Space Exploration Technologies (SpaceX)
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
Reusable-rocket cost position plus Starlink spectrum-and-scale remain the durability. Cursor's close and Grok 4.6 add a real application layer; they do not deepen the physical or regulatory moat, and they make a software interface newly applicable — intact, not strong.
SpaceX's durability is still physical and regulatory. The two weeks after the Q2 print added a software surface on top; they did not replace the engines underneath:
- Reusable-Launch Cost Position: H1 2026: 78 launches and 1,041 metric tons to orbit, most of it internal Starlink deployment. Q2 added 38 flights (10 customer / 28 internal). No competitor (Blue Origin, Rocket Lab, ULA, China) has matched orbital-class reuse at this cadence, so the cost-per-kg gap is still measured in years, not quarters.
- Spectrum + Orbital Slots: Starlink's FCC spectrum grants and ITU orbital-slot filings are scarce, first-mover, government-allocated assets. Q2 added the EchoStar transfer of 65 MHz of U.S. spectrum plus global mobile-satellite licences, and more than $6B of multi-year Starshield awards. A rival cannot simply out-spend its way past the regulatory queue.
- The Self-Funding Flywheel: SpaceX launches its own constellation at internal cost, so every Falcon flight makes Starlink cheaper to deploy, and Starlink cash flow funds the next tranche and Starship. Vertical integration competitors must buy launch on the open market to compete.
- Application Layer, Not a New Moat: Cursor closed on August 14 as a wholly owned subsidiary (389.3M Class A shares, $60.0B implied equity value). Grok 4.6, trained with Cursor workflow data, shipped two days earlier and launched first inside Cursor and Grok Build. That is a genuine distribution and data loop for SpaceXAI. It is not a substitute for spectrum, slots, or reusable launch — and Cursor's category is already contested (Copilot, Claude Code) with share that has been sliding, so the new surface is intact rather than strong.
Moat Verdict
SpaceX is a net AI beneficiary on the demand side — AI buildout drives launch, satellite connectivity, and, since the xAI combination, an in-house frontier-model and compute franchise that Q2 already printed at $2.56B of revenue and +$1.15B of segment adj. EBITDA. Grok 4.6 and the Cursor close add a coding application layer on that compute; Terafab is a 2028 attempt to own the chip supply underneath it. The core moats (regulatory spectrum/slots, scarce aerospace talent, reusable-launch cost position) are essentially AI-irrelevant and therefore AI-resilient. The new software surface is the exception: learned interfaces is now applicable and only intact, frontier weights commoditise, and Cursor competes on price in a category that is already losing share — so the application layer lifts growth optionality without deepening durability, which is why the moat score does not rise with the close. The honest limitation is that half the moat slate is still N/A, and the software legs that just became applicable are the AI-vulnerable ones. Durable physical/regulatory franchise with a newly attached, contested application layer.
81.1 resilient · 65.0 vulnerable · 80/20 = 77.9 · = 78
Open a moat to read its note.
Upgraded from na on the August 14 Cursor close. Cursor is a complex agent IDE — rules, memories, codebase index, team workflows — that developers invest in. That is now a SpaceX product. Kept at intact, not strong: the category is three years old, GitHub Copilot and Claude Code are substitutes, and Cursor's share of AI-coding spend has already been sliding. The interface is the agent, so this leg is AI-vulnerable.
Launch and Starlink remain procurement/utility relationships. Cursor enterprise rules and memories are real configuration, but they are not years-deep proprietary process logic — an engineering team can re-point an IDE in weeks. Not yet a business-logic lock-in.
the company does not monetise gated access to a public dataset.
Reusable-orbital propulsion, GNC and satellite-manufacturing engineers are the scarcest talent in aerospace, and the Cursor close adds a frontier coding-agent team. The talent shows up in the cost and cadence lead already rated strong under scale economics, so it rates intact here (one fact, one pillar).
Vertical integration is still the bundle: in-house launch deploys Starlink at internal cost, and ground network plus terminal plus connectivity is sold as one stack. Grok 4.6 launching first inside Cursor, trained on Cursor workflow data, is a new software bundle on the same compute (Colossus). Competitors must assemble these pieces on the open market. Not strong: Microsoft (GitHub Copilot + Azure) and Google (Gemini + Cloud) already sell the same shape of bundle.
Falcon flight-and-reuse telemetry and constellation operations data remain unique and compounding. The xAI combination added Colossus and Grok weights; Grok 4.6's model card states supplemental training on anonymised Cursor workflow data — a new flywheel now inside the company. Frontier weights still commoditise and Grok still competes on price, so this leg is AI-vulnerable and does not deepen durability. Kept intact: the data is real, but it improves the product rather than being a directly-monetised, defensible dataset.
FCC spectrum grants, ITU orbital-slot priority, FAA launch licences, and NASA/DoD national-security launch certification are scarce, slow, first-mover-advantaged assets. Q2 added the EchoStar 65 MHz transfer and more than $6B of multi-year Starshield awards. Switching launch providers requires re-certification, and spectrum/slots cannot be out-spent past the regulatory queue. This is the hardest moat to replicate and is AI-irrelevant.
Added users on a Starlink cell still congest shared capacity rather than improve it — the classic per-user effect is absent — but two ecosystem-level effects are present: the self-funding launch↔Starlink flywheel, and the direct-to-cell two-sided network signing carriers (SoftBank, NTT Docomo, Spark NZ in Q2). Cursor adds a developer network (training data from usage → better Grok → better Cursor) that is real and early. Not strong: D2C is early, the per-user Starlink effect never arrives, and Cursor's category share has been sliding.
SpaceX does not sit in a payment or transaction layer.
it is not the authoritative record for any external business function. Graphite (code review) sits inside Cursor; that is not a system of record.
Reusable Falcon launch and in-house Starlink production give the lowest cost per kilogram to orbit and per satellite of any operator: 78 launches and 1,041 metric tons to orbit in H1 2026, a cadence no rival launch provider approaches.
Buyers are enterprises choosing on switching cost, integration and performance, which the other pillars rate. The name carries reputation, not a price premium it could hold on brand alone.
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
Reusable-rocket cost position plus Starlink spectrum-and-scale remain the durability. Cursor's close and Grok 4.6 add a real application layer; they do not deepen the physical or regulatory moat, and they make a software interface newly applicable — intact, not strong.
Growth Score
The two weeks after the first public print changed the AI leg without changing the last reported quarter. Q2 2026 (8-K, August 4) was $7.81B of revenue (+92% YoY vs $4.07B), adj. EBITDA $3.54B, and a GAAP net loss of $(541)M — Connectivity $4.29B (+66% YoY) on 12.0M Starlink subscribers at a flat $66 ARPU, Space $962M (+29% YoY) on 38 launches, and AI $2.56B (+247% YoY) after $14.1B of cloud-service contracted sales, of which $1.6B was recognised in the quarter. AI segment adj. EBITDA flipped to +$1.15B; the GAAP AI operating loss was still $(1.26)B on $15.8B of AI capex. Then: August 6, SpaceX and Tesla sited Terafab in Grimes County, Texas, with a $16.8B first phase (construction this year, phase-one complete 2028; later phases $55–119B). August 12, Grok 4.6 shipped — a coding-and-agents model that matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index, priced at $2 / $6 per million tokens, trained with Cursor. August 14, the Cursor merger closed. The stock closed August 17 at $146.23. Starship Flight 14 is still the next operational test: first orbital attempt and first ship catch, tentatively late August, pending FAA approval.
Valuation Score
At $146.23 (August 17 close, +4.5% on the day, market cap ~$1.93T) SPCX sits just under the $155 base and well above the July 10 low near $145 — the post-IPO range has been $104.83–$225.64. The August 4 print was stronger than the first write-up recorded: revenue +92% YoY, adj. EBITDA $3.54B, AI already EBITDA-positive at the segment level. The two weeks since then added Cursor (389.3M new Class A shares, ~3% incremental supply at the June 30 share count) and a $16.8B Terafab first phase that extends the capex cycle into 2028. Trailing sales on a reconstructed TTM of ~$23B (FY2025 $18.7B − H1 2025 $8.1B + H1 2026 $12.5B) are ~84×; annualising Q2 implies ~62×. Earnings multiples remain meaningless on a $(541)M quarterly GAAP loss and H1 investing cash outflow of $34.5B against $3.5B of operating cash. The market is paying for the launch-and-spectrum position plus AI-compute and Cursor optionality — not current free cash flow.
The Reusability + Spectrum Moat
SpaceX's durability is still physical and regulatory. The two weeks after the Q2 print added a software surface on top; they did not replace the engines underneath:
- Reusable-Launch Cost Position: H1 2026: 78 launches and 1,041 metric tons to orbit, most of it internal Starlink deployment. Q2 added 38 flights (10 customer / 28 internal). No competitor (Blue Origin, Rocket Lab, ULA, China) has matched orbital-class reuse at this cadence, so the cost-per-kg gap is still measured in years, not quarters.
- Spectrum + Orbital Slots: Starlink's FCC spectrum grants and ITU orbital-slot filings are scarce, first-mover, government-allocated assets. Q2 added the EchoStar transfer of 65 MHz of U.S. spectrum plus global mobile-satellite licences, and more than $6B of multi-year Starshield awards. A rival cannot simply out-spend its way past the regulatory queue.
- The Self-Funding Flywheel: SpaceX launches its own constellation at internal cost, so every Falcon flight makes Starlink cheaper to deploy, and Starlink cash flow funds the next tranche and Starship. Vertical integration competitors must buy launch on the open market to compete.
- Application Layer, Not a New Moat: Cursor closed on August 14 as a wholly owned subsidiary (389.3M Class A shares, $60.0B implied equity value). Grok 4.6, trained with Cursor workflow data, shipped two days earlier and launched first inside Cursor and Grok Build. That is a genuine distribution and data loop for SpaceXAI. It is not a substitute for spectrum, slots, or reusable launch — and Cursor's category is already contested (Copilot, Claude Code) with share that has been sliding, so the new surface is intact rather than strong.
Moat Verdict
SpaceX is a net AI beneficiary on the demand side — AI buildout drives launch, satellite connectivity, and, since the xAI combination, an in-house frontier-model and compute franchise that Q2 already printed at $2.56B of revenue and +$1.15B of segment adj. EBITDA. Grok 4.6 and the Cursor close add a coding application layer on that compute; Terafab is a 2028 attempt to own the chip supply underneath it. The core moats (regulatory spectrum/slots, scarce aerospace talent, reusable-launch cost position) are essentially AI-irrelevant and therefore AI-resilient. The new software surface is the exception: learned interfaces is now applicable and only intact, frontier weights commoditise, and Cursor competes on price in a category that is already losing share — so the application layer lifts growth optionality without deepening durability, which is why the moat score does not rise with the close. The honest limitation is that half the moat slate is still N/A, and the software legs that just became applicable are the AI-vulnerable ones. Durable physical/regulatory franchise with a newly attached, contested application layer.
81.1 resilient · 65.0 vulnerable · 80/20 = 77.9 · = 78
Open a moat to read its note.
Upgraded from na on the August 14 Cursor close. Cursor is a complex agent IDE — rules, memories, codebase index, team workflows — that developers invest in. That is now a SpaceX product. Kept at intact, not strong: the category is three years old, GitHub Copilot and Claude Code are substitutes, and Cursor's share of AI-coding spend has already been sliding. The interface is the agent, so this leg is AI-vulnerable.
Launch and Starlink remain procurement/utility relationships. Cursor enterprise rules and memories are real configuration, but they are not years-deep proprietary process logic — an engineering team can re-point an IDE in weeks. Not yet a business-logic lock-in.
the company does not monetise gated access to a public dataset.
Reusable-orbital propulsion, GNC and satellite-manufacturing engineers are the scarcest talent in aerospace, and the Cursor close adds a frontier coding-agent team. The talent shows up in the cost and cadence lead already rated strong under scale economics, so it rates intact here (one fact, one pillar).
Vertical integration is still the bundle: in-house launch deploys Starlink at internal cost, and ground network plus terminal plus connectivity is sold as one stack. Grok 4.6 launching first inside Cursor, trained on Cursor workflow data, is a new software bundle on the same compute (Colossus). Competitors must assemble these pieces on the open market. Not strong: Microsoft (GitHub Copilot + Azure) and Google (Gemini + Cloud) already sell the same shape of bundle.
Falcon flight-and-reuse telemetry and constellation operations data remain unique and compounding. The xAI combination added Colossus and Grok weights; Grok 4.6's model card states supplemental training on anonymised Cursor workflow data — a new flywheel now inside the company. Frontier weights still commoditise and Grok still competes on price, so this leg is AI-vulnerable and does not deepen durability. Kept intact: the data is real, but it improves the product rather than being a directly-monetised, defensible dataset.
FCC spectrum grants, ITU orbital-slot priority, FAA launch licences, and NASA/DoD national-security launch certification are scarce, slow, first-mover-advantaged assets. Q2 added the EchoStar 65 MHz transfer and more than $6B of multi-year Starshield awards. Switching launch providers requires re-certification, and spectrum/slots cannot be out-spent past the regulatory queue. This is the hardest moat to replicate and is AI-irrelevant.
Added users on a Starlink cell still congest shared capacity rather than improve it — the classic per-user effect is absent — but two ecosystem-level effects are present: the self-funding launch↔Starlink flywheel, and the direct-to-cell two-sided network signing carriers (SoftBank, NTT Docomo, Spark NZ in Q2). Cursor adds a developer network (training data from usage → better Grok → better Cursor) that is real and early. Not strong: D2C is early, the per-user Starlink effect never arrives, and Cursor's category share has been sliding.
SpaceX does not sit in a payment or transaction layer.
it is not the authoritative record for any external business function. Graphite (code review) sits inside Cursor; that is not a system of record.
Reusable Falcon launch and in-house Starlink production give the lowest cost per kilogram to orbit and per satellite of any operator: 78 launches and 1,041 metric tons to orbit in H1 2026, a cadence no rival launch provider approaches.
Buyers are enterprises choosing on switching cost, integration and performance, which the other pillars rate. The name carries reputation, not a price premium it could hold on brand alone.
Growth Analysis
Growth Drivers
Key Risk
Starship is still the linchpin of the next constellation generation and the cost curve. Flight 14 — tentatively late August, pending FAA approval — is the first orbital attempt and the first ship-catch. H1 delivered Flight 12 (May) and Flight 13 (July 24); that is progress, not ≥15 launches/yr with reliable reuse. If that cadence is not demonstrated by the end of 2027, V3 Starlink mass deployment and the launch-cost step-down slip. Terafab ($16.8B phase 1, production 2028) and Q2 AI capex of $15.8B are observed spend already in the cash-flow picture, not a second unmaterialised risk.
Score Derivation
91.3 base + 4.0 trajectory + 4 margin − 10 risk = 89
Base 91 (30–40% CAGR, 35% midpoint) + 4 trajectory (Connectivity, Space, and AI all accelerating on the Q2 print) + 4 margin expanding (group adj. EBITDA $3.54B / ~45% vs $1.21B / ~30% a year ago; AI flipped to +$1.15B segment adj. EBITDA) − 10 high keyRisk (Starship cadence still unproven; Flight 14 is the next test) = 89.
Price Scenarios (12–24 Months)
Valuation Analysis
P/E is omitted — SpaceX is GAAP loss-making ($(541)M in Q2 2026; $(4.82)B in H1) and deeply FCF-negative (H1 operating cash $3.47B vs $28.5B of capex). Valuation is anchored on price/sales (~84× reconstructed TTM, ~62× annualised Q2) and EV/EBITDA (EV ~$1.87T after $100B cash and ~$39B of debt, against ~$14B annualised adj. EBITDA ≈ 130×). The premium is paid for reusable-launch cost position, spectrum, the Grok/Cursor application layer, and Starship — not current cash generation. $155 (base) — stock trades a few percent below fair value after the Cursor close and Terafab site decision.
Where We Are vs Targets
Loading live price…
Growth-stock derating meets lock-up supply, Cursor share keeps sliding, and Starship misses again: the market stops paying 60–80× sales for a GAAP-loss, dual-class-controlled company, and the stock re-rates toward ~$1.3T.
- Post-IPO lock-up expiries (2026–27) plus the 389.3M Cursor shares flood the thin float as the Nasdaq-100 forced-buying bid fades
- Starship Flight 14 slips or fails the ship-catch, and cadence stays well below 15/yr through 2027 — delaying V3 Starlink and the launch-cost step-down
- Cursor's AI-coding share keeps eroding against Anthropic and GitHub Copilot, so the $60B application layer does not convert the $14.1B compute book into a durable franchise
Starlink compounds at a flat ARPU, launch leadership holds, Grok 4.6 plus Cursor consolidate as a coding surface, and AI-compute revenue ramps — while Terafab and Colossus keep FCF negative and the stock holds near ~$1.9–2.0T.
- FY2026 lands in the high-$20Bs to low-$30Bs on Connectivity compounding (12M+ subs, enterprise mix lifting revenue faster than ARPU), Space at a $4B run-rate, and AI converting the $14.1B CSA book, with Cursor contributing from mid-August
- Starlink ARPU holds in the mid-$60s; enterprise and Starshield, not a consumer price hike, do the monetisation work — as Q2 already showed
- Grok 4.6 wins coding share on price and Cursor distribution, but the AI segment stays GAAP-negative on depreciation; Starship reaches orbital reuse below 15/yr, so the full cost-curve and FCF turn arrive later than bulls assume
The flywheel inflects on all cylinders: Starship hits cadence, Starlink and Starshield compound, and Grok 4.6 plus Cursor turn the compute book into an application franchise — re-rating the company toward ~$3.2T.
- Starship Flight 14 succeeds (orbital + ship-catch) and the vehicle reaches ≥15 launches/yr with reliable reuse, collapsing $/kg and enabling V3 Starlink mass deployment
- Grok 4.6's frontier coding scores and Cursor's enterprise footprint convert the $14.1B CSA book plus new attach into a durable franchise, and AI-plus-connectivity drives an exit-2026 run-rate toward $50B+
- Starlink crosses ~20M subscribers with enterprise mix lifting effective ARPU, Terafab stays a 2028 story rather than a 2026 cash crisis, and free cash flow turns toward breakeven