Launch, Connectivity & Compute
Reusable launch as the cost floor under everything else — Starlink today, Starship-enabled markets next, and orbital compute as the newest bet.
Personal notes published for reference. Figures are sourced from company filings, releases, and press coverage as of the date above — the interpretation is mine and may be wrong.
SpaceX priced its IPO at $135/share on June 12, 2026, opened at $150, and closed day one at $160.95 — the largest IPO on record, raising $75B at an implied valuation of roughly $1.77T (about $2.1T at the day-one close). Roughly 30% of the offering was reserved for retail. Everything below is now written against SPCX as a direct public holding rather than a fund or SPV wrapper.
The strongest operational bench of the Musk-associated names — Gwynne Shotwell running day-to-day operations alongside Musk's technical direction. Unlike Tesla, the key-person risk is buffered by a president who has run the business through two decades of execution.
Layered, and each layer is large on its own: commercial and government launch, satellite broadband, and whatever Starship's cost-per-kilogram unlocks that isn't economic today. The last category is the one that can't be sized — which is either the appeal or the problem depending on your temperament.
Q3 (reported Oct 20) beat again: $14.6B of revenue (+161% YoY, vs. ~$12.7B consensus), $8.8B of adjusted EBITDA, and a first GAAP profit of $4.4B, with the $100B annualized revenue target for year-end reaffirmed. The catch is capex: $19.6B in Q3 against ~$9.6B of operating cash flow, and a Q4 guide of $35–40B. The company is now funding AI compute, Starship, and Starlink at a scale operating cash flow doesn't cover (section 05).
The best track record on this watchlist by a wide margin. Orbital-class reusability and a global satellite broadband network both went from thesis to working business inside a decade. When this company says it will do something hard, the base rate favors it.
The business that turned SpaceX from a launch provider into a consumer and enterprise connectivity company — and the reason the equity story isn't purely cyclical government contracting.
Vertical integration is the whole advantage: SpaceX builds the satellites and launches them on its own reusable rockets. No competitor gets to launch at internal cost. That structurally caps what anyone else can charge and still make money.
Beyond residential broadband — maritime, aviation, enterprise, and direct-to-cell. Each is a higher-ARPU segment than consumer, and each is a market where the incumbent alternative is expensive and bad.
Government and defense connectivity has become a meaningful line, with the strategic value of a sovereign-adjacent constellation now well understood by every defense ministry.
Subscriber growth against capacity, ARPU mix shifting toward enterprise/mobility, and how much of Starship's launch capacity gets consumed by next-gen Starlink satellites versus external customers.
On Oct 8, 2026, SpaceX agreed to buy Grain Management's nationwide 800 MHz spectrum, up to 14 MHz of paired low-band airwaves, for a reported ~$8B (the WSJ's figure; terms weren't disclosed). Low-band signals get through walls, which satellite-to-phone service can't do on its own. Together with the $17B EchoStar spectrum purchase, it moves Starlink Mobile from a backup partner for carriers toward a standalone wireless operator. AT&T, Verizon, and T-Mobile fell 5–8% on the news. Caveats: it needs FCC approval, and spectrum gives coverage but not capacity. A competitive ground network could cost tens of billions more. It's a new TAM layer, and another capex line on top of Starship and orbital compute.
The variable that makes the rest of the thesis non-linear. If full reusability at scale works, cost-per-kilogram to orbit drops enough that entire categories of business become viable.
Every space business case that doesn't close today — larger constellations, orbital manufacturing, lunar infrastructure, meaningful mass to Mars — fails on launch cost, not on physics or demand. Starship is a bet that removing that constraint creates markets rather than just serving existing ones. That's why it can't be sized with a spreadsheet, and why it's the single highest-variance input to any SpaceX valuation.
Next-generation Starlink satellites are the anchor customer — bigger, more capable satellites that Falcon can't economically loft. That gives Starship a captive internal demand source while external customers develop.
Development timelines have slipped before and will again. Rapid reflight cadence — not any single successful flight — is the milestone that actually matters, and it's the hardest one. Flight Test 14, the first orbital attempt with the new Block 3 vehicle (Booster 21 / Ship 41), flew on Sept 28, 2026 after slipping from ~Sept 22.
Starship reached orbit for the first time and deployed 26 Starlink V3 satellites, all of which checked in. That's the first operational payload Starship has carried and the first proof that the anchor customer (next-gen Starlink) can actually be served. It wasn't clean, though. Two engines shut down early, and SpaceX brought the ship home about seven hours ahead of plan after its first orbital checkpoint, splashing down north of Hawaii. Both primary objectives were met; reliability isn't routine yet. The stock fell 2% on the day (it had run up on the launch-date announcement) and recovered 2.6% the next session.
The newest and most speculative leg just went from a conceptual Musk-ecosystem overlap to a named contract: using launch capacity to put AI compute in orbit, with NVIDIA as the anchor partner.
Terrestrial AI datacenters are increasingly constrained by power and cooling, not chips. Orbit offers continuous solar power and radiative cooling — and if launch cost collapses, the economics of putting racks up there stop being absurd.
SpaceX and NVIDIA are partnering on Starmind AI-1, a payload built around NVIDIA's next-generation Rubin GPUs and Vera CPUs — "datacenter-class compute" in orbit. This replaces the earlier xAI-overlap thesis with a real, disclosed customer relationship, though it's still a single payload, not a fleet.
Better than pure optionality now that NVIDIA is named, but still early — this depends on Starship economics working first, then on orbital compute being genuinely cheaper than another terrestrial datacenter, a claim still far from settled. I'd move this from near-zero toward a small, real weight in a base case.
Each update has pulled the date forward. The May 2026 IPO prospectus said orbital compute launches "as early as 2028"; the Aug 4 earnings call said "next year"; in early September Musk put the first NVIDIA-powered AI satellites at Q4 2027 with "significant scale" in 2028 — about a year ahead of what investors were told at the IPO. The new Gigasat factory is aimed at ~1 GW/year of orbital compute by late 2027, and Musk's stated "tentative target" is 20 GW of power and cooling capacity online by end of 2027. Google also booked a SpaceX launch for its Project Suncatcher orbital-AI test.
ARK — a SpaceX holder and the most aggressive public bull on this leg — has framed orbital datacenters as a ~$300B opportunity that could eventually dwarf Starlink, and Musk has publicly endorsed ARK's Brett Winton's argument that the 100th orbital gigawatt could cost a third of the first. Musk has now gone further on X, suggesting the build-out could land faster than even ARK projects. The usual Musk-timeline discount applies — but on this company, not Tesla, the base rate for "faster than the bulls expected" is better than it has any right to be.
The terrestrial side — SpaceXAI's Colossus sites in Memphis — is where the speed is visible today. In a Sept 25, 2026 post, Musk laid out Colossus 2 at 110k GB200 plus 440k GB300 chips, with another 220k GB300 coming online the following week, another 220k in November, and possibly another 220k by late December "if we get lucky." That would more than double the NVIDIA chip count by year-end and hit the old 1M-GPU target ahead of schedule, backed by a 1.2 GW permanent power plant replacing temporary generation. Caveat: satellite imagery earlier in the year showed Colossus 2's cooling capacity well behind the headline claims, so I'd count chips when they're drawing power, not when they're announced.
On Oct 6–7, the FT, Bloomberg, and Reuters reported that SpaceX is seeking about $40B of financing to buy Nvidia chips: roughly $10B of bank loans and $30B of investment-grade debt, with Apollo expected to lead and PIMCO reportedly involved. Musk has said SpaceX will use Nvidia hardware exclusively for its datacenters. Two caveats: it's early-stage and Bloomberg says it could end without a deal, and it's debt, not equity, so it doesn't dilute the share count, though it adds interest cost to a company already spending $19.6B a quarter. The stock fell 2.5% on the report. The real change is that the buildout is now reaching for outside capital on top of the $75B IPO proceeds and operating cash flow.
SpaceX's second public report flipped the headline from a loss to $4.4B of net income, with AI now more than half of revenue. It also guided Q4 capex to roughly double Q3's.
Revenue was $14.6B against a $12.7B consensus, adjusted EBITDA was $8.8B (up 306%), and adjusted EPS was $0.45 against $0.13 expected. Q1 2027 revenue guidance of $33–37B is far above the ~$22.2B consensus. Backlog is $63B, including $17.5B of deferred revenue, with about $42B due to be recognized within a year. Management reaffirmed the $100B annualized revenue target for year-end and put out a $500B–$1T target for end-2027, which I'd discount heavily.
The Super Intelligence segment (xAI plus Cursor) had $7.56B of revenue, up 830% and 52% of the total, including $608M from Cursor after the Aug 14 close. Its operating income was $431M after $1.2B of Cursor stock compensation, or about $1.6B before it. Connectivity (Starlink) had ~$5.2B of revenue, up 69%, and still earns most of the profit: $2.37B of operating income at a 55% gross margin, which includes about 3 points of one-time tariff refunds. Consumer subscribers reached 13.6M (1.6M net adds) at $69 a month, and Starlink Mobile passed 22M monthly users. Space revenue was $1.8B, up 6% YoY, and that segment lost $7M. The thesis in sections 02–04 now describes a minority of revenue.
Operating income was $2.79B. The other ~$1.6B came from below the line: $1.03B of other income (including $561M from a related party, not itemized), a $390M tax benefit, and $198M of net interest income ($873M earned on roughly $90B of cash and securities from the IPO, against $675M of interest expense). The $2.79B is the number to track, and it still includes the tariff refunds. One write-up called the headline flattering, and I agree.
Q3 operating cash flow was about $9.6B against $19.6B of capex. Over nine months, operating cash flow was $13.1B against $48.1B of capex, a gap of about $35B, with AI capex alone at $16.4B in Q3 and $40.0B year to date. Q4 is guided to $35–40B. The CFO said funding would come from operating cash flow, multiple debt pools backed by an investment-grade balance sheet, and value-chain partnerships, which fits the reported ~$40B Apollo-led financing, though the call excerpt I saw didn't confirm it. At this pace the IPO cushion covers only a few quarters; that's my estimate, not a company figure.
Compute online was 1.7 GW at quarter-end, nearly 2 GW by end of October, and is targeted at 2.5–3 GW by year-end. Musk expects about 13 GW by end-2027 with a rising chance of more than 20 GW, and a 440,000-chip Vera Rubin cluster by the end of Q2 2027. Starship Flight 15 aims to catch both booster and ship. Management put a test chip around mid-2027 and volume production around Q3 2028, with wide error bars.
The stock closed at $147.79 before the report and rose roughly 8–12% on Oct 21 (quotes ran $159–165; I couldn't confirm the official close). JPMorgan raised its target to $500 from $240, Cantor to $345 from $246, Wells Fargo to $274 from $214, TD Cowen to $224 from $200, and Goldman to $245 from $230, while Morgan Stanley held $300. The average target is about $225 (Moderate Buy).
The furthest-out leg, and the one where TAM is least meaningful as a concept.
Starship's selection as a crewed lunar lander is a real, contracted government program — the concrete piece of this. It's revenue and it's a forcing function for Starship development.
Building on the Moon — permanent infrastructure, in-situ resource use, a fuel depot supporting deeper missions — is a decades-long proposition with no revenue model I'd underwrite today. It belongs in a thesis as a statement about ambition and talent retention, not as value.
I own this for Starlink cash flows and Starship optionality. The lunar and Mars ambitions are why the best engineers work there, which indirectly protects the first two. I don't pay for them.
This is the first quarter of public 13F data SpaceX has ever had — not a trend yet, but a useful snapshot of who showed up on day one and how much of the float rolled in from pre-IPO strategic holders versus fresh institutional buying.
There's no prior quarter to sell out of — every institutional position is new by definition. The number worth watching is next quarter's, once there's an actual base to compare against.
NVIDIA's stake predates the IPO as a strategic investor and simply converted into public shares. Atreides Management disclosed SpaceX as its single largest position at roughly $4.7B. Alphabet and Harvard Management also reported large stakes, though the disclosed dollar values vary meaningfully across data providers this early — treat any specific figure beyond NVIDIA and Atreides as directional until filings settle.
Whether the lockup expiration produces net institutional selling once shares are freely tradable, and whether the initial strategic/insider-adjacent holders (NVIDIA, Musk himself) are still counted alongside genuinely new institutional buyers in next quarter's filings.
Two public quarters, two beats: the first was followed by a near-14% single-day drop to an all-time low, the second by a pop of roughly 10%. The difference was the profit, and the capex guide that came with it.
If this plays out: ~$110
Revenue and EBITDA beat handily on the first public print, and the stock still fell almost 14% in a session to an all-time low near $108 — the market repricing SpaceX as a capital-intensive AI/infrastructure spender, not a pure launch-and-broadband compounder. Q3 capex of $19.6B was more than twice operating cash flow, and Q4 is guided to $35–40B, while the $4.4B headline profit includes ~$1.6B from below the line. It's funding a third speculative bet (orbital compute) on top of two that haven't fully proven out, with financing still to be raised. Flight Test 14 reached orbit but cut the mission short after two early engine shutdowns — a reminder that "reached orbit" and "reflies weekly" are years apart, and the capex is being spent as if the second one is close.
If this plays out: ~$180
SpaceX beat revenue and EBITDA by a wide margin in both public quarters and has absorbed three lockup expiries without a selloff — early evidence the supply-overhang fear is more theoretical than real. Starlink and launch remain the reason to own this, now with a Q3 showing AI can earn money too ($431M of operating income, $1.6B before Cursor stock comp). The risk is the cash burn that comes with it. Flight 14 reaching orbit with a real Starlink V3 payload was the data point the base case needed; NVIDIA's Starmind AI-1 and reflight cadence are the next ones. Expect a stock that trades on capex disclosures and Starship outcomes rather than settling into a range this early in its public life.
If this plays out: ~$300
Morgan Stanley's framing is the cleanest version: at today's price, the market largely pays for Starlink and launch, while the orbital-compute bet — now anchored by a named NVIDIA partnership on next-gen Rubin/Vera silicon — is priced at close to nothing. Flight Test 14 got to orbit; if reflight cadence follows, the cost-per-kilogram unlock that makes orbital datacenters, lunar infrastructure, and bigger constellations real all move from thesis to roadmap together. A management team with the best execution record on this watchlist, guiding Q1 2027 revenue to $33–37B against a ~$22B consensus, is the setup behind Morgan Stanley's $300 target. JPMorgan ($500) and Cantor ($345) now sit above that; I'm keeping my bull case anchored to $300 because it still assumes the capex gets financed.
It closed at $147.79 on Oct 20, ahead of the Q3 report, and traded up roughly 8–12% on Oct 21 (about $159–165; the official close is unconfirmed). That reverses most of the slide from the Oct 6 high near $176 to ~$146 on Oct 16, which came with the AI hardware selloff in the Nasdaq's worst week since June and the reported $40B debt financing. At ~$160, my bear case of ~$110 is about 30% below and the base case of ~$180 about 12% above. Before the slide, on Oct 9 it closed at $162.57 after holding above $160 all week, a level it hadn't closed at since early July, about 48% above its Aug 5 low. It also absorbed the Oct 9 unlock of up to 328.4M insider shares and still closed up 1.25% that day, the third lockup expiry it has taken without a selloff. Barclays is at Overweight/$254. The week began with a ~5–8% jump on Oct 5 on three things: Morgan Stanley's Adam Jonas reiterating Overweight/$300 ("cheap and getting cheaper" on growth-adjusted multiples), Musk confirming talks with TSMC to build and run a Texas fab supplying chips exclusively to SpaceX, Tesla, and xAI, and the AI unit being renamed SpaceXSI. After the Oct 21 target raises (JPMorgan $500, Cantor $345, Wells Fargo $274, Goldman $245, TD Cowen $224, Morgan Stanley $300), consensus averages about $225. Q3 changed the story from capex sticker shock to profit, but the Q4 capex guide is the same story at twice the size. The scenario targets above are my own rough, directional estimates — not formal price targets — anchored to the sell-side range rather than a full model.
Musk's attention is split across SpaceX, Tesla, xAI, and more. Shotwell mitigates this operationally but not strategically.
$19.6B of Q3 capex against ~$9.6B of operating cash flow, with Q4 guided to $35–40B and a nine-month gap of about $35B. It's funding AI compute on top of Starship and constellation replenishment, and the company is reportedly seeking ~$40B of outside debt for chips. A rate-hiking Fed with the 10-year near 5% makes that debt more expensive, and the plan depends on the AI business continuing to earn what Q3 showed.
On the Oct 20 call, Musk said both companies have formed special committees of independent directors to discuss a merger, called it a "good idea," and said it isn't a certainty; he can't participate as a conflicted party. Barclays expects SpaceX to acquire Tesla in a stock-for-stock deal. No terms exist, and Tesla approval would need 50% of all its outstanding shares. If it happens, SpaceX is the issuer: its holders would take on Tesla's balance sheet, capex, and key-person profile at a ratio set by committees of two Musk-controlled companies. Until there's a filing, treat it as a source of volatility, not value.
Launch licensing, spectrum allocation, and orbital debris rules are all policy-dependent. Competing constellations and national programs are funded and motivated.
Day-one close valued the company near $2.1T; the stock has since round-tripped to an all-time low near $108, back to a high near $176 on Oct 6, and down to roughly $146 by Oct 16. A further lockup unlocked ~319M shares around Sept 9, 2026 (the 90-trading-day mark) — the market absorbed the first one without much damage, but each event is a live supply test on a name that's already shown it can move 14% in a session.