The Lens · Deep Dive
SpaceX: The Complete Picture — Every Lever That Moves SPCX
The full anatomy of the biggest IPO ever. After merging with xAI, SpaceX isn't a rocket company — it's four companies in one. Who owns it, every revenue stream, the variables that move the stock, the "Elon effect," and the honest risks. Hover or tap any underlined term.
Dragonfly Lens · June 9, 2026 · The biggest picture we can build — every claim sourced, fact separated from spin.
The short version
- It's not one company. The Feb 2026 xAI merger made SPCX a conglomerate: launch + Starlink + Starshield (defense) + xAI (AI compute). Nothing else on the market spans all four.
- Musk runs it outright: ~42% of the equity but 82.4% of the votes. Public buyers get economics, not control.
- Cash flywheel: ~$27–30B 2026 revenue. Starlink is the engine; defense is a $22B backlog; and AI compute is a brand-new stream — Anthropic pays $1.25B/mo, Google $920M/mo.
- The "Elon effect" is real but double-edged: unmatched execution speed and talent — and total key-man risk with no governance recourse.
- And it's bigger than SpaceX: the Terafab chip mega-fab (with Intel), a two-brained AI stack (Tesla's edge AI + xAI's data centers), and a real ~50–80% chance of a Tesla merger into a $3.4T empire.
The thing that changes everything: the xAI merger
In February 2026, Musk merged SpaceX with his AI company xAI — what CNBC called the largest corporate combination in history, a ~$1.25 trillion deal. So the company IPO'ing as SPCX is not the rocket business you remember. It's a conglomerate with four engines:
| Engine | What it is | Why it matters |
|---|
| Launch (Falcon, Starship) | The world's dominant rocket business — ~$5–6B/yr, most of the planet's payload to orbit. | The moat + the enabler for everything else. |
| Starlink | Satellite internet — the growth + profit engine the whole IPO leans on. | The cash flywheel funding the rest. |
| Starshield (defense) | Military version of Starlink + the new Space Data Network for the Pentagon. | Sticky, high-margin government revenue. |
| xAI (AI compute) | The Colossus data centers + Grok. Now a compute landlord to other AI labs. | A brand-new multi-billion revenue line — see below. |
That's the "1-of-1" you can't price with a normal comp: there is no other company that launches the rockets, owns the satellite network, holds the defense contracts, and rents AI compute to its rivals.
Who actually owns it (and what you'd be buying)
Per the S-1/A filed June 3, 2026:
- Elon Musk: ~42% of the equity but 82.4% of the voting power — via dual-class shares (Class B = 10 votes, Class A = 1). After 31 funding rounds of dilution, super-voting keeps him fully in control.
- Other major holders: Alphabet (~7%), Fidelity, Founders Fund, Sequoia, Andreessen Horowitz, EchoStar, sovereign wealth funds — plus, from the xAI side, Nvidia and the Qatar Investment Authority.
- Employees hold equity through stock-comp plans.
- You (the public) buy Class A — real economic exposure, virtually no governance power. If you don't trust Musk's capital allocation, there's no vote to change it.
The revenue map — every stream
2026 revenue is tracking ~$27–30 billion. Where it comes from:
| Stream | Scale (2026) | Notes |
|---|
| Starlink | The majority & the growth engine | SpaceX is "heavily reliant on Starlink for growth and profit" (CNBC). The number that matters most. |
| Launch services | ~$5–6B (20–25%) | Falcon cadence + Starship ramp. |
| Government / defense | ~$7B (Starshield ~$3.2B) | $22B cumulative federal backlog; 52 active contracts, $11.8B remaining. Space Force just added $2.29B (Space Data Network) + $4.16B (missile-tracking sats). The PLEO program ceiling jumped from $900M to $13B. |
| AI compute (xAI) | New — ~$26B/yr run-rate from two tenants | Anthropic: $1.25B/mo (Colossus, through May 2029, ~$40B+); Google: $920M/mo — confirmed Jun 2026: ~110,000 Nvidia GPUs at xAI data centers, Oct 2026–Jun 2029, ~$30B total. Together that's ~$2.17B/mo renting compute to its own AI rivals — SpaceX/xAI is now an AI-compute landlord. |
Fact vs spin — the Anthropic deal's durability. SpaceX's own S-1 describes Anthropic's payments running through May 2029. But Musk has publicly reframed the deal as short-term and cancellable (either side, 90 days' notice). Both can be "true" — but which framing you believe changes how much of that ~$40B you put in the valuation. We flag it; we don't pick the spin for you. The Google deal carries the same caveat: either side can walk after Dec 31 2026 on 90 days' notice, and Google can terminate if SpaceX misses its GPU-delivery milestone by Sep 30 2026. So both legs of that ~$26B are real — but neither is guaranteed.
Update — June 16, 2026: SpaceX is buying Cursor for $60B (all-stock). With xAI already folded in (Feb 2026), this closes the loop: SpaceX now owns the compute (Colossus), a frontier coding model (co-trained with Cursor), and the distribution (Cursor + Grok Build). It's no longer just renting GPUs to AI tenants — it's becoming a vertically integrated AI company. The honest caveat: $60B in freshly-IPO'd stock for one app is a rich, somewhat circular bet — paying for AI with paper whose value already assumes the AI works. Real strategic logic, real valuation risk. Both at once.
The hidden lever almost nobody prices: sovereign defense connectivity
The market prices what it can see: Starlink subscribers and launch cadence. What it can't see — and therefore under-prices — is Starshield, the government-only, end-to-end encrypted version of Starlink built for the NRO, Space Force, and classified agencies. It's the same constellation, sold as sovereign, secure comms — and that's a fundamentally stickier, higher-margin business than selling internet to households.
The pieces that are already public point at something much bigger than the line item suggests:
- Government/defense is tracking ~$7B (Starshield ~$3.2B), on a $22B cumulative federal backlog — and the Space Force just stacked on $2.29B (a military space-data network) and $4.16B (missile-tracking satellites) in a single month.
- The PLEO contract ceiling — the budget line these awards draw from — jumped from $900M to $13B. That's the government telling you how big it intends this to get.
- So much of it is classified that SpaceX's IPO filing reportedly required an unredacted version in a secured vault, readable by a handful of cleared people. The revenue you can't see is the tell, not the noise.
The multiplier the models miss — allied sovereign demand. Every close US ally faces the same problem: they want encryption-proof, resilient comms they control, and almost no one else can provide a 10,000-satellite secure network. That makes the addressable market not just the US defense budget but the US plus Europe, Australia, Canada, Japan, and a long tail of partner nations — each a sticky, classified, multi-year contract. Most valuation models can't include this because they literally can't see the contracts. It's the clearest example of the thesis underneath this whole company: connectivity is the core business, and the most valuable connectivity is the kind a government can't afford not to own.
The honest brakes on it: classified revenue is lumpy and unverifiable — you're trusting backlog figures you can't audit. Selling sovereign comms abroad runs into ITAR/export-control and political friction (and a CEO whose politics can complicate allied deals). And it's contested: Europe is funding its own sovereign constellation (IRIS²), and Eutelsat/OneWeb want the same buyers. The upside is real and under-priced — but it's optionality on top of the Starlink core, not a sure thing you can underwrite to the dollar.
The variables that move SPCX — the full lever board
You asked for every variable. Here's the honest board — what pushes the stock, which way, and why:
| Lever | Dir. | How it hits the stock |
|---|
| Index/passive inclusion | UP (near-term) | ~$22–27B forced buying, ~30% of float held by passives in 15 days. See the Aftermath piece. |
| Starlink subscriber + ARPU growth | UP | The core engine; the single most important number to track. |
| Starship reaching cheap reusability | UP | Collapses launch cost → unlocks bigger Starlink, more payload, off-Earth optionality. |
| New defense contracts | UP | Sticky, high-margin, validates Starshield as critical infrastructure. |
| AI-compute demand staying hot | UP | More tenants like Anthropic/Google = more of that ~$26B line. |
| First public earnings (the reveal) | BOTH | Real Starlink margins + Starship costs, finally disclosed. Re-rates hard either way. |
| Lockup expiry (~90–180d) | DOWN | Float multiplies; the early scarcity flips to a glut. |
| Anthropic/Google deals cancelled or not renewed | DOWN | That ~$26B compute line is concentrated in two tenants — and Musk himself calls one cancellable. |
| Starship failures / delays | DOWN | It's still a high-risk program; a major setback dents the whole thesis. |
| Key-man / Musk distraction or controversy | DOWN | 82% voting control = the company is Musk. Headline risk is balance-sheet risk here. |
| Macro / AI-bubble sentiment | BOTH | Markets already wobbled on AI-bubble fear ahead of the IPO; SPCX lists into that tape. |
| Regulatory (FAA, spectrum, antitrust, China) | DOWN | Launch licensing, Starlink spectrum, and self-dealing-with-tenants scrutiny are all live. |
The "Elon effect" — honestly
You're right that betting against Musk's execution has been a losing trade. It's a real, if unquantifiable, variable — worth naming on both sides:
The bull case for the man: a documented ability to compress timelines others call impossible (xAI's Colossus data center was stood up in a fraction of a normal build time), a magnet for top talent, and vertical integration that compounds cost and speed advantages. Markets pay a premium for that, and often it's been earned.
The other edge of the same sword: total key-man risk (82% of votes, no recourse), attention split across SpaceX, Tesla, xAI, Neuralink, The Boring Co. and X, recurring controversy/headline risk that moves the stock independent of fundamentals, and a governance structure that asks you to trust one person completely. "Don't bet against Elon" and "you have zero control if he's wrong" are both true at once.
The "Mars-Shot" pay package — how the control is actually built
Buried in the S-1 is the most ambitious compensation plan in corporate history — and the cleverest. Musk takes no traditional salary. Instead the board granted him 1.3 billion performance-based Class B super-voting shares (10 votes each), in two awards:
- 1 billion shares vest only if SpaceX hits 15 market-cap milestones climbing to $7.5 trillion and establishes a permanent Mars colony of at least 1 million people. Worth up to ~$1 trillion if it all hits.
- 300 million shares vest on 12 milestones ($1T–$6.6T) plus orbital data centers delivering 100 terawatts of compute per year — the off-Earth compute bet, written directly into his pay.
The governance fine print that matters more than the trillion: Musk votes all 1.3 billion shares immediately — even if he never earns a single one. The voting power is unconditional; only the money is performance-gated. That's a large slice of how the ~82–85% control is constructed. And the package is deliberately lawsuit-proofed after Delaware struck down his $56B Tesla award: it's disclosed in the prospectus itself (buying the IPO = accepting the terms) and SpaceX is incorporated in Texas, not Delaware. Read it both ways: incentives perfectly aligned with the most ambitious goals in capitalism — and a structure where you fund the mission while one person holds the controls regardless of whether the milestones ever arrive.
The wider Musk machine — where SPCX really sits
SPCX doesn't exist in isolation. It sits inside an interlocking machine worth seeing whole — because the pieces feed each other, and the biggest catalysts (and risks) live in the connections.
1. Terafab — the chip backbone (Tesla + SpaceX + xAI + Intel)
Announced March 2026, Terafab is a semiconductor mega-fab built by Tesla, SpaceX, xAI and Intel — initial ~$55B, up to $119B at full build. Intel is the primary manufacturing partner, bringing its advanced 18A process. The target is staggering: one terawatt of compute a year — roughly double current US capacity — for Tesla cars, Optimus robots, and AI data centers.
Why it matters: chips are the scarcest input in the whole AI race. Terafab is Musk vertically integrating his own chip supply — if it works, the empire stops depending on Nvidia/TSMC for the thing everyone is fighting over. It makes
Intel (INTC) a direct beneficiary, and it drags in the same
power-delivery bottleneck we mapped (a terawatt of compute needs a terawatt of power to run it).
2. The two-AI architecture (your instinct, made precise)
You spotted something real: the empire runs two different kinds of AI, and they're converging.
- Edge / real-time AI — Tesla FSD (cars) and Optimus (robots) run inference at the edge on Tesla's own chips (AI4, now AI5). Millisecond, physical-world decisions.
- Data-center AI — xAI's Colossus does the heavy training and centralized inference (Grok), then rents the spare capacity to Anthropic and Google.
The two are merging at the chip level: Tesla shut down its Dojo training supercomputer once "all paths converged to AI6," unifying on chips (AI5/AI6) that serve both edge and data center — and xAI takes ~30% of AI5 output. So the same Terafab/Intel chips power the cars, the robots, and the data centers. That's a full, self-supplied AI stack — chips → edge brains → data-center brains → the satellites (Starlink) that connect them. Nobody else owns all of it.
3. The Tesla–SpaceX merger — a real scenario (not a fact)
Label this clearly: it's a scenario, not a done deal. But it's a well-sourced, probability-weighted one — Musk has reportedly discussed folding the companies together. Wedbush's Dan Ives puts the odds near 80% by 2027; the Kalshi prediction market sits around 51–61%. A combined entity would be a ~$3.4 trillion empire.
The logic: Tesla's terrestrial AI (robots, cars, energy storage) + SpaceX's orbital reach (Starlink, space-based data centers) + the shared Terafab chip supply = one integrated AI-and-physical-world company. The catches: which company is the parent, how the stock swap is priced, shareholder fights, and the "self-dealing" criticism (it would be Musk's fourth billion-dollar deal between his own companies).
Why it matters for SPCX: if you buy SPCX you may be buying a future slice of a $3.4T everything-empire — or a company about to be re-priced and re-indexed in a contentious merger. Either way, the merger question is now part of the SPCX thesis, not separate from it.
Up- and down-stream: the logical next moves
- Upstream (what feeds it): chips/compute (Nvidia is now a shareholder), energy + grid for the data centers (the same power-delivery bottleneck we mapped), raw materials, and launch capacity.
- Downstream (what it enables): direct-to-cell Starlink, orbital/space-based compute (bypassing the terrestrial grid — see the off-Earth economy), more defense platforms, and Grok/AI products riding its own compute.
- Likely next steps: more compute tenants, a Starlink sub-IPO or segment disclosure, defense expansion, and using the IPO cash + stock to acquire or vertically integrate further.
The bear case — competition + the S-1's own risk factors
We've covered the optionality; honesty demands the other side. SpaceX's S-1 runs 38 pages of risk factors — including, by name, Elon Musk himself. The ones that matter most:
- It loses billions — for real. A $4.9B net loss in 2025 and a $4.3B net loss in Q1 2026 alone, against $29.1B of debt (much at variable rates). The $1.75T price sits on a deeply unprofitable, leveraged company — which is exactly why the S&P 500 can't touch it for years.
- It's really a Starlink bet. Roughly 60–70% of the enterprise value lives inside Starlink. If Starlink growth disappoints, the whole valuation deteriorates fast — launch and even AI compute are smaller slices of the price than the headline suggests.
- Competition is closing — right at IPO time. Amazon's Kuiper goes commercial mid-2026 (the same window as the listing), and OneWeb's revenue is up ~60% YoY. Amazon brings capital, cloud, devices and customer relationships — the one rival with the balance sheet to actually challenge Starlink.
- Everything hinges on Starship. The filing is blunt: any failure or delay in scaling Starship cascades into the V3 satellites, the mobile expansion, and the orbital-compute dream. It's still a high-risk program.
- Customer concentration. ~20% of 2025 revenue came from US federal agencies — exposed to politics and budget shifts.
- Chip dependency. SpaceX has no long-term contracts with any chip supplier — precisely why Terafab/Intel exists, but that's a bet, not a fix yet.
The balanced read: the bull case is enormous optionality; the bear case is a leveraged, unprofitable, Starlink-concentrated company facing real competition for the first time, with everything riding on one rocket program and one man — and a $1.75T price that already assumes much of the bull case. Both are true. That's why this is a conviction call, not a layup.
Short vs long-term trajectory
Short term (first ~6 months): dominated by mechanics, not fundamentals — the index-inclusion squeeze up, then the lockup supply flood, the first-earnings reveal, and high volatility throughout. Trade the schedule, not the story. (Full timeline in the Aftermath piece.)
Long term (years): the bet is whether the four engines compound — Starlink scaling, Starship working, defense entrenching, AI compute becoming a durable line — faster than the risks (key-man, tenant concentration, regulation, a $1.75T price that already assumes a lot) bite. It's a genuine moonshot conglomerate: enormous optionality, priced for a lot of it to go right, and impossible to value with a clean comp. The honest answer isn't a number — it's knowing exactly which levers to watch.
The whole picture, in plain English
We map the whole machine — not just the headline.
Dragonfly Lens breaks down every lever, sourced and hover-defined, and separates fact from spin — so you see the full board, not a hot take.
Join the Lens →
Sources: ownership / dual-class (Musk ~42% equity, 82.4% votes, S-1/A Jun 3 2026) — MEXC, KeepTrack; xAI merger (Feb 2026, ~$1.25T) — CBS News; Anthropic $1.25B/mo (Colossus, through May 2029) — TechCrunch, DCD; Google $920M/mo — TechCrunch; Musk reframing the deal as short-term — TechCrunch; "Mars-Shot" pay package (1.3B Class B shares, 15 tranches to $7.5T + 1M-person Mars colony, 300M shares on 100TW orbital compute, immediate voting of unvested shares, Texas incorporation / prospectus disclosure) — S-1 exhibit (SEC), Fortune, NBC News; revenue ~$27–30B + Starlink reliance — CNBC, NextBigFuture; defense contracts ($2.29B, $4.16B, $22B backlog, PLEO $13B) — SpaceNews, Fed-Spend; Terafab (Tesla/SpaceX/xAI/Intel, ~$119B, Intel 18A, 1 TW target) — EE Times, TechCrunch; Tesla–SpaceX merger speculation + odds (Wedbush ~80%, Kalshi ~51–61%, ~$3.4T) — CNBC, Fortune, Motley Fool; two-AI architecture / AI5 / Dojo — Applying AI, TechCrunch; S-1 risk factors, financials ($4.9B FY25 + $4.3B Q1'26 losses, $29.1B debt), Starlink concentration, Kuiper competition — Yahoo Finance, Trending Topics (S-1 breakdown).
Educational research, not personalized investment advice. Dragonfly Lens is not a registered investment advisor. Facts as of June 2026; figures are from public reporting and the S-1/A — verify against primary sources before acting. SPCX is named to explain the business, not as a buy or sell recommendation. Past performance does not guarantee future results.