per-share:
| Anchor | Whole-company valuation | Per share | Nature |
|---|---|---|---|
| Morningstar fair value (2026-05) | ~$780B | ~$60 | Independent third party, the only public FV model; already discounts xAI (calls its finances "reckless") |
| 2025-12 cash tender (split-adjusted) | $800B | $84.2 | A real-money market-clearing price from six months ago -- the same stock, not a model |
| Merger-implied body (this report's Section I basis) | $1.25T | ~$95 | The upper-bound reading of the body: includes xAI's full $250B while xAI is burning $1B/month |
EIP fair value: $60-95, midpoint $75-85
The three anchors bracket a range of $60-95/share, midpoint $75-85 -- mutually consistent with where the gray market actually trades ($125-129, a price still carrying IPO-scramble sentiment, yet below the offer) and with the conditional re-entry zone of the EIP $135 IPO vet report ($60-85). Against the $135 offer price: a premium of ~60-80%, i.e. fair value sits -37% to -44% below the offer. RESET-scenario floor: if the AI capex clearing plays out, the $250B for xAI needs writing down and the Starlink multiple compresses in tandem; the bear anchor slides to $50-60. Note this fair value already gives the body full credit -- it contains zero orbital-data-center option value, because Sections V-VI established that as a difficulty-9, outside-the-circle option worth time value only.
Where the most credible future growth lives
One criterion only: proven unit economics x penetration expansion -- growth that needs no new physics, no new materials, and no extension of the extortion economy is the only growth that can be discounted into a valuation. Ranked by that standard:
| Growth line | Current evidence | Credibility |
|---|---|---|
| 1. Starlink broadband subscriber ramp | Subscribers 8.9M -> 10.3M (single quarter), 2025 revenue $11.4B (+50% YoY), 39% operating margin; international penetration plus aviation / maritime access still early | Highest -- a curve already running; pure execution |
| 2. Direct-to-cell (D2C) | Sells coverage to carriers off the existing constellation, near-zero-marginal-cost TAM expansion; needs no new technology gate | High -- asset reuse, revenue-share model |
| 3. Starshield government / military | Dedicated-network expansion on a ~20% US-government revenue base; stickiest contracts, least price-sensitive | High -- but also the flip side of single-point political risk |
| 4. Starship cost curve compressing launch prices | Difficulty 5, known path; but global commercial launch TAM is only ~$10-15B/yr | Medium -- a margin story, not a revenue story; its real role is cutting costs for lines 1-3 |
| xAI AI revenue / orbital data centers / Mars | A trailing GPU player burning cash + a difficulty-9 materials gate + no revenue model | Not counted -- these three are precisely the narrative prepaid inside $135 |
Sum lines 1-3: the Starlink complex (broadband + D2C + Starshield) is the only growth engine that has real cash flow, real growth rates, and zero physics gates all at once -- every cent of the fair value is effectively pricing it. This is also the basis for the vet report's "conditional re-entry re-anchored on Starlink": after a RESET, what you want to buy is not the SpaceX story, it is Starlink's cash flow plus a launch monopoly thrown in for free.
Verdict: Buy the Body, Not the Narrative
The SpaceX body is a good business, worth pricing off that $1.25T-scale set of knowable cash flows and cost curves. The $1.75T IPO target, by contrast, pre-discounts into this year's offer price an option that is 9/10 in difficulty, sits outside SpaceX's circle, and may have no solution at all. Buy the body; do not buy the option the narrative has already discounted -- its true constraint is materials physics, neither inside SpaceX's engineering circle of competence nor within reach of its capital to accelerate. Watch the wide-bandgap compute-chip maturity curve closely; until it genuinely moves, the option remains just an option.
Data Quality
Valuation / IPO figures (verified live, 2026-05-28): $800B/$421 tender = Fortune 2025-12-13 (hard); xAI $1.25T merger = CNBC/Bloomberg 2026-02 (hard); Starlink 10.3M subs / $11.4B 2025 revenue / $4.42B operating profit = CNBC/Bloomberg 2026-05-21 (hard); SpaceX 2025 total company revenue $18.67B / 165 launches = electroiq/Sacra/Wikipedia 2026 (hard); Starship 12 flights + V3 first flight 5/21 = CNN 2026-05-22 (hard); FCC million-satellite orbital data center filing + Davos "2-3 years" quote = Decrypt/Data Centre Magazine 2026-01 (hard).
Post-S-1 re-verification (2026-06-09, multi-source web cross-check): $135 fixed price / 555.6M shares / $75B / ~$1.77T / books close 6-10, pricing 6-11, listing 6-12 / SPCX = CNBC + Fortune + Reuters 06-03 (hard); all primary, no existing holders selling = SpaceNews 06-04; use of proceeds led by AI compute + $20B bridge = S-1 / NCSU 06-04; gray market Forge $128.90 (06-09) / NPM $125 (05-26); FY25 revenue $18.7B / net loss $4.9B / Q1'26 revenue $4.69B / net loss $4.27B = S-1, PitchBook/Morningstar breakdown 05-20; Morningstar fair value ~$780B = Morningstar 2026-05. First-draft soft-number notes retained for the record: $1.75T / $40-80B / 6/12 were all pre-S-1 reported targets at the time.
Sell-side claims and interest structure (researched 2026-06-10): GS lead left + 23-bank syndicate + $800M-$1B fee pool = Yahoo Finance 2026-06; GS roadshow AI revenue model ($3.2B -> $322B / total revenue $474B / EBITDA $352B) = Yahoo / 247wallst / Motley Fool 2026-06-04 (soft, roadshow relayed, no named analyst); MS $3.4T/2040 = Stocktwits / BigGo 2026-06-05 (soft); $20B bridge loan led by GS, refinancing X/xAI 12.5% debt, repayable within 6 months of the IPO = Octus / IFR 2026-05 (hard); 2x oversubscription / ~$150B orders = Seeking Alpha 2026-06-09; Cramer $4T = 247wallst 2026-06-04; ARK $2.5-3.1T = Fortune 2025-06-11 (year-old pre-IPO model); lockup ladder (20% at 2 days post-earnings / ~45% at 135 days / Musk 366 days) = CNBC 2026-05-21. GS/MS proprietary pre-IPO holdings: no public record found.
Subscription-multiple comparison set (verified 2026-06-10): ARM 10-15x = Bloomberg Law / Fortune 2023-09; Aramco 4.65x overall / 2.95x institutional book + break of issue at 3 months = Nasdaq 2019-12 / CNBC 2020-03; Alibaba 14-17x (accounts vary) = CNBC 2014-09; Facebook institutional >5x + -50% in 4 months = CLS Blue Sky / Wikipedia; Reddit 4-5x = Fox Business 2024-03; Rivian multiple undisclosed, -80% to date = Crunchbase / stockanalysis 2026-06-09; CoreWeave downsized, +250% = Bloomberg 2025-03 / CNBC; order-inflation custom = Yahoo Finance 2026-06-09 (strategist quote on the SpaceX deal) + Cornelli & Goldreich, Journal of Finance 2003; fair value per-share conversion uses fully diluted ~13.1B shares ($1.77T / $135) [model].
Technical coordinates (web search, 2024-2026, reliable): WBG ~400 C / 250 C limit, GaN D flip-flop at 160 C, missing p-type GaN, 17% data-center penetration, Intel prototype stage = arXiv 2404.03510 / MIT thesis / TrendForce / Wiley pssa. 838 W/m^2, ~100 t/LEO, FCC acceptance = search recall. Difficulty 1-10 ranking, three-way mutex, feasibility criteria chain, time windows = framework + mechanism reasoning (estimate).