Built by Claude Fable

SPCX | Thematic Exposure: 6/10

Deep dive pages: Overview | Business Model | Financial Trends | Thematic Exposure | Management Quality | Investor Sentiment | Concerns, Catalysts & Risks

Prepared 2026-10-29, revised after the PM Review the same day. Fundamentals from Daloopa (company_id 247912), cited inline as daloopa.com/src/ID. Fiscal year = calendar year. Last print Q3 2026 (2026-10-20, 10-Q filed 2026-10-22); next print about 2027-02-25.

Bottom line. SPCX is #1 in the businesses that are shrinking as a share of revenue and about #6 in the business that is becoming the company. It clears the oligopoly gate easily on launch (about half of all orbital launches worldwide and more than 80% of all mass to orbit since 2023, per the S-1) and is a near-monopoly in satellite broadband. But AI compute became the primary segment in Q3 2026 (52% of revenue) and is roughly 80% of the Q1 2027 guide. There SPCX added only about 7% of new global AI IT capacity in 2026 on a like-for-like basis, about #6 among builders. Its 2027 target implies about 22% (Brooker's simple framing, 10 of 40 GW, is 25%), which would make it a top-2 builder in a competitive field, not an oligopolist. Scored on the forward revenue mix, the primary segment fails the >30% test, so the score is 6, down from 7.

Corporate structure (verified from filings, not assumed): SpaceX acquired X.AI Holdings (xAI, which owned X/Twitter and SpaceXAI Corp.) on 2026-02-02. It was a common control merger, so history is recast (10-Q Q3 2026 p.10). Cursor was acquired around mid-Q3 2026 and contributed $608M of Q3 AI revenue (Q3 call). Reporting segments are Space, Connectivity and AI.

The two oligopoly tables come first: launch share (Lens A) and share of new AI GW (Lens B). Segment mix follows.

Oligopoly table 1 (headline). Lens A: launch market share (count, mass, US, ex-Starlink)

SpaceX launch counts are from Daloopa KPIs (2024: 138, 2025: 170, Q1 26 40, Q2 26 38, Q3 26 35). World totals use McDowell/Wikipedia attempt counts via SpaceQ, Payload, BryceTech, Breaking Defense and Spaceflight Now. The 2026 YTD figures come from Orbital Radar as of 2026-10-29. Counts differ by about 1 to 3% across trackers depending on attempts vs successes.

Mass share for 2020 to 2022 is not disclosed (n.d.): the S-1 states only that SpaceX launched "more than 80% of mass to orbit for the world each year" since 2023. Customer and internal launch splits for 2023 onward are Daloopa KPIs (Falcon only; the remainder of the total is Starship test flights); 2020 to 2022 splits are public Starlink mission tallies and are approximate. Ex-Starlink detail (2024 and 2025 rows are now anchored to the Daloopa customer launch counts):

Customer payloads flown: 282 in 2024 and 312 in 2025, against internal (Starlink) payloads of 1,418 and 1,901. BryceTech says SpaceX launched about 85% of all spacecraft deployed in 2025.

Satellite cross-check: Starlink had 11,100 broadband and mobile satellites in LEO at Q3 2026 (up from 9,600 at Q1). That is about 66 to 69% of all active satellites (McDowell and LookUp indices, about 15,700 to 16,800 active). Amazon Leo has about 400 in orbit.

Read-through. On count, SpaceX is about 50% of world launches, and with CASC (about 21%) two players hold about 70%. On mass, SpaceX is about 85 to 90% alone. The lens clears the >30% gate by a wide margin. Two caveats. (1) Excluding Starlink, SpaceX is only about 23% of world launches, behind China in aggregate. Much of the count share is captive demand. (2) Falcon cadence fell in 2026 (113 launches in 9M vs 128 in 9M 2025; mass 1,517 t vs 1,627 t) as the fleet shifts to Starship. Share by count dipped to about 44 to 49% YTD. Mass share, the economically relevant measure, is unchanged at about 90%.

Oligopoly table 2 (headline). Lens B: share of new AI compute by GW added

Inputs. SPCX discloses "nameplate compute draw" (end-2024 0.3 GW, end-2025 0.8, Q1 26 1.0, Q2 26 1.4, Q3 26 1.7). The Q3 call adds about 2.2 GW at end-October, 2.5 to 3.0 GW at end-2026, and "highly confident" of +10 GW in 2027 (13 GW total). Management also cited a 20 GW power-and-cooling plan with a "small but non-zero" chance of over 20 GW of compute. One quarter earlier (Q2 call), the end-2027 target was "closer to 10 than 5" GW total. That is roughly +5 to 7 GW in 2027, so the 2027 target rose about 50% in one quarter. No orbital datacenter capacity exists yet. Starmind satellites start flying in 2027.

Basis conversion. The house file (research/ai-economics/debate/gw_added.md) treats xAI figures as facility-plus-margin, about 1.4x IT (300 MW per 110k GB300). Cross-check: 200k GB300 per month (Q3 call) x 2.7 kW per GPU on xAI's basis = 0.55 GW per month. That matches the observed +0.5 GW nameplate in October (1.7 to 2.2). On the house IT basis of 1.95 kW per GPU, the same chips are 0.39 GW. So SPCX nameplate / 1.4 = AI IT GW. Denominators are the house view (research/ai-economics/debate/gw_added.md), energized AI-only IT GW: 2025 11 (9.5 to 12.5), 2026 19 (15 to 23), 2027 32 (24 to 43). The US is my derived estimate at about 70% of global AI IT adds (2025 about 7.5, 2026 about 13, 2027 about 22). The US 2027 high case is SemiAnalysis +38 GW all-DC x 85% AI = about 32.

Brooker's simple framing (stated SPCX GW over a headline "GW added" number, mixing bases):

Feasibility check (my analysis). On target, 7.1 GW IT in 2027 at the house's about $37B per GW for NVIDIA 2027 systems is about $260B of accelerators. That is roughly 37 to 50% of the house's 14 to 19 GW of NVIDIA output for CY2027. Hitting the target requires SPCX to take a share of NVIDIA supply no single buyer has held. A chip-feasible case of about 4 to 5 GW IT gives about 12 to 16% global share. Delivered reality so far is about 5% of 2026 global adds (0.64 GW IT energized through Q3).

Competitor context for 2026 adds (house file, mostly facility basis, total about 26 GW): AWS about 6, Microsoft about 4.3, Google about 3.5 (est), Oracle about 2.9, Meta about 2.5 (est), xAI/SPCX about 2.0, China AI about 4, CoreWeave 0.85. SPCX ranks about #6 in 2026. The top 3 hold about 53%, so this is not a market where 3 players hold over 70%. On its 2027 target, SPCX would plausibly be the #1 or #2 single builder, but at about 22% of global IT adds in a field of 5 to 7 large builders.

Read-through. The AI GW lens fails the gate on energized reality: 3% (2025) and about 7% (2026). It passes only if you use Brooker's mixed-basis framing or US-only share on the 2027 target (25 to 32%), or the stretch case. Those are company targets that rose 50% in one quarter and look chip-constrained against the house view.

Segment revenue mix (Daloopa)

Concentration: Customer B was 34.5% of Q3 2026 revenue (about $5.0B, consistent with the AI cloud services ramp). Customer A was 16.6%. The company guided Q1 2027 revenue to $33B to $37B, mostly from AI compute (Q3 call). Holding Connectivity near $6B and Space near $1B for the quarter (my estimates), the AI segment (which includes X advertising) is about $28B of the $35B midpoint, roughly 80% of revenue. FMP quarterly consensus for Q1 2027 is $32.0B (9 analysts), below the guide's low end.

Required segment table

Key competitors. Launch: CASC and Chinese commercial providers (LandSpace, Galactic Energy and others), Rocket Lab, ULA, Arianespace, Blue Origin, Roscosmos, ISRO, MHI. Connectivity: Amazon Leo (about 400 satellites, beta service), Eutelsat OneWeb, Viasat, Hughes, China's Guowang and Qianfan, plus terrestrial fiber and 5G in dense markets. AI compute: AWS, Microsoft, Google, Meta, Oracle, CoreWeave, OpenAI/Stargate and Chinese hyperscalers. Frontier models: OpenAI, Anthropic, Google, Meta.

Oligopoly gate questions

1. Competitors with >15% share, per segment. Launch by count: two (SpaceX about 50%, CASC about 21%). Launch by mass: one (SpaceX about 85 to 90%). Satellite broadband: one (Starlink). New AI compute GW: two or three at most (AWS about 23% and Microsoft about 17% of 2026 facility adds; SPCX about 7%). Five or more meaningful builders compete, so the market is competitive, though not technically fragmented. Frontier models: three leaders (OpenAI, Google, Anthropic) above 15%; Grok is below. None of the moat segments is fragmented.
2. Could a customer replace SPCX within 12 months? Launch: no. No other provider has the cadence or cost. Amazon had to buy Falcon launches for Leo, and Ariane 6, Vulcan and New Glenn fly a handful a year each. Starlink in rural, maritime, aviation and government markets: no. Amazon Leo is in beta with about 400 satellites, and Starlink is contracted on about 12,000 aircraft, roughly a quarter of the fleet (Q3 call). Starlink in dense urban markets: yes (fiber, 5G). AI compute: yes. Contracts are deliberately short-term at spot pricing ("rent at effectively the spot market rate", Q3 call). Customer B (34.5% of Q3 revenue) could shift to other clouds at renewal.
3. Price setter or taker? Launch: setter. Falcon pricing has risen with no share loss. Starlink: setter. ARPU fell from $91 (FY24) to $69 (Q3 26), but that was a deliberate mix and emerging-market strategy, not competitive pressure. AI compute: taker of a tight spot market. Pricing is "slightly above the high end of the $30 to $50 per watt range" today, which is favorable but set by scarcity, not by SPCX.

Gate verdict: PASS. SPCX has over 30% share in launch (about 50% by count, about 85 to 90% by mass) and in satellite broadband (about two-thirds of active satellites, over 80% of satellite broadband subscribers). Both are meaningful segments, 48% of Q3 2026 revenue combined, but only about 20% of the Q1 2027 guide. The AI GW lens does not independently pass on energized data.

Why the position is durable, who buys, what could replace it

• Durability. Reusability plus internal demand. Starlink buys most Falcon flights, which funds cadence no competitor can match. Starship (first orbital flight and V3 deployment on Flight 14, a tower catch of both stages targeted for Flight 15) extends the cost lead by an order of magnitude. Flight 14 carried 26 Tbps, about 10 Falcon 9 launches' worth, at roughly half load. Starlink's 11,100-satellite constellation, spectrum (newly acquired low-band, FCC approval for 15,000 Mobile V2 satellites) and kit scale form the second moat. The AI claim of durability (in-house turbine blade casting, giving power 12 to 18 months faster) is plausible but unproven.
• Buyers. Launch: NASA, the DoD/NRO (6 NSSL missions in Q3, 18 new NSSL awards), commercial operators, and SpaceX itself. Connectivity: about 13.6M consumers (Daloopa), airlines, maritime, enterprises, and the US government through Starshield. AI: hyperscalers and labs renting compute (about 50% external by design), Grok and Cursor users, and X advertisers.
• What could replace it. Launch: China's state and commercial reusables (LandSpace landed an orbital booster in August 2026), but they are walled off from Western payloads. New Glenn and Neutron are years behind on cadence. Connectivity: Amazon Leo for enterprise and government, Chinese constellations outside the West, and terrestrial networks in cities. AI: anyone with chips and power. The AI segment is a commodity capacity business today, and its moat (speed of energizing capacity) erodes as the grid and turbine bottlenecks ease.

Scoring

• Rubric 10 requires the largest segment to have >50% share. The largest segment is AI (52% of Q3 revenue, about 80% of the Q1 2027 guide), with about 7% share. Fails.
• Rubric 7 to 8 requires >30% share in the primary segment. Scored on the business being bought, the forward revenue mix, the primary segment is AI, where SPCX is about #6 with single-digit energized share. Fails. (The first draft reached 7 by switching to a 9M basis where Connectivity was 47%; Q3 and every forward period say otherwise, so that basis was dropped.)
• Rubric 5 to 6: the oligopoly gate passes on secondary segments that are still large and dominant (launch, about 50% of world launches and >80% of mass; Starlink, about two-thirds of active satellites), the themes are growing fast (AI GW +68 to 73%, Connectivity +70% YoY), and the AI segment is profitable (65.5% segment EBITDA margin in Q3). That earns the top of this band.
• Why not lower: the launch and Starlink moats feed the AI business (power, speed to deploy, orbital compute optionality), and on the 2027 target SPCX would be a top-2 AI builder. Why not higher: under the principle "don't settle for #2, worse don't look at the 3rd", a #6 position in the primary segment cannot score 7. Score 6.

Assessment. SpaceX owns two of the clearest oligopolies in public markets. In launch it holds about 50% of world orbital launches, more than 80% of mass to orbit (about 90% per the Q3 call) and about 90% of US launches. In satellite broadband it holds about two-thirds of all active satellites and an estimated 80%+ of subscribers. Both sit on growing themes, both come with real pricing power, and both clear the hard gate with room to spare. But the xAI merger changed what the company actually is. AI compute was 52% of Q3 2026 revenue and is about 80% of the Q1 2027 guide. There, SPCX added about 3% (2025) and about 7% (2026) of new global AI IT capacity on a basis-consistent measure, about #6 among builders. Its 2027 target of +10 GW nameplate (about 7 GW IT) would be about 22% of global and about 32% of US adds; Brooker's simple framing (10 of 40 GW) is 25%. That would make it a top-two builder, but in a competitive market where it rents at spot and could be replaced by any well-funded cloud, and the target rose 50% in a quarter and needs an unprecedented share of NVIDIA supply. Stated plainly: SPCX is #1 in the businesses shrinking as a share of revenue and about #6 in the business becoming the company. Score 6/10 (was 7): gate passed on launch and Starlink, scored on the forward mix where the primary segment lacks a moat.

Notes: Daloopa expert call transcripts are not enabled for this firm, so no BlueHeron color was available. World launch counts, satellite share and launch/satcom TAM come from web sources linked above. Starlink mission counts excluding Starlink, US launch totals for 2021 to 2023, and the 70% US share of AI GW adds are approximations and are labeled as such.

Data sourced from Daloopa.