SpaceX — Research Report (2026-08-16)

Source note before you start. The harness flagged "no EDGAR match," but that is now out of date: SpaceX listed on Nasdaq as SPCX on 12 June 2026 and files with the SEC under CIK 0001181412 (S-1, Q2 earnings release). SEC.gov and several data sites blocked direct automated fetching in this session, so filing figures below are taken from named secondary sources that quote the S-1 and the Q2 10-Q, plus the company's own Q2 2026 earnings call transcript. Every number is labelled reported / estimated / speculative. Nothing here is investment advice.

TL;DR

• What it is: the world's dominant rocket launcher (Falcon 9), the largest satellite broadband network (Starlink, 12m subscribers), and — since the February 2026 xAI merger — a top-tier AI compute business, all inside one Nasdaq-listed company.
• My take: two genuinely monopoly-grade assets (launch, spectrum + Starlink) are being used to fund a third business where SpaceX is a price-taker, and the market is paying monopoly prices for all three.
• Educational decision: WATCH — Medium conviction. The business is real; the price assumes the AI bet works and pays for itself.
• $1.85T market cap on ~$38.9B of 2026E revenue = ~46× EV/Sales (Nvidia trades at ~17× — this is priced for a decade of flawless execution).
• H1 2026 free cash flow ≈ −$25.0B (it burns cash faster than any company in history).
• $100B cash vs ~$16B/quarter burn rate = roughly 4–6 quarters (a capital raise is a when, not an if).
• Starlink ARPU $66/mo, down 22% year-over-year (growth is now coming from cheaper customers).
• Strongest bull point: connectivity earned a 38.6% operating margin in Q2 and is the only segment making money — a compounding cash engine with no equivalent competitor.
• Strongest bear point: "adjusted EBITDA" of $3.5B ignores the depreciation from $18.4B of quarterly capex; that bill arrives over the next 3–5 years.
• Watch next: the 2027 capex guide and how it is funded (Q3 report, est. early November 2026).

1. Revenue engine — how the cash register actually rings

So what: three totally different businesses share one balance sheet — one prints cash, one is a subsidised national asset, and one is a capital furnace. Only the first one is profitable today.

SpaceX is no longer "a rocket company." After absorbing xAI (which itself already contained X, formerly Twitter) in an all-stock deal on 2 February 2026, and agreeing to buy AI coding firm Cursor/Anysphere for $60B on 16 June 2026, the listed entity is a three-headed conglomerate (CNBC, CNBC).

The cash-register sentences

Connectivity: A rural household, airline, ship, enterprise, or mobile carrier pays SpaceX a monthly subscription plus an upfront terminal price for bandwidth from the Starlink constellation, sold direct online or wholesale through carriers like T-Mobile — producing recurring subscription revenue plus transactional hardware revenue.

Space: NASA, the US Department of Defense, and commercial satellite operators pay SpaceX a fixed price per launch, or per contract milestone, to deliver mass to orbit and crew to the ISS — producing project revenue recognised against milestones and backed by a multi-year backlog.

AI: Enterprises and AI labs (Google and Anthropic were named on the Q2 call) pay per committed block of compute capacity; consumers pay a subscription for Grok/X Premium; advertisers pay per impression on X — producing a mix of contracted, subscription, and advertising revenue.

Revenue architecture

The driver equations

Connectivity = (subscribers × monthly ARPU × 12) + (terminals × price) + carrier wholesale fees
Space        = (launches × price/launch) + government program milestones (HLS, Commercial Crew, Starshield)
AI           = (GW deployed × utilisation × realised $/GW-hour) + subscriptions + (impressions × CPM)

A number the company does not print (my calculation, estimate). Q2 connectivity revenue was $4.3B across roughly 11.15m average subscribers — about $129/month of connectivity revenue per subscriber, versus the $66 consumer ARPU management quotes. The ~$63/month gap implies roughly half of connectivity revenue is not consumer subscriptions — it is hardware, aviation/maritime, government Starshield, and direct-to-cell wholesale. That mix is higher-margin and much harder for Kuiper or a Chinese constellation to attack than the consumer dish business. (Caveat: segment definitions may not align exactly; treat as directional.)

So what: connectivity is the only stream creating equity value today. The stream most likely to disappoint is AI, and the KPI that reveals it first is cloud contracted sales ($14.1B signed in Q2) failing to grow alongside capex.

2. Financials — the cash reality

So what: revenue is compounding beautifully and cash is haemorrhaging. Both are true, and the second one is what sets the risk.

"Profit is opinion, cash is fact."

Annual (S-1 basis; xAI consolidated retroactively — reported)

Sources: Morningstar, PitchBook, Yahoo/PitchBook charts. n/d = not disclosed in the sources I could reach.

Quarterly 2026 (reported)

Sources: The Transcript / S-1 Q1 data, CNBC Q2, Q2 call transcript, TradingKey. H1 FCF is my arithmetic.

Three things to notice.

The Q1 net loss of $4.28B versus Q2's $541M is not a turnaround — Q1 carried heavy non-cash charges from the merger and from cancelling and re-granting Musk's xAI award in March 2026 (Forbes). Read the two quarters together, not separately.

Adjusted EBITDA of $3.5B is a pre-depreciation number. Depreciation is how the accounting system spreads the cost of that $18.4B of hardware across the years it is used. Almost none of the AI capex has started depreciating yet. When it does — GPUs are typically written off over ~5 years — roughly $3B+ per quarter of new depreciation arrives for every $15B quarter of AI capex sustained. That converts today's "adjusted EBITDA growth" story into GAAP losses for years.

Balance sheet (Q2 2026, reported): cash and marketable securities $100B, debt and finance leases $39.4B → net cash ≈ $60B. Funded by an IPO that raised $85.7B including the over-allotment, plus $25B of investment-grade bonds.

3. Operating KPIs

So what: every operating metric is going up except the one that determines profitability per customer.

Management guided to a $100B annualised revenue run-rate by December 2026 and pulled the $1T revenue target from 2031 to 2030 (Q2 call). Treat both as management ambition, not guidance you can underwrite.

4. Valuation

So what: at ~46× forward sales, the price already contains the Starship success, the Starlink monopoly, and an AI business that becomes a hyperscaler. There is no margin for one of the three failing.

Which multiple, and why

The company is GAAP loss-making, so P/E (price per $1 of annual profit) is undefined and useless. EV/EBITDA (whole-business value versus operating cash proxy) is misleading here because EBITDA deliberately excludes the depreciation from a historic capex wave that hasn't landed. That leaves EV/Sales — enterprise value divided by revenue — the standard tool for high-growth, pre-profit companies. It is a blunt instrument, and I'm using it because the alternatives are worse.

Sector benchmarks (§4b — always give a multiple its yardstick)

Honest caveat: there is no clean single-sector comp for this company. It is simultaneously a satellite operator, a defence contractor, and an AI cloud. I did not gather sourced multiples for mature satcom (Viasat, Iridium, EchoStar) in this pass — that gap is stated, not filled with a guess.

What a reasonable buyer might pay (illustrative sum-of-the-parts — estimate, not a model)

The market is paying roughly 2× my illustrative SOTP. That gap is the option value on Starship, Mars, orbital data centres, and Musk. Whether that option is worth $900B is the entire debate.

5. Value-chain position — bottleneck or commodity?

So what: SpaceX is the bottleneck in space and a price-taker in AI. It is spending the profits from the first to compete in the second.

The clearest tell that the AI position is weak: SpaceX paid $60B — about 15× revenue — for Cursor, after Grok failed to compete with Anthropic's Claude Code and OpenAI's Codex (CNBC). Monopolists don't buy their way into product-market fit.

Competition scoreboard (mid-2026)

6. Ownership, incentives, and the share-supply schedule

So what: one person holds 85% of the votes, has a $760B pay package tied to colonising Mars, and the shares that back it start hitting the market in stages between now and 2027.

Read the incentive honestly. The pay package rewards market capitalisation and Mars, not free cash flow, not return on invested capital, not dividends. An incentive scheme that pays on valuation milestones tells you what management will optimise for. That is a feature if you're buying the narrative and a warning if you're buying the cash flows.

Lockup release schedule (the supply overhang)

That the 6 August unlock was absorbed upward is a genuine bull data point. It is also only the second of nine stages.

7. PE / quality lens

So what: a private-equity buyer would love the connectivity annuity and refuse to underwrite the rest.

What a buyer would love: ~39% operating margin, month-to-month recurring revenue with real switching costs (you own the dish), a physically unreplicable constellation, a launch monopoly that lowers its own cost of goods, and $12.1B of deferred revenue (cash collected before service is delivered — negative working capital, which is a gift).

What a buyer would fear: capex intensity with no visible terminal state; ARPU declining 22% a year; ~20% revenue concentration in a single politically volatile customer (the US government); 85% voting control with no board check; and the fact that the profitable business is being used as collateral for the unprofitable one.

ROIC vs WACC — the moat in one number — is currently not computable in any meaningful way: the invested capital was deployed in the last four quarters and the returns haven't arrived. That absence is itself the finding.

8. Bull / base / bear — attributed to named institutions

So what: the sell side disagrees by roughly $1 trillion. When professionals disperse that widely, price is being set by narrative, not analysis.

🐂 Bull — Morgan Stanley (Adam Jonas), Overweight, PT $300

• Initiated 7 July 2026, citing "AI-era upside" and a $3.3 trillion revenue outlook by 2040 (Seeking Alpha).
• Morgan Stanley has separately laid out a path to $600 in its upside scenario (Motley Fool, 11 Aug 2026).
• More bullish still: IPO underwriter Raymond James initiated at Strong Buy, PT $800 (Motley Fool) — note the underwriter conflict.

😐 Base — Goldman Sachs (Eric Sheridan), Buy, PT $205

• Also initiated 7 July 2026. Goldman and Morgan Stanley — the two lead underwriters — landed ~$1 trillion apart in implied valuation (Quartz).
• Implies ~46% upside from $140: growth delivered, but the AI multiple normalises toward compute-peer levels.

🐻 Bear — Morningstar, 1-star, fair value $62

• ~58% downside; Morningstar has maintained this through the post-earnings period (Investing.com forecast page).
• The bear mechanism: capex overwhelms cash generation, a dilutive raise follows, and the sales multiple compresses toward CoreWeave-like levels.

Consensus: average 12-month target $227, range $62–$800; roughly 28 buy / 5 hold / 2 sell (aggregator counts differ between TradingView and MarketBeat — treat as approximate).

Catalysts

What would prove the thesis wrong

My cautious view is wrong if: free cash flow turns positive within four quarters without a capital raise; connectivity operating margin holds above 38% while subscribers keep compounding; and the AI segment sustains the 43% adjusted-EBITDA margin it just hit after depreciation starts flowing through the income statement. Any two of those three and the $46-per-dollar-of-sales price becomes defensible.

§5 — Practitioner decision layer

An educational framework for making reasoning falsifiable. Not investment advice.

5a. Reverse expectations (Mauboussin)

This is a multiple-and-consensus approximation, NOT a discounted cash flow model.

Base-rate test. Sustaining >35% revenue CAGR for seven years while crossing $100B of revenue is close to unprecedented. Amazon and Nvidia are the canonical examples — a sample of roughly two out of thousands of large-cap attempts.

Judgment: STRETCHED, not implausible. The exit-2026 run-rate guide of $100B, if delivered, would put the company most of the way there by itself. But that guide is management ambition, not contracted revenue — and only $47.5B of backlog stands behind a $1.85T price.

Assumptions dated 16 Aug 2026. Consensus revenue from Motley Fool, 12 Jul 2026.

5b. Variant perception

1. Consensus believes: SpaceX is a launch monopoly and a Starlink cash machine now supercharged by a credible AI compute business — a genuine three-engine compounder worth an average $227.

2. This report's differentiated view: the market is applying hyperscaler economics to an AI segment where SpaceX is a structural price-taker with no silicon edge and a losing model franchise, while the durable monopoly assets (launch, Starlink, spectrum) plausibly account for only $500–800B of the $1.85T. The remainder is an option on capex converting into returns it has not yet demonstrated.

3. Evidence: the AI segment lost >$6B in 2025 and −$2.5B in Q1 2026; connectivity was the only segment with an operating profit in Q2; SpaceX paid 15× revenue for Cursor because Grok lost to Claude Code and Codex; Starlink ARPU has fallen $99 → $66 in three years; H1 2026 free cash flow was −$25B against $100B of cash.

4. Why the market may be wrong: the effective float has been artificially scarce — 48% of shares are Musk's and locked to June 2027, and the first major unlock only occurred on 6 August. Post-IPO index inclusion and momentum flows set the marginal price. The clearest tell is dispersion: when the two lead underwriters publish targets $1 trillion apart ($205 vs $300, with a third underwriter at $800), the range is not analysis — it is the absence of it.

5c. Probability-weighted expected value

EV = (0.20 × $400) + (0.45 × $200) + (0.35 × $70) = $80 + $90 + $24.5 = ~$194

Read this honestly. The expected value (~$194) sits above the $140 price, which mechanically argues for buying. I am not hiding that. But the outcome band is $70 to $400 — an EV built on subjective probabilities across a 5.7× spread carries almost no information. Position sizing and entry price matter far more than this point estimate. This is a scenario discipline, not a forecast.

5d. Pre-mortem — it is August 2028 and the stock is at $70

Cause of death #1 — The capex/dilution spiral. AI capex ran at $60–100B/yr toward the 10 GW target while cloud revenue lagged. A large equity raise landed in 2027, and simultaneously the depreciation from 2026's $28B of capex hit the income statement, turning "adjusted EBITDA growth" into widening GAAP and free-cash-flow losses. The sales multiple compressed from 46× to 15×.

• Earliest signal: Q3 2026 report — 2027 capex guide rising while cloud contracted sales fail to exceed $14.1B/quarter. Any equity or convertible announcement.
• What would reduce this risk: two consecutive quarters of positive FCF, or capex funded entirely from operating cash and the existing $100B.

Cause of death #2 — Starlink's margin erodes from underneath. ARPU continued falling below $50 as growth shifted to low-income geographies, while Qianfan (satellite costs down 96%) and Amazon Leo undercut price in the profitable markets. Connectivity's 38.6% operating margin fell to the low 20s, and the segment funding everything else stopped funding it.

• Earliest signal: connectivity operating margin below 35%, or net adds decelerating while ARPU still declines — the worst combination.
• What would reduce this risk: direct-to-cell wholesale (T-Mobile, EchoStar spectrum) and aviation/maritime growing as a share of connectivity revenue, since those carry premium pricing.

Cause of death #3 — The key-person / political shock. Musk's attention fragmented across Tesla, politics, and Mars; there is no key-person insurance, and 85% voting control means no board can correct course. A Washington shift turned ~20% of revenue from a moat into a liability, with FAA/NASA conflict-of-interest scrutiny escalating into contract re-competes.

• Earliest signal: NASA or DoD re-competing sole-source awards; congressional action following the conflict-of-interest letters; Musk pledging or selling shares after June 2027.
• What would reduce this risk: a named, credible operating CEO for the space business; genuine independent directors; commercial revenue growing faster than government revenue.

5e. KPI driver tree and sensitivity

                    SPCX equity value
                            │
        ┌───────────────────┼───────────────────┐
   CONNECTIVITY            SPACE                 AI
   (the cash engine)   (the moat)          (the capex bet)
        │                   │                    │
  subs × ARPU        launches × price      GW × utilisation × $/GW-hr
  (12.0m × $66)      (78 in H1)            (1.4 GW today)
  + hardware         + gov't milestones    + subs + ad impressions
  + D2C wholesale

Rough sensitivities (estimates, stated assumptions — not a model):

The asymmetry is the whole point. A 10% ARPU move flows almost entirely to profit. A 10% compute move flows almost entirely to cash out the door today and profit only years later. Which is why I watch ARPU for the thesis and capex for the risk.

5f. Scuttlebutt proxies

Decision (Educational — not investment advice)

• Decision: Watch
• Conviction: Medium
• Core assumption: By the Q4 2026 report, SpaceX will still not be covering capex from operating cash flow, and will have announced or clearly signalled incremental capital beyond the $25B of bonds already issued.
• Review date: Q3 2026 earnings — estimated early November 2026
• Controlling question: What is the 2027 capital expenditure guide, and how is it funded — internal cash, debt, or equity?
• Review trigger: Any equity or convertible issuance announcement; or two consecutive quarters of positive free cash flow; or connectivity segment operating margin falling below 35%.

Why Watch and not Long, given EV > price? Because the next four quarters contain two known negatives with dated schedules — seven remaining lockup tranches through June 2027, and a capital raise that the arithmetic makes near-certain — and neither is in the price yet. Interest band: below ~$110 (≈36× 2026E and ≈19× 2027E EV/Sales), which is also near the $104.83 post-IPO low. Horizon: 2–3 years, because that is when the AI depreciation and the Starship cadence both become visible facts rather than forecasts.

📚 What this company teaches

1. "Adjusted EBITDA" and free cash flow can point in opposite directions for years — and the gap has a name. SpaceX reported adjusted EBITDA of +$3.5B and free cash flow of −$16.0B in the same quarter. The bridge between them is capex: money spent today on assets that will be expensed as depreciation over the next five years. A company in a capex ramp will always look profitable on EBITDA and terrible on cash — and then, three years later, look the reverse as depreciation peaks while spending slows. Always ask which phase of that cycle you are buying.

2. Being the bottleneck in one industry does not make you the bottleneck in another. SpaceX has near-absolute pricing power in launch — Amazon had to buy rockets from its own competitor. It has essentially none in AI compute, where Nvidia sets the input price and the models are commoditising. The lesson from §0 of this vault's playbook: find the monopoly bottleneck in the value chain. The corollary is that a company can be the bottleneck and the commodity simultaneously, in different segments, and the market will often price the whole thing at the bottleneck multiple.

3. Read the pay package to find out what management is actually optimising. Musk's $760B award vests on market capitalisation milestones and a Mars colony — not on return on invested capital, free cash flow, or margins. Incentives are the most honest document a company publishes, because unlike the strategy deck, someone has to pay out on them. When the scoreboard is valuation, expect decisions that grow valuation, which is not always the same as decisions that grow value.

🎯 Self-test question: SpaceX's connectivity segment earned a 38.6% operating margin in Q2 2026, while ARPU fell 22% year-over-year and subscribers doubled. Explain how both can be true at once — and then say which of the two trends you would rather see reverse, and why.

📖 Glossary

Adjusted EBITDA — Earnings before interest, tax, depreciation and amortisation, with certain items (usually stock compensation and one-offs) also excluded. A cash-ish profit proxy; here it flatters results because the depreciation from a huge capex wave hasn't started yet.

ARPU — Average Revenue Per User, per month. Subscribers × ARPU = subscription revenue.

Backlog — Contracted future revenue not yet delivered. Comparing it to market cap tells you how much of the price is already locked in (here: 2.6%).

Base rate — How often something has historically happened across a large sample. The antidote to being persuaded by one compelling story.

Capex (capital expenditure) — Cash spent on long-lived assets (satellites, GPUs, factories). Doesn't hit profit immediately; hits cash immediately.

D&A (depreciation & amortisation) — The accounting method that spreads capex across an asset's useful life. Non-cash today, but it reflects a real past cash outflow.

Deferred revenue — Cash collected before the service is delivered. A negative-working-capital gift: customers fund the business.

Dual-class shares — Two classes with different voting rights. SpaceX's Class B carries 10 votes to Class A's 1, giving Musk ~85% voting control on a 48% economic stake.

EV (enterprise value) — Market cap plus debt minus cash. What it costs to buy the whole business, debt-neutral.

EV/EBITDA — Enterprise value ÷ EBITDA. The private-equity standard (typically ~7–12× for mid-market). Distorted here by the capex wave.

EV/Sales — Enterprise value ÷ revenue. Used for pre-profit or loss-making companies where P/E and EV/EBITDA don't work.

FCF (free cash flow) — Operating cash flow minus capex. The cash actually left over. The anti-story metric.

Float — Shares actually available to trade. A small float can make a price move on little real conviction.

Greenshoe (over-allotment) — An option letting IPO underwriters sell up to ~15% extra shares. Explains why the "$75B" IPO raised $85.7B.

LEO (low Earth orbit) — ~300–2,000 km altitude. Where Starlink, Kuiper and Qianfan operate; low latency, but requires thousands of satellites for coverage.

Lockup — A post-IPO period during which insiders cannot sell. Expiries create known share-supply events.

Monopsony — A market with one dominant buyer (here, the US government for national-security launch). The mirror image of a monopoly, and just as much of a pricing risk.

Net debt / net cash — Total debt minus cash. SpaceX has ~$60B net cash, so the usual leverage ratios don't apply; quarters-of-runway does.

OCF (operating cash flow) — Cash generated by day-to-day operations, before capex.

P/E (price-to-earnings) — Price per $1 of annual profit. Undefined here because the company is loss-making.

Pre-mortem — Assume the investment already failed, then work backwards to the causes. Surfaces risks that forward-looking optimism hides.

Price-taker — A buyer or seller with no power to set price. SpaceX is a price-taker on GPUs.

Quiet period — The ~25 days after an IPO during which underwriting banks cannot publish research. It ended 7 July 2026, which is why all the initiations cluster on that date.

ROIC vs WACC — Return on invested capital versus the cost of that capital. If ROIC > WACC, each $1 reinvested creates value. The moat in one number.

SOTP (sum-of-the-parts) — Valuing each segment separately and adding them up. Appropriate when segments have different payers, margins and capital needs — exactly SpaceX's situation.

Take rate — The percentage a platform keeps from transactions it enables. Not a SpaceX metric (it sells capacity directly), included because it's the standard unit-economics term for marketplace businesses.

Variant perception — An explicit, evidence-backed statement of where and why your view differs from consensus. If you can't write one, you agree with the market — which is a valid answer.

Sources: SpaceX S-1 (SEC, CIK 1181412) · Q2 2026 earnings release (SEC) · Q2 2026 earnings call transcript · CNBC — Q2 2026 results · CNBC — IPO day · CNBC — xAI merger · CNBC — Cursor acquisition · CNBC — Starlink reliance · Morningstar — 6 charts on S-1 financials · PitchBook — S-1 financials · Sacra — SpaceX revenue · Bloomberg — $760B Musk pay package · Forbes — Musk milestones · Fortune — Mars-shot pay · CII letter to SpaceX · Quartz — Goldman vs Morgan Stanley · Seeking Alpha — MS/GS initiations · Motley Fool — analyst targets · Motley Fool — MS $600 path · Motley Fool — 2026/27 revenue estimates · Motley Fool — $14.1B cloud contracts · Investing.com — consensus estimates · TradingView — SPCX quote · MarketBeat — SPCX forecast · CNN — Aug 6 lockup · TradingKey — Musk stake · TradingKey — Q2 analysis · The Transcript — Q1 2026 figures · Barchart — EchoStar spectrum · EchoStar IR — spectrum sale · CGTN — Qianfan cost reduction · SpaceNews — Chinese megaconstellations · NASASpaceflight — Starship Block 3 / HLS · multiples.vc — Nvidia · multiples.vc — CoreWeave · multiples.vc — Rocket Lab · Glassdoor — SpaceX reviews · startup.jobs — SpaceX openings · T-Mobile — T-Satellite · CNBC — xAI Saudi data centre · Washington Technology — risk factors · Senate letter on Musk conflicts