SPCX Spot and Perp Total Returns

Quarter in one view

• Q3 FY2026: revenue $14.6B (+161% y/y from $5.6B), first GAAP profit — net income $4.4B (vs. -$1.1B y/y), adjusted EBITDA $8.8B (+306% y/y). Capex $19.6B; backlog $63B ($17.5B deferred revenue; ~$42B expected recognized within 12 months).
• Segment mix: Space $1.8B (+84% q/q, +6% y/y); Connectivity $5.2B (+69% y/y, +22% q/q); SI $7.6B (+$5B q/q, +$6.7B y/y) — SI is now >50% of revenue.
• Quarter defined by: Flight 14 reaching orbit and deploying operational V3 Starlink satellites; Cursor acquisition closed mid-quarter ($60B equity value, $608M in-quarter revenue, $23B retention SBC over ~3 years); Q4 capex guided to $35–40B (roughly double the prior run rate); first explicit forward revenue guide — Q1 2027 revenue $33–37B.
• Compute targets escalated again: 1.7GW at quarter end, ~2.2GW by end of October, 2.5–3GW by year-end, "highly confident" of +10GW incremental in 2027 (13GW total), with a 20GW incremental power/cooling plan and a "non-zero" chance of >20GW compute online next year. Musk floated $500B–$1T ARR by end of 2027.

What management is focused on

• Full Starship reusability: Flight 15 next month will attempt catch of both booster and ship — Musk: "more likely than not" to succeed; FAA license expected to include return-to-launch-site and tower catch for both stages. Refly of second stage targeted early next year; on-orbit refueling attempt early next year; weekly cadence in H1 next year, twice-weekly+ by end of next year.
• Compute build-out as the core business: Musk spending ~4 days/week in Memphis on Colossus II; ~200,000 GB300s/month coming online for the rest of the year, shifting to Vera Rubin next year; 440,000-GPU Vera Rubin training cluster targeted online by end of Q2 (2027). Power/cooling is the stated constraint; mitigation includes in-house casting of turbine blades/vanes (claimed 12–18 month advantage) and a proprietary 20GW power plan Musk declined to detail.
• TerraFab: first disclosure of concrete chip-fab milestones — test chip ~May/June 2027 (Austin research fab), volume production aspiration Q3 2028 at Givens Creek; EUV equipment deliveries being accelerated by ASML and suppliers (implicitly confirming orders, though not stated directly).
• Starlink Mobile as a US carrier: FCC approval for 15,000 Mobile V2 satellites; new low-band spectrum acquisition announced; Mobile V2 constellation flies through next year, service early 2028 (slipped from prior "end of 2027" framing). Musk claimed a "clear path to several hundred billion dollars a year" in connectivity revenue but deferred the detail.
• Grokbot enterprise push: launched August; 800,000 paid weekly users (doubled in a month); wide enterprise deployment "within a few weeks"; bundled plan combining Grok, Grokbot, Cursor, and X Premium.
• Tesla merger: both companies have independent special committees; no agreement on terms; requires 50% of all Tesla shares. Musk supports it but stressed it is not a certainty.

Key numbers and quarter mechanics

• Revenue $14.6B (+161% y/y); net income $4.4B; adjusted EBITDA $8.8B (+306% y/y).
• Space: $1.8B (+84% q/q, +6% y/y) on favorable customer mix; 35 launches (21 Starlink F9, 2 Starship, 1 Falcon Heavy carrying NASA's Roman Space Telescope, 12 customer missions incl. 6 NSSL). Won 3 Dragon human-spaceflight missions (through 2028), Artemis III Starlink comms contract, 18 new NSSL missions, a Starship lunar payload delivery contract, 3 Starfall missions.
• Connectivity: $5.2B (+69% y/y, +22% q/q); 13.6M consumer subs (+1.6M, vs. 1.7M record in Q2 — slight deceleration); ARPU +5% q/q to $69 (mix-driven); 174 markets (added Vietnam, Uganda); enterprise/government +91% y/y, +18% q/q; gross margin 55% including ~3pts one-time tariff-refund benefit; ex-one-time EBITDA margin +~2pts y/y. Aviation: ~12,000 planes contracted (~25% of available fleet), 3 of 4 major US carriers signed; production quadrupled YTD, to double again by year-end. Starlink Mobile: 22M monthly users (+37% q/q).
• SI: $7.6B revenue (+$5B q/q); $7.0B from SI solutions/infrastructure, of which $608M from Cursor (closed mid-quarter); ex-Cursor SI solutions/infrastructure +192% q/q. Segment adj. EBITDA $5.0B (+$3.8B q/q); segment operating income $431M, or $1.6B excluding $1.2B Cursor-related SBC. Advertising +19% y/y. Grok Enterprise customers >$1M run-rate (incl. Cursor, pro forma) +50% q/q.
• Compute: 1.7GW nameplate at quarter end; ~2.2GW by end of October; 2.5–3GW year-end target.
• Capex: $19.6B in Q3; Q4 guided to $35–40B — a step-change, not flagged at this magnitude last quarter.
• Backlog: $63B (vs. $47.5B in Q2); $17.5B deferred revenue; ~$42B to be recognized within 12 months.
• Guidance: $100B ARR by December (annualized December month) reaffirmed; Q1 2027 revenue guided to $33–37B; Q4 gross margins guided back toward Q2 levels (i.e., Q3 margin flattered by one-timers and mix).
• Pricing: compute pricing "slightly above the high end" of the prior $30–50/watt range; paybacks still <1 year, with Musk musing about 6-month paybacks next year.

Product and launch scorecard

• Starship V3: Major milestone delivered — Flight 14 reached orbit and deployed operational V3 satellites (26 Tbps from a ~half-loaded vehicle, claimed = 10 Falcon 9 launches). Flight 15 dual-catch attempt next month is the next gate; Musk's "more likely than not" framing is appropriately hedged. Refly of ship and on-orbit refueling both due early next year — stacked execution risk.
• V3 Starlink satellites: Now operational on orbit. Consumer-noticeable improvement revised to "a few hundred" satellites by mid-next-year — materially earlier and lower than the prior ~1,000 satellites / ~Q2 2027 framing. Each Starship launch ramping toward >20x Falcon 9 downlink capacity.
• Starlink Mobile: 22M monthly users (+37% q/q); new carrier markets (Uganda, DRC, Kazakhstan, Costa Rica); FCC approval for 15,000 V2 sats; low-band spectrum deal announced. Service timing now "early 2028" (prior: end of 2027) — a slip. S-band to carry majority of traffic, low-band for building penetration. Still no terrestrial capex disclosure.
• Grokbot: 800K paid weekly users, doubling monthly; enterprise wide deployment within weeks; per-user usage claimed to grow over time. Early but real adoption metrics — the most concrete consumer-AI traction disclosed to date.
• Cursor: Closed mid-quarter; $608M revenue contribution in ~half a quarter; $23B of the $60B price allocated to retention SBC over ~3 years (a durable GAAP drag); integration "progressing well."
• TerraFab: New program with dates: test chip May/June 2027, volume Q3 2028 aspiration — Musk himself flagged "insanely fast time frames" and big error bars. Long-term ambition: "a terawatt of SI chips per year."
• Louisiana spaceport: Second land purchase; internal aspiration first launch 2028, "highly likely" at scale by 2029; vision of 20+ towers and millions of tons/year.
• StarMind / Mobile V2 / orbital compute: satellites begin flying 2027 (per Shotwell).

Sell-side read-through

• Only five analyst questions; several key topics went unpressed.
• Jonas (Morgan Stanley) opened with TerraFab — extracted the first fab timeline disclosure (test chip mid-2027, volume Q3 2028) and implicit confirmation of EUV orders via "appreciative of ASML... accelerating equipment deliveries." Musk declined process-detail questions ("not give away the shop").
• Anmuth (JPMorgan) pressed the 13GW confidence and funding. Musk's answer: ~3GW nearly done, deployment-rate transparency, turbine blade/vane casting as a 12–18 month moat, and a deliberately undisclosed 20GW power plan. Bret's funding answer was generic — operating cash flow, IG debt pools, vendor/customer partnerships — no quantum, no structure, no committed facilities disclosed despite capex nearly doubling.
• Yu (Deutsche Bank) asked the Tesla-merger question; Musk confirmed special committees exist, no terms agreed, 50%-of-all-shares hurdle on the Tesla side. Notably, this is now a live, disclosed process — new information.
• Mural (Evercore) asked satellite/terrestrial traffic mix for Starlink Mobile; got an 85–93% Wi-Fi framing and S-band-majority/low-band-coverage split, but the capital-intensity question from last quarter remains unanswered — Musk deferred the "several hundred billion" connectivity case to a future forum.
• Sheridan (Goldman) asked internal-vs-external compute allocation. Key disclosure: SpaceX is deliberately contracting compute short-term/spot rather than long-term, betting $/watt rises as intelligence-per-watt improves; claims renewals are coming at higher $/watt and H100s rent above original prices. This is a strategic choice that increases upside and cyclicality — and it contradicts the "contracted visibility" framing used to support the $100B ARR guide.
• Not asked: no questions on the $35–40B Q4 capex step-up, the $23B retention SBC, backlog composition/duration, counterparty concentration, or the $500B–$1T ARR-by-2027 remark. The investor Q&A (V3 threshold, Louisiana timing) was soft.

Management credibility

• Delivered this quarter: Flight 14 orbit + V3 deployment (as promised); Cursor closed (as promised); ~2.2GW October compute on track per CFO; $100B ARR guide reaffirmed with a defined basis; first GAAP profit; backlog +$15.5B q/q.
• Helpful specificity: CFO gave ex-one-time margin bridge (55% GM incl. ~3pts tariff refunds), ex-SBC segment operating income ($1.6B), Cursor revenue contribution, deferred revenue, and 12-month backlog conversion (~$42B of $63B). Q4 gross-margin guide-down is candid.
• Concerns:
• Target escalation continues: compute framing moved from "closer to 10GW than 5GW by end-2027" (Q2) to "highly confident of 13GW total, non-zero chance of >20GW" (Q3); ARR framing moved from $100B by Dec 2026 to "$500B–$1T by end of 2027" delivered offhandedly ("I don't want to give Bret a heart attack"). Each quarter's ceiling becomes the next quarter's floor.
• Internal tension on revenue quality: the $100B ARR and backlog narrative implies contracted visibility, yet Musk disclosed a deliberate short-term/spot pricing strategy. Both can be true, but the risk profile is different from what "backlog" and "ARR" language suggests.
• Slipped items: Starlink Mobile service moved from end-2027 to early 2028; V3 consumer-impact threshold quietly revised from ~1,000 satellites to "a few hundred" (favorable direction, but the prior figure was presented with confidence).
• Unverifiable superlatives persist: "most advanced casting capability on Earth," "hounded relentlessly by every hyperscaler," cheese-shop anecdote, airline load-factor claims, "fastest growing product I've ever seen."
• Funding opacity: capex is guided to roughly double q/q with no committed financing detail beyond "pools of debt capital we are talking to."
• TerraFab timelines self-flagged as "insanely fast" — credit for the caveat, but volume production in ~2 years from a standing start is an aspiration, not a plan of record.

What changed versus the prior quarter

• Profitability inflection: net income +$4.4B vs. -$541M net loss in Q2; adjusted EBITDA $8.8B vs. $3.5B. SI segment op income positive even including $1.2B Cursor SBC.
• SI became the majority of revenue: $7.6B (52% of total) vs. $2.6B (33%) in Q2; Cursor consolidated from mid-quarter.
• Capex guide stepped up sharply: Q4 guided $35–40B vs. Q2's "next two quarters at ~$18B levels" — Q3 came in at $19.6B (roughly on guide), but Q4 is ~2x the prior run-rate framing.
• Backlog: $63B vs. $47.5B; first disclosure of deferred revenue ($17.5B) and 12-month conversion (~$42B).
• Compute targets raised again: year-end 2.5–3GW (vs. >2GW prior); 2027 framing now 13GW base case with 20GW upside (vs. "closer to 10 than 5" prior).
• Starship progressed from test to operations: V3 satellites deployed on Flight 14; dual-catch attempt and FAA license for tower catch expected on Flight 15.
• Starlink Mobile timing slipped to early 2028; V3 consumer-impact threshold lowered to a few hundred satellites by mid-2027.
• New disclosures: TerraFab program with dates; Tesla merger special committees confirmed; Louisiana spaceport expansion; Q1 2027 revenue guide ($33–37B) — the first forward quarterly revenue guide.
• Consumer adds decelerated slightly (1.6M vs. 1.7M record) while ARPU rose 5% to $69 (prior: flat at $66) — mix shift toward developed markets.

Bull case

• The AI compute thesis is converting to GAAP results: $5B segment EBITDA, $1.6B ex-SBC segment operating income, pricing above the prior $30–50/watt range, claimed renewals at higher $/watt, and <1-year paybacks — with demand described as several times available supply.
• Backlog $63B with ~$42B converting within 12 months plus a $33–37B Q1 guide gives unusual near-term revenue visibility; $100B ARR by December looks attainable on the disclosed ramp (2.2GW already online in October).
• Flight 14 de-risked the Starship→V3 deployment loop; a few hundred V3s by mid-2027 could lift Starlink capacity and unlock high-demand-market sales well before the prior ~1,000-satellite framing implied.
• Connectivity shows operating leverage (gross profit doubled y/y; ARPU up; aviation contracted on ~25% of global fleet with production quadrupling) while enterprise/government grows 91% y/y toward surpassing consumer.
• Vertical-integration moats are compounding: turbine casting (12–18 month power advantage), in-house chips (TerraFab), launch (Starship), and spectrum (EchoStar + new low-band) — each reinforcing the others' economics.
• IG balance sheet, $63B backlog, and short-payback assets give multiple funding levers for the $35–40B quarterly capex pace.

Bear case

• Capex is now guided to $35–40B in a single quarter (~$150B+ annualized) with no committed financing disclosed; the plan depends on compute pricing staying at or above ~$50/watt while management deliberately keeps contracts short-term — maximum exposure to any pricing reversal.
• The short-term-contract strategy cuts both ways: if intelligence-per-watt gains deflate $/watt (as Musk himself argued happens with better models), renewals reprice down and the payback math inverts on a fleet depreciating over ~3 years.
• Earnings quality needs scrutiny: Q3 gross margin included ~3pts of one-time tariff refunds; SI operating income required adding back $1.2B of Cursor SBC, and $23B of retention SBC will hit GAAP over the next ~3 years; Q4 margins guided down.
• Guidance inflation is now a pattern: $500B–$1T ARR by end-2027 was delivered casually with no bridge; 13GW/20GW/23GW figures shifted within a single answer; TerraFab volume production in 2028 is self-described as insanely fast.
• Execution stack is extreme and correlated: Flight 15 dual catch, ship refly, on-orbit refueling, weekly cadence, 440K-GPU training cluster, 10GW incremental compute, Mobile V2 constellation, and a Tesla merger process — all within ~4 quarters, all sharing the same management bandwidth.
• Starlink Mobile slipped to early 2028 and the terrestrial capex question remains unanswered for a second consecutive quarter; consumer adds decelerated and Q4 activations depend on holiday retail channel execution.
• Tesla merger introduces governance overhang: a conflicted-controller transaction requiring 50% of all Tesla shares, with terms not yet agreed.

Next-quarter watchlist

• Flight 15: dual-catch outcome, FAA license scope, and whether ship refly and on-orbit refueling hold "early next year" timing; cadence evidence toward weekly flights in H1 2027.
• Q4 capex vs. the $35–40B guide and any disclosure of committed financing (debt facilities, vendor/customer partnerships) behind the 2027 build.
• $100B ARR confirmation: December annualized revenue disclosure; composition between contracted backlog conversion and spot/short-term compute revenue; renewal pricing data points.
• Compute trajectory: does year-end land at 2.5–3GW; monthly GPU activation pace (~200K/month) sustained; progress on the 440K Vera Rubin training cluster by end of Q2 2027; any detail on the proprietary 20GW power plan.
• Q1 2027 revenue vs. the $33–37B guide — the first test of forward guidance credibility.
• SI revenue quality: contract tenors, counterparty concentration, Cursor revenue run-rate beyond the $608M partial quarter, and the SBC drag trajectory.
• Connectivity: Q4 consumer activations vs. the retail-stocking setup; ARPU trend after the 5% mix-driven bump; aviation activation pace into 2027; V3 satellite count vs. the "few hundred by mid-2027" threshold.
• Starlink Mobile: low-band spectrum deal closing, Mobile V2 launch cadence, and any terrestrial capex disclosure (asked twice, unanswered twice).
• Tesla merger: special-committee progress, terms, and any financing/interdependency implications for SpaceX capex.
• Language discipline: whether $500B–$1T ARR-by-2027, 20GW+ compute, and TerraFab 2028 volume dates are repeated, qualified, or walked back; whether the "several hundred billion" connectivity presentation materializes.

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.