Corbanu
Quarter in one view
- SpaceX's second reported quarter post-IPO: revenue $7.8B (+92% y/y), net loss of $541M (improved $467M y/y), adjusted EBITDA $3.5B (+191% y/y).
- Three segments: Space ($962M rev, -$205M adj. EBITDA), Connectivity/Starlink ($4.3B rev, $2.6B adj. EBITDA, $1.7B op income), AI ($2.6B rev, $1.1B adj. EBITDA, first positive segment EBITDA).
- Quarter defined by the capital markets transformation: ~$85.7B IPO net proceeds, $25B inaugural IG notes (5.855% weighted coupon, 11.7yr avg maturity), $100B cash on hand, $47.5B backlog.
- Massive AI compute build-out is now the story: $15.8B of $18.4B Q2 capex went to AI compute; management guided to $100B+ ARR by December 2026 and pulled an internal $1T revenue target from 2031 to 2030.
What management is focused on
- Starship V3 operationalization: 2 successful V3 flights in 90 days; Flight 14 targets first V3 Starlink satellite deployment; ship catch attempt possible on next flight (end of August, pending regulatory approval). Musk: heat shield problem "solved"; goal of ≥1 flight/day within a year.
- AI compute land grab: 1.4GW nameplate at quarter end, target >2GW by year-end, "closer to 10GW than 5GW" of cumulative compute by end-2027; Musk separately floated a 20GW power/cooling target with ~15GW expected. Exclusive NVIDIA build (Vera Rubin). "Starmind" orbital compute satellites launching next year.
- Starlink scale-up: V3 satellites (~10x capability of V2, ~10x launch volume claimed → ~100x bandwidth); service improvement requires ~1,000 V3s on orbit, expected ~Q2 2027.
- Starlink Mobile: next-gen mobile satellites fly next year, service end of 2027, leveraging 65MHz EchoStar spectrum (FCC-approved; $856M paid in Q2 under spectrum credit agreement) plus terrestrial build-out via distributed femtocell concept; Shotwell explicitly sized the US Big 3's ~$600B revenue as the target pool.
- $100B ARR by December 2026 — framed by Musk as achievable "if we basically did nothing," driven mainly by AI cloud services ($6.7B contracted in first weeks of Q3, ramping October).
Key numbers and quarter mechanics
- Revenue $7.8B (+92% y/y from $4.1B). Net loss $541M. Adj. EBITDA $3.5B (+191% from $1.2B).
- Space: $962M rev (+55% q/q, +29% y/y); adj. EBITDA -$205M on higher Starship R&D (+$389M y/y costs). 78 launches and 1,041 tons to orbit in H1.
- Connectivity: $4.3B rev (+32% q/q, +66% y/y); 1.7M net consumer adds (record; vs. 1.4M in Q1); ARPU flat at $66; 167 markets; ~10,200 operational satellites, ~800 Tbps downlink; op income $1.7B (+79% y/y, ~3pts margin expansion); costs +$970M (+58% y/y).
- AI: $2.6B rev (+213% q/q, +247% y/y); $1.6B incremental from new cloud hosting deals (Colossus I/II); adj. EBITDA +$1.1B, first positive quarter; net op loss narrowed to $1.3B; advertising +7% q/q; Grok 4.5 token consumption tripled post-July launch.
- Capex: $18.4B in Q2; CFO guided next two quarters at similar levels (~$37B H2 implied). Claimed <1-year payback on new compute deployments.
- Balance sheet: $100B cash + securities; $47.5B backlog; $25B notes partially repaid $20B bridge.
- Note: Musk said "$100 million ARR in December" in one passage — plainly a slip for the $100B figure repeated elsewhere; treat $100B as the guidance.
Product and launch scorecard
- Starship V3: On plan. Flights 13/14 framing: Flight 13 demonstrated orbital capability and ocean-intact heat shield (ship floating, to be recovered). Flight 14 = first V3 satellite deployment; catch attempt next flight pending FAA approval. Cadence claim (1+/day in a year) is aggressive and unproven.
- Starlink V3 satellites: Passed laser-link speed test on recent Starship flight; operational deployment on upcoming missions; consumer-visible service improvement gated to ~1,000 satellites (~Q2 2027). No revenue contribution yet.
- Starlink in-flight connectivity: Signed American Airlines; activated Southwest, Virgin Atlantic, Iberia, Aer Lingus. Claimed <10% aviation penetration. Anecdote about passengers booking connecting flights for Starlink — colorful, unverifiable.
- Starlink Mobile: Carrier partnerships launched (SoftBank, NTT Docomo, Spark NZ); service currently on ~5MHz via telco partners; EchoStar 65MHz integration later next year. No terrestrial capex number disclosed — Shotwell explicitly declined.
- AI/Grok: Grok 4.5 released July; 4.6 "probably next week," 4.7 in 3–4 weeks, Grok 5 by year-end (trained on all SpaceX data). Cloud deals with Google and Anthropic ramping. Cursor acquisition close to closing, "through almost all regulatory hurdles" — road map deferred.
- Government: >$6B in US contracts won in Q2 (Space Force comms/sensing).
- HLS/Moon: Propellant transfer milestone flagged as critical; Artemis III docking mission next year; uncrewed lunar cargo; "boots on the moon in 2028."
Sell-side read-through
- Analysts pressed hardest on sustainability of AI economics: Anmuth (JPMorgan) asked directly whether premium pricing persists; Musk answered with a supply/demand argument (memory supply +20%/yr vs. demand +200%/yr) and a "$30–$50 per watt" Rubin monetization "guess." No contractual pricing, tenor, or counterparty concentration disclosed for the $6.7B or $1.6B hosting revenue.
- Backlog conversion (Sheridan/Goldman): management gave qualitative stickiness ("never lost an enterprise customer") but no book-to-bill, duration, or conversion schedule for the $47.5B backlog or the $6B government wins.
- Capex trajectory (Godyn/Citi): CFO pinned H2 2026 at Q2-like levels but declined longer-term guidance; no response on terrafab/vertical integration specifics.
- Mobile capital intensity (Maral/Evercore): the ~$100B fourth-carrier question was deflected with the femtocell cost-efficiency argument; Shotwell explicitly refused to discuss capex. This is the largest unanswered capital question on the call.
- Compute split (Hodulik/UBS): Musk estimated ~10% of compute reserved for Grok training over time — useful modeling input but a loose figure.
- Musk's $1T-revenue-by-2030 comment came unprompted in a capex answer; analysts did not challenge it directly.
Management credibility
- Strengths: Segment-level disclosure is detailed (margins, cost deltas, ARPU, megawatts). Connectivity operating leverage (+79% op income on +66% revenue) supports the scale-economics claim. AI segment EBITDA inflection coincides with disclosed hosting revenue, which is checkable. H2 capex guide is specific.
- Concerns:
- Guidance layering: "closer to 10GW than 5GW" compute by 2027 was immediately escalated under Jonas's questioning to a 20GW power target / ~15GW expected — targets shifted within the same answer.
- "$100B ARR by December" was characterized as achievable doing "basically nothing," yet depends on ramps from Google/Anthropic deals, the $6.7B October ramp, and Cursor closing. The December-month annualization basis flatters the figure.
- $1T revenue by 2030 ("non-zero chance" of 2029) is aspirational framing with no supporting bridge; segment mix said to be "similar to what we shared pre-IPO" (not in evidence here).
- "Heat shield problem solved" rests on one flight and an unrecovered vehicle; Musk hedged ("don't want to jinx it") but the claim outruns the data.
- "We have never lost an enterprise customer" and the airline passenger anecdote are unverifiable superlatives.
- Starlink "majority of the world's Internet in less than 10 years" and 1M–10M tons/year to orbit are vision statements, not guidance; the 10x-revenue math assumed flat demand against 100x bandwidth.
- Consistency check: "over 2 gigawatts" year-end compute target is consistent between Musk's remarks and CFO; 1.4GW → 2GW in two quarters implies decelerating adds vs. 1.0→1.4GW in Q2.
What changed versus the prior quarter
No prior-quarter summary was supplied; sequential deltas from the transcript only:
- Revenue mix shifted sharply toward AI: segment revenue +213% q/q to $2.6B (33% of total vs. ~15% implied in Q1).
- Starlink net adds accelerated (1.7M vs. 1.4M in Q1) with ARPU flat at $66.
- Compute capacity +0.4GW q/q to 1.4GW.
- Space segment revenue +55% q/q on launch mix.
- IPO and $25B bond offering completed within the quarter — capital structure is entirely new vs. pre-IPO baseline.
- V3 Starship flew twice in the trailing 90 days, moving the program from development narrative to deployment precursor.
Bull case
- AI hosting is monetizing spare compute at high incremental margins with claimed <1-year paybacks; $6.7B already signed for October ramp suggests demand visibility into 2027; supply/demand imbalance (memory-constrained) supports pricing.
- Connectivity is a self-funding growth engine: $2.6B quarterly EBITDA, record adds, and V3 satellites offering a step-change in capacity per launch cost.
- Enterprise/government optionality is large and underpenetrated (<10% aviation; >$6B Q2 government wins; maritime TAM expansion).
- Starship V3 success would simultaneously cut Starlink deployment cost, enable mobile constellation, and open HLS revenue — multiple shots on goal from one program.
- $100B liquidity plus IG market access removes financing risk for the capex plan.
Bear case
- $18.4B quarterly capex (~$74B annualized, ~86% into AI) is an enormous bet on compute pricing persisting; Musk's own "$30–50/watt" figure is self-described guesswork, and he acknowledged rapid intelligence-per-watt gains that historically deflate compute pricing.
- AI revenue quality unknown: no disclosure of contract tenors, renewal terms, counterparty concentration, or GPU depreciation assumptions behind the "high incremental EBITDA margins."
- $100B ARR and $1T-by-2030 are annualized/projection constructs presented with confidence disproportionate to disclosed backing.
- Execution stack is extreme: Starship catch, V3 satellite ramp, mobile service by end-2027, Cursor integration, and 2GW+ compute all due within ~4 quarters.
- Net loss persists ($541M) despite $3.5B adjusted EBITDA — gap between adjusted figures and GAAP (D&A on the compute fleet, stock comp, interest) bears scrutiny as the fleet grows.
- Starlink ARPU flat and management concedes blended ARPU will decline with geographic mix; consumer growth must carry margin while V3 R&D spends.
Next-quarter watchlist
- Flight 14 outcome: V3 satellite deployment and any ship-catch attempt; FAA approval timing; progress toward 1,000 V3 satellites by ~Q2 2027.
- AI deal disclosure: terms/tenors of the $6.7B October ramp and Google/Anthropic agreements; whether <1-year payback claims hold as deals scale; GW trajectory toward >2GW year-end.
- $100B ARR tracking: any monthly/annualized revenue disclosures; Cursor close and its revenue contribution.
- Capex vs. guide: does Q3 land near ~$18B as CFO indicated; any mobile terrestrial capex emergence post-EchoStar close (spectrum payments continuing — $856M paid in Q2).
- Connectivity economics: net adds vs. 1.7M record; ARPU trend as new markets activate (management teased announcements "as early as today"); enterprise/government revenue growth vs. the 108% y/y mark.
- Space segment EBITDA: whether Starship R&D keeps the segment negative and how launch cadence/mix evolves.
- Language discipline: whether 10GW/15GW/20GW compute framing converges; whether $1T/2030 and "heat shield solved" claims are repeated, qualified, or walked back.