The Operating Loss Almost Disappeared and Nobody Noticed: Upgrading SpaceX Into the Unlock

Results vs. Consensus

This is SpaceX's first reported quarter as a public company. The IPO priced at $135 on June 11 and the shares began trading June 12, which means the June-quarter results span nineteen days of public-company life and seventy-two of private. Sell-side coverage is roughly seven weeks old and the consensus dispersion reflects it: published EPS estimates ranged from a loss of $1.26 to a profit of $0.33. The point estimates below are indicative rather than tight, and we have weighted our read of the quarter accordingly.

Q2 2026 Scorecard

Every revenue line beat, and the one line that missed is the one the market traded on. That asymmetry is the entire explanation for a session that closed up 9.4% into the print and then gave it all back after hours.

Year-over-Year Comparison

Sequential Comparison

Revenue Assessment

Ninety-two percent growth at a $31B annualised run-rate is the headline, but the composition is what changes the investment case. A year ago this was a satellite-broadband company with a launch business attached and a loss-making AI subsidiary. This quarter, Connectivity was 54.9% of revenue, AI was 32.8%, and Space was 12.3%. The AI segment went from 18.1% of revenue to nearly a third in four quarters, and it did so through a single line item: AI solutions and infrastructure revenue of $2,194M against $311M a year ago.

The sequential shape is more informative than the annual. AI revenue rose 213% quarter-on-quarter off a Q1 base of $818M, and management disclosed that $1,600M of that was the initial ramp from Cloud Services Agreements signed during the quarter. That is a genuine step-function, not a trend, and it means the Q2 exit rate rather than the Q2 average is the right anchor for Q3. It also means the durability question is not academic: this is the first quarter in which a third of consolidated revenue sat on contracts the company had only just signed.

Margin Assessment

Consolidated adjusted EBITDA margin of 45.3% against 29.8% a year ago is the cleanest evidence yet that the combined entity has real operating leverage. The mechanism is visible segment by segment. Connectivity expanded operating margin 293bp to 38.6% while growing revenue 66%. AI went from a negative 206.8% operating margin to negative 49.1%. Only Space deteriorated, and it did so deliberately, absorbing a 55% increase in Starship research and development.

The caution is the gap between adjusted EBITDA and operating income, which widened from $2,184M to $3,681M year on year. That gap is depreciation and share-based compensation, and both are compounding: D&A rose 86.6% and SBC rose 79.5%. With $18.4B of capex landing in a single quarter, most of it on compute assets that depreciate over a handful of years, the D&A line is going to keep climbing steeply. Adjusted EBITDA will continue to look better than operating income for as long as the build lasts, and investors who anchor on the former will consistently overestimate the economics.

EPS Assessment

The reported $(0.09) beat a roughly $(0.25) consensus, but the beat is mostly the operating improvement showing up below the line rather than any tax or share-count artefact. The bridge from a $(143)M operating loss to a $(541)M net loss is $629M of interest expense less $340M of interest income, less $86M of other expense and $23M of tax. Of that interest expense, $327M was owed to a related party, which we address separately below.

What the per-share figure conceals is the half-year picture. H1 net loss was $(4,817)M, of which $(4,276)M landed in Q1, driven by $(1,876)M of other expense that the 10-Q defines as gains and losses on digital assets, foreign currency and debt extinguishment. Net loss attributable to shareholders for the half was $(5,488)M, $671M worse than the net loss line, reflecting pre-IPO preferred mechanics that will not recur. Q2 is the first clean quarter in this company's reported history, and it is the one investors should model from.

Segment Performance

Connectivity: The Business That Pays For Everything

Connectivity produced $1,656M of operating income, which is $1,799M more than the consolidated company earned. Every dollar of Space and AI investment is funded by this segment plus the balance sheet. Its margin expansion, 293bp of operating margin on 66% revenue growth, is the single most important number in the report for anyone underwriting a path to consolidated profitability.

The mix shift underneath is the part the market has not fully priced. Enterprise and government revenue grew 108% year on year and 63% sequentially, taking it from a third of the segment to 42.1%. That is a higher-ARPU, stickier, contract-backed revenue pool replacing consumer growth at the margin, and it is why segment operating margin expanded even as the disclosed consumer ARPU fell 22%.

"Enterprise revenue is quite sticky. We have never lost an enterprise customer. They're quite happy with the Starlink capability that we have right now. Bret mentioned, we are only 10% penetrated in the aviation industry, which definitely represents a lot of headroom for revenue going forward." — Gwynne Shotwell, President and Chief Operating Officer

The consumer side is a genuine tension. Subscribers doubled to 12.0 million while disclosed ARPU fell from $85 to $66, and management was explicit that geographic expansion will continue to press it. Consumer revenue of $2,485M against an average subscriber base of roughly 11.15 million works out to about $74 per subscriber per month, above the $66 the company discloses as ARPU, a gap that is consistent with hardware and activation revenue sitting in the same line. The company did not reconcile the two figures, and the definition matters for anyone building a subscriber model.

"While our geographic expansion may drive down blended ARPU over time, we anticipate continued subscriber momentum to deliver strong revenue growth." — Bret Johnsen, Chief Financial Officer

Assessment: Connectivity is doing exactly what our initiation said it needed to do, and the enterprise mix shift is happening faster than we modelled. The 64bp of adjusted-EBITDA margin compression against 293bp of operating margin expansion is worth watching: it means depreciation is growing more slowly than revenue right now, which will reverse when the V3 constellation starts capitalising. We would rather own this segment at 38.6% operating margin with 42% enterprise mix than at 61% EBITDA margin with 33% enterprise mix.

Space: Revenue Up, Cadence Down

Space revenue grew 29% on a 17% decline in total launches. The reconciliation is mix: customer launches rose from 9 to 10 while internal Starlink launches fell from 37 to 28, and launch services revenue per customer launch rose roughly 19% to about $64.8M. Management characterised this as "a higher number of large customer launches and a favorable customer shift," which is accurate and also the reason the segment's revenue line and its operating statistics point in opposite directions.

Two facts from the release deserve more attention than they received on the call. First, first-half Space revenue of $1,581M was down 1.9% against $1,611M a year ago, so the quarter's 29% growth is a recovery from a weak Q1 rather than a trend. Second, internal payload mass fell 29.5% year on year while the Starlink subscriber base doubled. The constellation is being asked to serve twice the customers on materially less new deployment mass, which is the operational reason the V3 satellite matters on a timetable rather than as an ambition.

"We're currently launching at our highest Falcon cadence and are on the precipice of operationalizing Starship." — Gwynne Shotwell, President and Chief Operating Officer

Segment R&D of $1,076M was 112% of segment revenue, up 55% year on year, and is the direct cause of the widening operating loss. This is a deliberate allocation, not a deterioration, and it is the cleanest available proxy for Starship spend.

Assessment: the Space segment is now unambiguously a development cost centre with a profitable launch business inside it, and management is not pretending otherwise. The claim that Starship will "reduce the cost to orbit by 99% or more relative to the historical average" is in the earnings release itself, which is a notably specific commitment for a company that discloses no Starship revenue. We treat Space as a roughly $4B-a-year revenue business carrying a development programme of comparable size: first-half segment R&D of $2,006M annualises just above the segment's own revenue run-rate, which is the cleanest statement of what this segment currently is. We would rather see the cadence statistics recover before crediting it with operating leverage.

AI: The Step-Function, and What Sits Underneath It

The AI segment crossed into adjusted-EBITDA profitability at $1,146M, which management framed as the quarter's marquee achievement and which the release describes as cutting the operating loss 49% sequentially. Both statements are true. The segment nonetheless lost $1,257M at the operating line, and the entire distance between those two numbers is $1,885M of depreciation plus $516M of share-based compensation. With $15,828M of capex booked in the same three months, the asset base generating that depreciation roughly doubled during the quarter, which means the D&A charge investors are adding back today is a fraction of the charge they will be adding back a year from now.

"We meaningfully narrowed our AI segment net operating loss to $1.3 billion and turned adjusted EBITDA positive for the segment in the second quarter at $1.1 billion. The incremental revenue from new hosting deals generated high incremental EBITDA margins as we monetized available compute capacity." — Bret Johnsen, Chief Financial Officer

Underneath the headline, the segment is really two businesses moving in opposite directions. AI solutions and infrastructure revenue grew 605% to $2,194M. Advertising revenue fell 13.8% year on year to $367M, and is down 18.4% for the half. Management noted the sequential 7% advertising improvement and did not mention the annual decline. The legacy X advertising business is shrinking while the compute business is exploding, and only the second of those was discussed.

The concentration is the part that was not on the call at all. The 10-Q discloses that a single AI-segment customer, absent from the disclosure a year ago, was 19.5% of consolidated revenue this quarter, and that a second customer spanning all three segments was 18.3%. Together, 37.8% of the company's revenue came from two counterparties. The same filing states that cloud services agreements "may be terminated by either party upon 90 days' notice" after an initial ramp period, which is the contractual reality sitting behind the "$14.1 billion of contracted sales" headline.

Assessment: the AI segment is the reason to own this stock and the reason to size the position carefully. The revenue is real, the customers are named on the call as being among the largest AI developers, and the incremental margins are genuinely high. But this is the first quarter it existed at scale, one customer is most of it, the contracts run on ninety days, and the depreciation freight train has barely left the station. The disclosure we asked for in May has arrived; the track record has not.

Key Topics & Management Commentary

Overall Management Tone: assured to the point of expansiveness, with the confidence concentrated in the two executives least accountable for the capex number. Prepared remarks ran heavily to multi-decade framing while the quarter's most consequential disclosures, customer concentration and ninety-day contract terms, appeared only in the filings. Analyst pushback was narrow and almost entirely about capital intensity; the answers were framed as payback economics rather than as a funding plan, and that was the only place management sounded like it was managing rather than describing.

1. The Operating Loss Nearly Vanished

The quarter's most important financial fact went almost unmentioned on the call. A $970M operating loss became a $143M operating loss, and it happened while the company was absorbing a 55% increase in Starship R&D and a 94% increase in AI R&D. On a business generating $7.8B of quarterly revenue, that is 1.8% of margin away from the GAAP breakeven that our May initiation identified as the central execution question.

"The combination of our global launch leadership, recurring Starlink revenue and rapid growth in AI infrastructure continues to drive improved operating leverage." — Bret Johnsen, Chief Financial Officer

Management led with adjusted EBITDA and net loss, both of which are true and both of which are less demanding measures than operating income. The choice is defensible for a company mid-build, but it means the single cleanest evidence that the combined entity works was left for readers to find in the tables.

Assessment: this is the datapoint that changes our rating. A company can grow 92% and still be uninvestable if the cost base grows with it. This one grew costs 58% against revenue growth of 92%, in a quarter carrying peak development spend across two programmes. That is the operating leverage the thesis required.

2. Capital Expenditure: $18.4B, and Two More Like It

Capex of $18,369M against $7,814M of revenue is a ratio of 235%. It exceeded the ~$13.2B the Street had modelled by 39%, and it is the proximate cause of the after-hours reversal. Asked directly about the trajectory, the CFO gave the guidance that did the damage.

"If you look at the rest of this year, to answer specifically on CapEx, I think you should probably think that the next 2 quarters are very similar to the current quarter from a CapEx level perspective." — Bret Johnsen, Chief Financial Officer

Taken literally, that puts full-year 2026 capital expenditure near $65B, against first-half operating cash flow of $3,466M and first-half free cash burn of roughly $25.0B. Management's defence is that the character of the spending has changed.

"What we're seeing right now is a situation where we're able to deploy, specifically on the AI compute side, we're able to deploy capital in such a way that we're getting less than a 1-year payback. It's almost like your capital is becoming almost an item that you would put otherwise in COGS because it's moving so quickly and as far as monetization and moving into revenue." — Bret Johnsen, Chief Financial Officer

The sub-one-year payback claim is the load-bearing assertion of the entire capital-allocation case, and nothing in the filings substantiates it. If it is right, spending more is obviously correct and the market's reaction was wrong. If it is optimistic by even a factor of two, the company is committing $65B a year to assets that depreciate over five.

Assessment: we take the payback claim seriously because the AI segment's incremental EBITDA margins support directionally aggressive economics, and because $100B of cash and securities means the company does not need it to be true this year. We do not take it on faith. This is the number to audit each quarter, and the audit is straightforward: AI capex of $15,828M in Q2 against AI adjusted EBITDA of $1,146M is currently a ratio of 13.8 to one.

3. The $100 Billion ARR Target and the Arithmetic It Requires

The most consequential forward statement on the call was a run-rate target for December.

"We believe this puts us on a trajectory, including contribution from Cursor, to reach $100 billion of ARR or annualized revenue run rate by the end of this year based on our expected revenue in the month of December of this year." — Bret Johnsen, Chief Financial Officer

A $100B annualised run-rate measured on December revenue requires roughly $8.33B of revenue in that single month. Q2 revenue of $7,814M works out to about $2.60B a month, so the target implies a 3.2x increase in monthly revenue over six months. Connectivity and Space cannot plausibly deliver more than a fraction of that. If Connectivity is running at roughly $2.0B a month by December and Space at $0.4B, AI has to produce nearly $5.9B a month, against the $731M a month that AI solutions and infrastructure generated in Q2. That is roughly an eightfold increase in half a year.

The target is not arbitrary. It is broadly consistent with 2 GW of compute monetised at the $30 to $50 per watt management cited. It is not consistent with the roughly $7 per watt currently being realised. The entire bridge is the monetisation rate, and management did not walk investors across it.

"Yes, to be clear, the $100 million ARR in December is not a question mark, but that's what we would achieve if we basically did nothing. So like, so I think it may be higher than that. It probably will be higher than that." — Elon Musk, Chief Executive Officer

The figure discussed throughout the call was $100 billion; the "$100 million" is a verbal slip. The substance is the framing: a number described as achievable "if we basically did nothing" requires the AI business to grow roughly eightfold in two quarters.

Assessment: we do not model to this. We model the contracted portion, which is $14.1B of Cloud Services Agreements plus the $6.7B signed in July, and we treat the remainder as upside. Investors should be alert to the possibility that the target is met on paper by a December that includes Cursor and a full month of newly-ramped capacity, and that the run-rate proves to be a peak rather than a base.

4. What "Contracted Sales" Actually Means

The release headlines "$14.1 billion of contracted sales" from Cloud Services Agreements and management added a further commitment on the call.

"For example, in the first few weeks of the third quarter, we've already contracted an additional $6.7 billion of cloud services revenue over a 6-month period that begins ramping starting in October of this year." — Bret Johnsen, Chief Financial Officer

The 10-Q defines contracted sales as the value of contracts "for the non-cancellable, enforceable period." It also discloses that cloud services agreements are generally terminable by either party on 90 days' notice after an initial ramp. Those two disclosures constrain each other, and neither was discussed on the call. Separately, the company's $47,461M of backlog is described as relating primarily to Space agreements and Connectivity enterprise and government contracts, with 56% expected to convert within one year. The durable, long-dated contract book sits in the segments that are not growing 247%.

Assessment: the AI revenue is contracted in the sense that it is signed and ramping. It is not contracted in the sense that Connectivity's government backlog is contracted. Analysts modelling AI revenue as an annuity are extrapolating from a monthly-fee arrangement with a quarterly exit. This is the most under-discussed risk in the quarter.

5. Starlink Subscriber Growth Against ARPU Compression

Starlink added a record 1.7 million net consumer subscribers, reaching 12.0 million and doubling year on year, while holding ARPU flat sequentially at $66. Against the prior-year quarter, ARPU is down 22.4% from $85.

"During the second quarter, we added net more than 1.7 million Starlink subscribers globally, consumer. This reflects our best quarter of new customers to date, an increase from the 1.4 million we added during the first quarter. We achieved this by holding ARPU, which is the average revenue per user, stable at $66 per month compared to the first quarter." — Gwynne Shotwell, President and Chief Operating Officer

Our May initiation modelled blended ARPU stabilising in the $90 to $95 range by FY2028 as enterprise mix grew. The enterprise mix thesis is working, at 42.1% of segment revenue against 33.5% a year ago. The consumer ARPU thesis is not: at $66 the disclosed figure is materially below where we expected it, and management guided it lower still.

Assessment: the trade is working in aggregate because segment operating margin expanded 293bp while ARPU fell 22%. But the two halves of the Connectivity story should now be modelled separately. Consumer is a volume business with declining unit economics in a deliberate land grab; enterprise and government is a contract business growing 108% with pricing power. Blending them into a single ARPU line obscures both.

6. Starship V3: The Heat Shield Claim

Two Starship V3 flights occurred in the past ninety days, and the second produced the most consequential technical statement of the call. Flight 13, in July and therefore subsequent to the quarter end, deployed 20 production V3 satellites, demonstrated in-space relight of a Raptor engine and executed what the release calls "the softest ever splashdown of Starship, providing critical views of an intact heatshield."

"Look, I don't want to jinx it or anything, but I think I'd consider the heat shield problem solved at this point. Now we want to take a close look at the ship that is currently floating in the ocean." — Elon Musk, Chief Executive Officer

He went further, stating that "we do not see any technical obstacles at this point to achieving full and rapid reusability." A tower catch attempt is tentatively scheduled for the end of August subject to regulatory approval, and management expects at least one flight a day roughly a year from now.

Assessment: our May initiation identified in-orbit refuelling, not the heat shield, as the binding constraint on Starship monetisation, and nothing this quarter changes that. Refuelling remains undemonstrated and the lunar architecture requires eleven or more tanker flights per mission. But the heat shield was the gating item for reusability, and reusability is the gating item for the V3 constellation deployment that Connectivity now urgently needs. If the claim holds, the Starship option moves from long-dated to medium-dated.

7. Starlink Mobile and the EchoStar Spectrum

The FCC approved the EchoStar licence transfer for 65 MHz of US spectrum plus certain global Mobile Satellite Service licences, and the transfer closing occurred in May. Management framed the resulting capability step in unusually concrete terms.

"Currently, the system that we're operating uses about 5 megahertz of bandwidth, and we operate that through local telco providers. We have 65 megahertz of bandwidth available to us through the EchoStar spectrum, which gives massive increase in capability. Another way to look at it is we were also probably 10x the number of satellites. So we, simplifying, you could look at the Starlink mobile leveraging this, the next-generation satellite and the EchoStar spectrum as being 100x better." — Gwynne Shotwell, President and Chief Operating Officer

Next-generation mobile satellites begin flying next year with service targeted for the end of 2027. The addressable market was framed against the three large US carriers at roughly $600 billion a year between them. The financial mechanics disclosed in the 10-Q are less discussed: total consideration is approximately $19.6B, comprising roughly $11.1B in equity through the issuance of about 261.8 million Class A shares at a fixed reference value of $42.40 per share, plus up to $8.5B related to the payoff of designated EchoStar debt, with any shortfall payable in cash. Payments under the associated spectrum credit agreement run $1,241M in 2026 (of which $856M was paid through June 30), $828M in 2027, and a possible further $827M if closing slips to late 2028.

Assessment: the spectrum is a genuine structural asset and the FCC approval removes the principal execution risk. Two things to hold in mind. The share consideration is fixed in share count rather than in value, so the economic cost of that leg moves with the stock. And the up-to-$8.5B cash component lands on a balance sheet already committed to $65B of annual capex. This is a 2027 and 2028 revenue story funded with 2026 cash.

8. Exclusivity With NVIDIA and the Starmind Satellite

The CEO disclosed a vendor commitment that was not in the release.

"And going forward, we've decided to build exclusively on NVIDIA because we think the Vera Rubin architecture is the best architecture. We think it's the best AI computer, and we greatly value our close cooperation and partnership on many levels with NVIDIA. So we're exclusive to NVIDIA." — Elon Musk, Chief Executive Officer

He added that "our understanding with NVIDIA is that we will receive a very significant percent of their GPUs next year." Alongside it came the Starmind satellite, described as essentially an optimised NVL72 rack designed for orbit, expected to begin launching next year and to be deployed on the ground as well as in space.

Assessment: exclusivity buys allocation, which in a supply-constrained market is worth more than price flexibility. It also removes a negotiating lever and concentrates architectural risk in one vendor roadmap. For a company whose stated advantage is hardware engineering, committing exclusively to a merchant silicon supplier is a meaningful strategic choice, and the Starmind concept is the hedge: if the rack design is genuinely better, the company captures value in the balance of system rather than in the accelerator.

9. Compute Scale Targets, and the Gap Between Watts and Revenue

Nameplate compute reached 1.4 GW from 1.0 GW in Q1 and 0.4 GW a year ago, with more than 2 GW targeted by year end. The forward numbers were considerably larger.

"So we're actually aiming to far exceed that gigawatt number in terms of power online, power cooling and electrical equipment. So our tentative target is to actually have 20 gigawatts at the power and cooling level online by the end of next year. Now I don't think we're going to achieve 20 gigawatts, but we want to have a series of projects that cumulatively come to 20 gigawatts by the end of next year." — Elon Musk, Chief Executive Officer

He settled on "something close to 15 gigawatts" at the power-plant level as the realistic case, and made the strategic logic explicit: the goal is to hold far more power and cooling than GPUs, because balance-of-system is cheap relative to accelerators. On monetisation he was candid about the uncertainty.

"If you say like what's my best guess for monetization per watt of, say, Rubins, my guess would be somewhere between $30 and $50. This is just a guess. I'm just, take a guess as a guess." — Elon Musk, Chief Executive Officer

Assessment: the honesty is welcome and the number is the single most important input in any SPCX model, which is an uncomfortable combination. At $40 per watt, 2 GW supports roughly $80B of annual revenue and the December target makes sense. At the roughly $7 per watt currently realised, it does not. The company disclosed a compute KPI this quarter, which is exactly the disclosure investors need to track the convergence, and we will hold management to it quarterly.

10. Related-Party Financing: $13.3 Billion From a Director's Firm

Not mentioned in the release and not raised on the call: the 10-Q discloses that the company's AI infrastructure hardware is being financed through equipment lease agreements with Valor Equity Partners, whose founder and chief executive serves on the SpaceX board. The arrangements are accounted for as failed sale-leasebacks, meaning the debt sits on balance sheet. The balance was $13,329M at June 30 against $4,507M at December 31, and it generated $327M of interest expense in the quarter.

To put that in proportion: related-party debt is 33.9% of the company's $39,364M of total debt and finance leases, it grew 2.96x in six months, and the related-party interest charge was 52.0% of the quarter's total interest expense. The company also purchased $295M of Tesla Megapack products during the quarter, recorded in property, plant and equipment.

Assessment: the disclosure is proper and the arrangements may well be arm's length and competitively priced. But a third of the debt stack of a newly public company sourced from a director's investment firm, tripling in six months, is a governance fact that belongs in the discussion rather than in Note 17. Combined with the controlled-company exemption we flagged in May, this is the second structural item where public shareholders are asked to take related-party economics on trust. We would like to see pricing terms disclosed.

11. Cursor: Dilution That Increases as the Stock Falls

The $60B Cursor acquisition is signed and expected to close in the third quarter. Management declined to discuss the product roadmap.

"I think we'll leave that to, we'll probably make some announcement or discussion of that outside of the earnings call. But we don't want to jump the gun on regulators for the closing the acquisition, so we're trying to get the acquisition closed as quickly as possible." — Elon Musk, Chief Executive Officer

The mechanic that matters is in the 10-Q. Consideration is Class A stock "based on an implied equity value of Cursor of $60 billion," priced off the seven-day volume-weighted average close immediately preceding closing. The dollar amount is fixed and the share count floats, which means the dilution is inversely geared to SPCX's share price. At the June 16 peak close of $201.80 the deal would have cost roughly 297 million shares, or 2.26% of the 13,176 million outstanding. At $135, roughly 444 million shares. At current levels, roughly 522 million shares, or 3.96%.

Assessment: the 43% de-rating since June has increased the share count SpaceX must issue for Cursor by roughly three quarters. Add the fixed 261.8 million shares owed for EchoStar and pending issuance runs to roughly 784 million shares, about 5.9% of the current base. That is a real cost of the share-price decline that does not appear in any income statement, and it argues for closing the Cursor transaction sooner rather than later.

Guidance & Outlook

SpaceX did not issue formal quantitative guidance for revenue or earnings. What it gave instead was a set of forward commitments across the call, which we treat as guidance for scorecard purposes.

Implied quarter-over-quarter ramp: reaching a $100B December run-rate from a Q2 monthly average of $2.60B requires monthly revenue to compound at roughly 21% a month for six consecutive months. The company has done that in one segment for one quarter. It has never done it consolidated.

Street at: our own model carried FY2026 revenue near $22.6B and treated external compute leasing as unmodelled optionality, on the basis that the IPO prospectus disclosed no contracted external compute revenue. That optionality has now materialised: H1 revenue alone is $12,508M and the Q2 run-rate annualises to $31.3B. Consensus entering the print sat near $37B for FY2026. Both our model and the Street are stale in the same direction.

Guidance style: unusual, and investors should calibrate to it. This management team does not guide quarters; it states multi-year destinations and revises them upward in public. The $1 trillion revenue date moved forward a full year on this call with no supporting bridge. Treat the specific near-term commitments, capex, compute capacity, contracted cloud revenue, as the accountable ones, and treat the destinations as expressions of conviction.

Analyst Q&A Highlights

Converting the Enterprise and Government Backlog

The first question of the company's first public Q&A went to the durability of the Connectivity segment's contract book rather than to Starship or AI, which is a reasonable read of where the earnings actually are. Management answered with stickiness and penetration rather than with a conversion schedule, then volunteered a forward claim about mix that was more aggressive than anything in the release.

Q: "Maybe just one on the Connectivity segment. Can you discuss the backlog of enterprise and government contracts that segment of the business and how we think about that backlog, converting it to revenue and what that might mean for the pace and cadence of growth in the next couple of years?"
— Eric Sheridan, Goldman Sachs

A: "So we actually, we're building up a large enterprise sales team just to go out there, meet with enterprise, corporate and government customers, show them the incredible uptime and low latency of the Starlink system today. So they realize that they can actually treat Starlink as a primary provider, not sort of as a backup provider. This requires just going out there, talking to people, educating them about the current state of the Starlink system and where it's going. And I would expect enterprise revenue to substantially exceed consumer revenue."
— Elon Musk, Chief Executive Officer

Assessment: the answer sidestepped the conversion-cadence question, which the 10-Q answers anyway at 56% of the $47.5B backlog within one year. The volunteered claim that enterprise revenue will "substantially exceed" consumer is the more valuable disclosure, and at 42.1% of segment revenue growing 108% year on year, it is closer than most models assume.

Line of Sight on the 2027 Compute Ramp

A pointed question about whether the multi-gigawatt targets were bounded by permitting, silicon or turbines produced the call's largest number and a claim of preferential allocation from the sole supplier the company has now committed to exclusively.

Q: "Just confirming, you said you expect to be closer to 10 gigawatts of nameplate compute by end of '27, closer to 10 than 5. Do you have line of sight? How confident is your line of sight on permitting, chips and turbines for that?"
— Adam Jonas, Morgan Stanley

A: "So our goal is to have far more power, cooling and electrical equipment than we have GPUs. That's a logical thing to do given the relative expense of GPUs versus balance of system. SpaceX as a company is incredibly good at hardware. In fact, I'd say probably outside of China, SpaceX and Tesla are the 2 best companies on Earth at hardware. Yes, and I think perhaps even including China. So we're taking a small amount of the expertise that we use for rockets and satellites and applying that to scaling terrestrial data centers. And then our understanding with NVIDIA is that we will receive a very significant percent of their GPUs next year."
— Elon Musk, Chief Executive Officer

Assessment: the answer addressed power and cooling convincingly and silicon only through an "understanding" with a supplier. Deliberately over-building balance of system is the right call when GPUs are the scarce input, but it also means a meaningful share of the $65B capex programme buys capacity that generates no revenue until accelerators arrive. The permitting question went unanswered, which matters given the Clean Air Act litigation over the turbines powering Colossus II.

Capital Intensity and the Character of the Spend

The exchange that moved the stock. The question was framed generously, crediting historical capital efficiency before asking where the money now goes, and the answer reframed capex as something closer to a cost of goods sold than an investment.

Q: "One of the hallmarks of SpaceX to date is how efficiently the organization has used capital so far in its journey. But it's also no secret that the capital needs are significant to support your vision from here. Obviously, the company is moving fast. So I was hoping you could talk a bit more about the CapEx outlook across the segments, touch upon any areas where buying versus building might accelerate growth."
— John Godyn, Citigroup

A: "If you look at the rest of this year, to answer specifically on CapEx, I think you should probably think that the next 2 quarters are very similar to the current quarter from a CapEx level perspective. But I do think as you look longer term, I think CapEx, really, I think, from our perspective is all CapEx is not the same, right? What we're seeing right now is a situation where we're able to deploy, specifically on the AI compute side, we're able to deploy capital in such a way that we're getting less than a 1-year payback."
— Bret Johnsen, Chief Financial Officer

Assessment: this is the most important sixty seconds of the call and the direct cause of the after-hours reversal. Two more quarters at $18.4B was a genuine surprise against a ~$13.2B consensus. The payback defence is the right argument to make and is entirely unverifiable from outside; no filing discloses utilisation, contract pricing, or asset-level returns. Until it does, the market will keep pricing the capex as risk rather than as return.

Starship Flight 13 and the Path to Reusability

A question about what the July flight actually taught produced the most quotable engineering claim of the quarter, and one with direct commercial consequences for the Connectivity segment's capacity constraint.

Q: "Could you walk through what you learned on Starship launch 13, what was better than expected? What was worse? And then how do we think about this in the journey to kind of rapid reusability, the heat shields and so on and so forth?"
— Ronald Epstein, Bank of America

A: "But this flight actually went incredibly well. The heat shield appears to be very robust. The ship is still floating in the ocean, and we expect to recover the ship and bring it back for analysis. So I would say things look very good. And that's why we, assuming we receive regulatory approval to do so, we'll attempt to catch the ship with the tower on the next flight which is tentatively scheduled for the end of this month."
— Elon Musk, Chief Executive Officer

Assessment: management engaged the question directly and gave a dated, falsifiable commitment, which is the most useful kind. A tower catch attempt in August is a binary near-term catalyst with a clear read-through: it is the gating item for the V3 constellation deployment that Connectivity needs, and internal payload mass fell 29.5% year on year while the subscriber base doubled.

Durability of Compute Pricing

The sharpest commercial question of the call asked whether the premium pricing on the new cloud deals survives contact with supply. The answer never addressed contract structure and instead argued from the supply side of the memory market.

Q: "And then totally appreciate the $6.7 billion in incremental compute deals. Can you talk more about just how you expect the supply-demand environment to persist going forward and whether you'll be able to keep this kind of premium pricing for these deals and sustainability?"
— Douglas Anmuth, JPMorgan

A: "Then look at the rate at which logic and memory is being produced. And one must always consider the limiting factor here. The limiting factor currently is memory. The memory output is increasing by around 20% per year. Now normally, that would be fantastically fast and amazing for any large mature industry. But ask yourself, is the demand increasing by 20% a year? No, the demand is increasing by 200% a year, maybe higher. So if you've got demand increasing much faster than supply, then Economics 101 would suggest that the price increases. It does not decrease."
— Elon Musk, Chief Executive Officer

Assessment: a coherent industry argument that does not answer the question asked. Pricing durability for this company is a function of its own contracts, and those run on ninety days' notice after ramp. If the supply argument is right, the ninety-day terms are an asset because they allow repricing upward. If it is wrong, they are the mechanism by which revenue disappears. Nobody asked which way the option cuts.

Unpacking the December Run-Rate

The only question that directly interrogated the $100 billion figure got a segment-level attribution rather than an arithmetic bridge, and management used the follow-up to pull a long-range target forward by a year.

Q: "I wanted to come back on the comment about the $100 billion in ARR. Is there any way you can unpack that a little bit and perhaps kind of what it takes to get there?"
— Edison Yu, Deutsche Bank

A: "Well, on the unpacking, I think the key there is it's really growth in all 3 of the business segments. But the biggest piece is, and we already talked about additional business we've already closed in the first couple of weeks of July on the AI cloud services side. But certainly, as you know, we've already announced some previous deals with Google and Anthropic that start to ramp either later this quarter or in October. And so there's a lot of momentum from that side of the business as well as we're seeing strength in our own models and Cursor coming into the fold as well."
— Bret Johnsen, Chief Financial Officer

Assessment: the question deserved a bridge and received a list. "Growth in all 3 segments" cannot be right as a description of a 3.2x increase in monthly revenue when two of the three segments are growing at a fraction of that rate. The disclosure of named counterparties is genuinely useful and the ramp timing is specific. The magnitude remains unexplained, and this was the one place where a follow-up would have been worth more than the question that replaced it.

How Much Compute Must Be Withheld to Train Grok

A modelling question, asked as such, produced the single most useful input of the call for anyone building an AI-segment forecast. Management recognised it as a modelling question and answered in those terms.

Q: "It looks like you're going to, you have a big ramp in compute coming in 2027. How should we think of that capacity being leased versus used for, to train Grok? I guess, in other words, how much of that has to be withheld to train Grok versus that can be leased out to third parties?"
— John Hodulik, UBS

A: "So for Grok training, we do expect the percentage of compute that we use internally for Grok training to decrease over time and to have far more of the compute used for inference or to, for renting, training inference compute to others. So I would expect maybe 10% of our compute to be used for Grok training, something like that."
— Elon Musk, Chief Executive Officer

Assessment: the most model-ready statement on the call, and it cuts both ways. Ninety percent leasable capacity is what the December run-rate target requires. It also implies that the current realised monetisation of roughly $7 per watt is being suppressed by internal training consumption today, which is the most credible explanation available for the gap to the $30 to $50 forward assumption. This is the number to track alongside nameplate compute each quarter.

What They're NOT Saying

1. Customer concentration. Not in the release, not in the prepared remarks, not raised in Q&A. The 10-Q discloses that two customers were 18.3% and 19.5% of consolidated revenue, together 37.8%. The 19.5% customer is AI-segment-only and was below the 10% threshold a year ago, meaning essentially all of the AI step-function came from one counterparty. Management discussed the AI segment for several minutes without mentioning that it is, functionally, one contract.
2. The ninety-day termination right. The phrase "contracted sales" appeared in the release, on the call, and in the CFO's forward commentary. The disclosure that cloud services agreements "may be terminated by either party upon 90 days' notice" appeared only in the risk factors. These belong in the same sentence.
3. Advertising revenue is shrinking. The CFO noted advertising "grew 7% sequentially" and did not mention that it fell 13.8% year on year, or 18.4% for the half. Presenting the favourable comparison and omitting the unfavourable one is a choice.
4. Launch cadence and mass to orbit both declined. The prepared remarks celebrated "78 total launches and 1,041 tons of mass to orbit delivered in the first half." Both figures are down year on year, from 84 launches and 1,102 tons. Framing a decline as a leadership statistic without the comparative is the sort of thing a first-time public company gets away with once.
5. Related-party financing. $13.3B of AI infrastructure debt owed to a firm run by a sitting director, tripled in six months, generating 52% of the quarter's interest expense. Disclosed in Note 17, absent from the release and the call. No pricing terms were provided.
6. No free cash flow discussion. The company disclosed six-month operating cash flow of $3,466M and six-month capex of $28,476M, implying roughly $25.0B of free cash burn in the half. Neither figure was mentioned on the call, and no question addressed the cash trajectory directly. For a company guiding to two more quarters at this capex level, the absence is conspicuous.
7. The ARPU definition. Disclosed ARPU of $66 does not reconcile to consumer revenue divided by average subscribers, which works out near $74. The gap is plausibly hardware and activation revenue, but the company did not define the metric or bridge it, and it is the primary KPI for 32% of consolidated revenue.
8. Terrestrial mobile capital expenditure. Asked directly what building out the terrestrial component of the EchoStar spectrum would cost, management declined three times in one answer: "I'm not going to talk about the CapEx for that right now." Given that the questioner framed the comparison at "well over $100 billion of mobile CapEx over time," a range would have been useful.
9. Cursor's financial contribution. The acquisition is $60B, roughly 4% dilution at current prices, and is included in the December run-rate target. No revenue, margin, or ARR figure for Cursor was provided, and the roadmap question was explicitly deferred. The 10-Q notes the concurrent compute arrangement with Cursor was "not material" this quarter.
10. The Colossus II turbine litigation. The 10-Q discloses a Clean Air Act suit seeking to enjoin operation of the mobile gas turbines powering Colossus II, with a preliminary injunction motion filed in May. Colossus II is named repeatedly in the release as the source of the incremental cloud revenue. An injunction is an operational risk to the segment carrying the growth, and it was not mentioned.

Market Reaction

• Pre-print setup: the shares closed at $125.33 on August 4, up 9.43% on the session on 144.1 million shares against a recent daily average nearer 60 to 80 million. That rally followed a 5.68% gain the prior session off a post-IPO low close of $108.37 on July 31. Entering the print the stock was down 22.6% over trailing thirty days, down 37.9% from its June 16 closing peak of $201.80, and down 7.2% against the $135 IPO price. The S&P 500 was up 13.0% year to date.
• After-hours move: shares fell roughly 7% in extended trading to about $115.98, giving back the session's gain while remaining marginally above the prior close of $114.53. The move was attributed almost universally to the capex figure and the guidance that the next two quarters would look similar.
• Wednesday's session: at the time of writing the shares are trading near $114.93, having opened at $112.37 with a range of $109.21 to $117.50 on 92.9 million shares. The reaction session has not closed and these are intraday levels.

The two-day price path tells the story more clearly than either day alone. The market bought a 9.4% rally into a print it expected to be strong, got a 13% revenue beat and a 69% adjusted-EBITDA beat, and sold the whole rally on a single line item. That is not a verdict on the quarter; it is a verdict on the funding requirement the quarter revealed.

The positioning context matters. This is a company seven weeks public with a very small float relative to its market capitalisation, and it has now given back roughly 43% from its June peak. A rally of 9.4% into a print on nearly double normal volume, in a stock that had fallen for four consecutive weeks, has the signature of short covering and event positioning rather than accumulation. When the capex number landed, there was nothing underneath it.

Then there is the calendar. The first tranche of the staggered lock-up releases up to 911.5 million shares on August 6, which is 1.43 times the entire 638.9 million-share IPO float, and holders of those shares are marked at cost bases far below today's price. An early-release provision that would have freed an additional 10% required the stock to trade at least 30% above the $135 IPO price on five of ten consecutive days before this earnings release; it did not come close. Any read of this week's price action that ignores a supply event forty-eight hours away is incomplete.

Street Perspective

Debate: Is the capital expenditure value-creating or value-destroying?

Bull view: management asserts a payback of under one year on new compute deployments, and the behaviour is consistent with the claim, given $6.7B of additional cloud revenue contracted within weeks of quarter end. On that arithmetic, under-spending would be the error, and the capex "overshoot" is a demand signal rather than a cost problem. The balance sheet supports it: $100B of cash and securities against $39.4B of debt is a net cash position of $60.6B.

Bear view: $18.4B of capex against $7.8B of revenue, guided to repeat twice, means roughly $65B of full-year spending against first-half operating cash flow of $3.5B. Nothing in the filings substantiates the payback claim, which is unaudited and unquantified. The most bearish sell-side view frames negative free cash flow persisting to the end of the decade, and at that horizon the cash pile funds perhaps two years of the current pace before the company returns to markets.

Our take: the bulls have the better argument on direction and the bears have the better argument on evidence. Incremental adjusted-EBITDA margin of 62% on the year-on-year revenue increase is consistent with attractive asset-level returns, and the net cash position means this is a solvency non-issue in 2026. But the company has disclosed a compute KPI and a segment capex line without disclosing utilisation or contract pricing, so the payback claim cannot be checked. We would assign the benefit of the doubt for two more quarters and require the disclosure after that.

Debate: How durable is the AI revenue?

Bull view: $14.1B of contracted cloud agreements, counterparties named on the call as among the largest AI developers, $6.7B more signed in July, and a segment that reached adjusted-EBITDA profitability earlier than anyone modelled. Compute is supply-constrained and the constraint is worsening, which argues for pricing power on renewal rather than pricing risk.

Bear view: one AI customer accounted for 19.5% of consolidated revenue this quarter, up from nothing a year ago, and the agreements terminate on ninety days' notice after ramp. A single counterparty decision can remove roughly a fifth of the company's revenue within a quarter. Buyers at this scale are also the most credible candidates to build their own capacity, which is a risk the filings name explicitly.

Our take: the bear framing is correct about the contractual position and probably wrong about the outcome. Hyperscale AI buyers are capacity-constrained and switching away from a supplier who is adding gigawatts is not obviously in their interest during a shortage. The asymmetry worth pricing is timing: ninety-day terms make the revenue reprice quickly in both directions, which means SPCX will look like a much better business than it is in a tight market and a much worse one in a loose market. Underwrite the segment at a discount to its current run-rate, not at a premium to it.

Debate: Does the August 6 unlock change the investment case?

Bull view: the supply event is dated, sized and known, and the 43% de-rating from the June peak has already discounted much of it. Fundamentals are materially unchanged and arguably better after this print, so a technically-driven decline into the unlock creates the entry point that the IPO price never offered. The most widely circulated sell-side view on the unlock takes precisely this position.

Bear view: 911.5 million shares is 1.43 times the IPO float, released into a stock that has fallen five weeks out of the last seven, held by investors whose cost bases sit far below the current price. The main 180-day tranche follows in December, and the founder's holding unlocks in June 2027. This is the first of at least three supply events, not the last one.

Our take: both are right about different horizons and the bears are right about the next fortnight. Our May initiation assumed a single 180-day lock-up expiring around December 12; the actual structure is staggered and the first tranche arrives four months earlier than we modelled, which is a genuine negative surprise relative to our own work. We are upgrading anyway, because a twelve-month rating that flinches at a dated forty-eight-hour supply event is not a rating. Investors sizing a position should expect to be able to add lower and should plan for it.

Model Update Needed

Our SPCX model was built in July from the IPO prospectus and deliberately excluded external compute leasing, on the stated basis that the prospectus disclosed no contracted external compute revenue and the Hold thesis carried it as optionality. That optionality converted this quarter, and the model is now stale in several places at once.

Valuation impact. At roughly $114.93 the equity is about $1.51 trillion on 13,176 million shares, and enterprise value is near $1.45 trillion after $39.4B of debt and $100.0B of cash and securities. That is 46.5 times the Q2 annualised revenue run-rate and 102.7 times annualised adjusted EBITDA. Both are demanding. Both are also an order of magnitude below the roughly 94 times trailing revenue and 266 times EBITDA we calculated against the $1.75 trillion IPO indication in May, and the compression is mostly revenue rather than price.

We are raising our fair-value range to $1.65 to $1.95 trillion, or roughly $125 to $148 per share, from the $1.35 to $1.55 trillion we published on the S-1. The revision reflects three changes: the AI segment converting from unmodelled optionality into a $2.2 billion quarterly infrastructure business, Connectivity's operating margin expanding 293bp on 66% growth with enterprise mix at 42%, and consolidated operating loss narrowing to $143 million. Against the current price that range implies roughly 9% to 29% upside. It is not a wide margin of safety, and it depends on the AI revenue proving more durable than its ninety-day contract terms imply.

Thesis Scorecard Post-Earnings

Scored against the standing thesis established in our May 20 S-1 initiation.

Overall: thesis strengthened. Three of the four bull pillars were confirmed, one materially so, and the two bear points that mattered most to our Hold rating, valuation and the absence of AI segment disclosure, both improved. The offsetting deterioration is concentrated in structure rather than in operations: related-party financing, customer concentration and a lock-up schedule that is earlier and larger than we assumed. That is a different risk profile, not a worse business.

Action: buy, with position sizing that anticipates volatility through the August 6 unlock and the December tranche. We would rather establish a position into a dated, mechanical supply event than chase the stock after the AI segment prints a second confirming quarter.