← All research中文
01Space Exploration Technologies · NASDAQ: SPCX

SpaceX, taken apart

The question is not whether $151 is cheap or expensive. It is: if you buy at $151, which assumptions are you paying for? Believers buy, sceptics pass, but either way you should know what you are believing.

The approach: one company, three completely different businesses. Take them apart, value each against its own peers, then add them back together.

SOTP · Sum-of-the-Parts: split a company into its separate businesses, value each one on the multiples its own industry peers trade at, then add the pieces up and add net cash to get a value for the whole. It suits companies whose businesses are too different to be captured by a single multiple.
Market cap · 2026-09-10
$2.06T
$151.61 × 13.57B shares
Three segments
Space · Connectivity · AI
The three reportable segments disclosed in the 10-Q
Data basis
Q2 2026 annualised
Latest quarter × 4, no seasonal adjustment

Disclaimer · This material is for personal research and educational discussion only. It is compiled from public information and does not constitute investment advice, an offer, or a recommendation, nor a basis for buying or selling any security. The presenter holds a position in SPCX and may have a conflict of interest. Investing involves risk; make your own judgement and bear the consequences of your own decisions.

02The businesses

Three businesses, eight names

Founded 2002 · HQ Starbase, Texas · 22,000 employees · 2025 revenue $18.7B · Listed 2026-06-12

Space · Launch and spacecraft
Falcon 9Reusable rocket · 165 launches in 2025 · a single booster flown 33 times
DragonCrew and cargo capsule · currently the only operational US crewed spacecraft
StarshipNext-generation, fully reusable · 100 t+ to orbit · V3 first flight 2026-05
Connectivity · Satellite communications
StarlinkSatellite broadband · 12M subscribers · 155+ countries · 10,000+ satellites in orbit, 65% of all active satellites
StarshieldGovernment and defence · communications, reconnaissance and missile-warning constellations · $6B+ in US Space Force contracts
AI · Models, platform, compute
XSocial platform · advertising revenue · merged into xAI 2025-03
xAIGrok frontier models · merged into SpaceX 2026-02 · the core of the AI segment
CursorAI coding tool · all-stock acquisition 2026-08 · $60B
ColossusSelf-built compute cluster · 1.4 GW nameplate · leased out and used to train Grok
03Since the IPO · Price and lock-ups

Priced at $135, a full round trip in three months

Closing price (FMP daily) · grey area is the future · ▲ = lock-up expiry · dashed line = IPO price $135
UnlockSharesBasis
2026-08-06912MDone
2026-08-20319MDone · fell below $135 that day
2026-09-09319MDone
2026-09-1059MDone
2026-12-09Undisclosed180-day lock-up expires · inferred from 6/12
Cursor 389M sharesIssued 8/14 · lock-up terms undisclosed
EchoStar 262M sharesSpectrum deal pending closing
  • Four unlocks so far total 1.61B shares, about 12% of shares outstanding; the float has grown from 640M shares at the IPO.
  • Sell side, 10 analysts: 9 buy / 1 hold / 0 sell. Consensus target $207, range $75–450.
04Last two quarters · Q1 & Q2 2026

Revenue +66% quarter on quarter, from consolidating xAI

xAI was consolidated on 2 February 2026, so Q2 is the first full quarter. Watch four lines: revenue, adjusted EBITDA, capex, free cash flow.

Consolidated ($M)Q1'25Q2'25Q1'26Q2'26Q2 YoYQ2 QoQ
Revenue4,0674,0714,6947,814+92%+66%
Gross margin51.8%49.1%55.3%
R&D1,5573,5143,54845% of revenue
Adjusted EBITDA1,7021,2141,1383,538+191%+211%
Net income−528−1,008−4,276−541
Operating cash flow7271,0472,419
Capex4,1402,82510,10718,369×6.5×1.8
Free cash flow−3,413−9,060−15,950

Adjusted EBITDA = operating income + D&A + stock-based compensation (this method gives 3,536 for Q2 against the disclosed 3,538; Q1 is derived the same way). Q2'25 has only the four items disclosed in the earnings 8-K. Period-end cash and marketable securities $100.0B, interest-bearing debt $39.5B.
Full-year base: 2025 revenue $18.7B (Starlink $11.4B, 61%); Q2 annualised is $31.3B, +67% on full-year 2025.

05Segment structure · 10-Q notes

Three pieces: Space · Connectivity · AI

SpaceConnectivityAIlight = Q2'25
Q2 2026 vs Q2 2025 · $M · Source: Q2 earnings 8-K
Q2'26 ($M)RevenueYoYAdj. EBITDAMarginCapex
Space962+29%−205−21%1,174
Connectivity4,291+66%2,59760.5%1,367
AI2,561+247%1,14644.7%15,828
Total7,814+92%3,53845.3%18,369

Revenue mix 55 / 33 / 12 · Capex mix 7 / 6 / 86

06Segment one · Connectivity (Starlink + Starshield)

12 million subscribers, 60% EBITDA margin

Subscribers
12.0M
6.0M a year ago
ARPU / month
$66
$85 a year ago
Q2 annualised revenue
$17.2B
Consumer 61% · enterprise and government 39%
Q2 annualised EBITDA
$10.4B
Margin 60.5%
PeerMarket capEV/SalesEV/EBITDAProfile
IRDM Iridium$5.0B7.4×15.5×Mature satcom · revenue +5%
VSAT Viasat$10.2B3.3×8.6×Mature satellite broadband
TMUS T-Mobile$188.7B3.3×10.7×Wireless carrier
GSAT Globalstar$10.6B37.6×169×Apple-backed · thin margins
ASTS AST SpaceMobile$25.2B225×Direct-to-cell satellite · near-zero revenue
Valuation range · EV/EBITDA 12× / 20× / 30×
$125B · $208B · $312B
Floor = T-Mobile's multiple; ceiling = a high-growth premium. No peer combines a 60% margin with +66% growth.
07Segment two · AI (xAI + X platform + Cursor + compute)

Already profitable, but eating 86% of capex

Nameplate compute
1.4 GW
Q1 1.0 GW · 0.4 GW a year ago
Cloud contracts
$14.1B
Non-cancellable portion · customer not named
Q2 annualised revenue
$10.2B
Compute and infrastructure 86% · advertising 14%
Q2 capex
$15.8B
$63B annualised
PeerMarket capEV/SalesEV/EBITDAProfile
CRWV CoreWeave$49.9B12.5×37.3×AI compute cloud · revenue +168%
NBIS Nebius$55.0B42.1×59.7×AI compute cloud
ORCL Oracle$449B8.5×17.9×Cloud
META Meta$1,654B7.7×16.4×Advertising · comparable for X

Cursor closed on 8/14, $60B all-stock, not in the Q2 numbers; valued separately at the deal price.

AI revenue · four ways to count itFigureMeaningSource
Recognised revenueQ2 $2.56B → $10.2B annualisedActually booked in Q2: compute $2.19B + advertising $0.37B8-K
Signed non-cancellable contracts$14.1BThe portion of contracts that cannot be cancelled · customer not named8-K
Management run-rate$3.75B / quarter → $15B annualisedQuarterly level once the compute business reaches full capacity · customer not namedProspectus FWP · CFO interview
Anthropic contract, total valueAbout $45B over three years$1.25B per month through 2029-05 · i.e. the row above × 3 yearsReported second-hand in a video · not verified against the filing

Actual Q2 compute revenue was $2.19B, short of $3.75B, because the contract ramps with Colossus II capacity. $14.1B is a hard number; $45B only materialises if fully delivered.

Valuation range · EV/Sales 8× / 20× / 40× + Cursor $60B
$142B · $265B · $470B
EV/Sales is used because EBITDA has only just turned positive. Floor = Oracle / Meta, ceiling = Nebius, base case in between.
08Segment three · Space (launch services + Starship)

Smallest revenue, biggest imagination

Q2 launches
38
46 a year ago · 28 internal / 10 customer
Mass to orbit
485 t
652 t a year ago
Q2 annualised revenue
$3.8B
EBITDA −$0.8B
Backlog
$47.5B
56% to be recognised within a year
PeerMarket capEV/SalesEV/EBITDAProfile
RKLB Rocket Lab$36.6B44.9×negativeLaunch · revenue +38% · EBITDA loss
LMT Lockheed$121.9B1.8×14.1×Defence contractor
NOC Northrop$73.9BDefence / space

Starship flight 12 in May, flight 13 in July. Management's framing: "cut the cost to orbit by more than 99%".

The economics of compute in orbit (as broken down by a former SpaceX executive)
1 GW ≈ 10,000 units of 100 kW ≈ 100 Starship launches. Rideshare pricing today is about $7,000/kg; it needs to fall to $100–200/kg just to break even and $10–20/kg to make sense.
Cooling by radiation (the T⁴ law), radiation handled by shielding: the physics all works. He personally bets on reaching 1 GW by 2029.
Conclusion: "physically feasible, economically not yet." That is exactly why Starship is treated as a separate option.
Valuation range · EV/Sales 10× / 25× / 45× + Starship option $0 / $100B / $300B
$38B · $196B · $473B
Starship has no revenue and no comparable, so it is valued separately as an option; the three tiers are three levels of faith.
09Sum of the parts

Three scenarios: $366B · $730B · $1,316B

ConnectivityAICursorSpaceStarship optionNet cash
$B · Q2 2026 annualised · dashed line = current market cap $2,058B
$BConnectivityAICursorSpaceStarshipNet cashTotalPer share
Bear · 12× / 8× / 10× / $0125826038060366$27
Base · 20× / 20× / 25× / $100B208205609610060730$54
Bull · 30× / 40× / 45× / $300B31241060173300601,316$97

Current price $151.61. On the latest quarter annualised, none of the three scenarios reaches the market cap, which tells you the market is not pricing today's level of business.

10Working backwards · Which year is the market pricing?

The same multiples, applied to the next three years of consensus revenue

Base-case multiples × sell-side consensus revenue → per share · dashed line = current price $151.61 · revenue mix as in Q2
ConsensusRevenueRangeAnalystsPer share
2026E$43B35 – 4816$61
2027E$104B70 – 14016$134
2028E$177B60 – 27917$221
  • The current price sits between 2027E and 2028E, closer to 2027E.
  • The 2028E consensus range is $60B – $279B, a 4.6× spread from low to high.
  • $177B is 5.7 times today's annualised revenue.
11From $31B to $177B

What each segment has to deliver

Connectivity
$17B → $97B
Subscribers from 12M to about 60M at constant ARPU. Verifiable: subscriber counts are disclosed every quarter.
AI
$10B → $58B
Compute from 1.4 GW to about 8 GW. Bought with capital: roughly $10B+ per GW.
Space
$4B → $21B
Starship commercialises and opens new markets. Bought with faith: no revenue today.

$100B of cash lasts about 1.4 years at the Q2 pace ($73B annualised capex); operating cash flow is $2.4B a quarter.

12Risks

The ones you can read straight off the filings

RiskNumberWhat to watch next quarter
Capex vs cash$18.4B / quarter · cash $100BWhether Q3 capex keeps climbing
Starlink ARPU decline$85 → $66Whether it stabilises or keeps trading price for volume
Fewer launches46 → 38 · mass to orbit −26%Q3 launch count, Starship flight 14
Widening Space segment loss−93 → −205Whether Starship R&D keeps consuming launch profits
Unlocks and dilution13.18B → 13.84B shares (pro forma)Around 12/9 · Cursor and EchoStar shares
Cursor, $60B all-stockRevenue undisclosedIts contribution once first consolidated
GovernanceMusk holds about 85% of voting powerMandatory arbitration for shareholder claims, no class actions, derivative suits require a 3% stake (about $45B); three pension funds have already written jointly

The governance terms come from the prospectus, which also acknowledges a "high degree of dependence" on Musk personally.

13Method and data sources

Where every number comes from

Primary sources
· 10-Q for the quarter ended 30 June 2026 (segment revenue, share count, balance sheet, backlog)
· Q2 2026 earnings 8-K (segment adjusted EBITDA, subscribers/ARPU, launch count, compute, segment capex)
· 8-K, 2026-08-14 (Cursor acquisition terms)
· Price: IBKR real-time, 2026-09-10 · daily prices and peers: FMP · sell-side consensus: FMP aggregate · lock-up schedule: as reported by GuruFocus

Secondary sources (a Silicon Valley 101 video's reading of the prospectus; the S-1 was not checked directly)
· Anthropic contract size and term, 2025 full-year revenue and Starlink share, governance terms: "The largest IPO in history: SpaceX is no longer just rockets"
· Economics of compute in orbit, launch-cost range, 1 GW by 2029: "E239 | SpaceX wants to take compute in orbit from science fiction to reality, but does it pay?"

Derived figures
· Q1 2026 segment revenue = six-month figure − Q2
· Q1 adjusted EBITDA = operating income + D&A + stock-based compensation
· Annualised = Q2 × 4
· Share count = Class A + B on the 10-Q cover, 13.18B, + Cursor 391M = 13.57B
· Net cash = cash and marketable securities $100.0B − interest-bearing debt $39.5B
· 180-day lock-up expiry = 6/12 + 180 days

A note on method
· Every multiple is a judgement. The ranges show where peers trade, not what the number should be.
· Starship is valued separately as an option because it has no revenue and no comparable.
· Cursor is carried at the deal price because there is no better information.
· Page 10 assumes the revenue mix (55/33/12) and the Connectivity margin stay constant.
· The sell-side 2028 consensus range is 4.6× wide.

Disclaimer · This material is for personal research and educational discussion only. It is compiled from public information and does not constitute investment advice, an offer, or a recommendation, nor a basis for buying or selling any security. The presenter holds a position in SPCX and may have a conflict of interest. Investing involves risk; make your own judgement and bear the consequences of your own decisions.