Space Exploration Technologies
Elon Musk's rocket company, which also runs Starlink satellite internet and, since February 2026, his AI company xAI with the X social network.
The scores
Growth3/4 ↑
Starlink subscribers rose fivefold in 30 months on its own power, but Q2's jump leans on one AI renter worth 19.5% of revenue.
Business quality3/4 →
Reusable rockets give a measured cost and share lead in launch and broadband, but the AI renters can build their own and Amazon is arriving.
Financial position2/4 ↓
$60.6B of net cash covers about three quarters of $18B spending; the case for 3 is investment-grade notes with nothing due before 2031.
Execution record3/4 ↑
One public print beat on revenue and Starship reached orbit, but a record this short as a listed company caps the score.
Per share capture2/4 ↓
xAI and Cursor were paid in shares, EchoStar is owed 261.8M more, and a controlling holder decides any Tesla merger terms.
Potential2/4 →
Two very large markets, but Starship's V3 cadence, renters staying and power arriving must all go right, and little is firmly signed.
Risk profile2/4 →
Two customers are 37.8% of revenue, one renter can leave on 90 days' notice, and one person controls 82.4% of the votes.
Valuation1/4 ↑
86 times trailing revenue assumes a $7.62B yearly loss becomes lasting profit on rentals customers can end on 90 days' notice.
Evidence quality3/4 →
Segment figures and contract terms are filed, but the $100B pace and V3 timing are graded management claims and Google's terms are reported.
If it works 1.0× · if it fails you keep up to 17%.
The Verdict
We rate SpaceX unattractive. It is three businesses: reusable rockets that carry half the world's launches, Starlink internet, its only profitable part, and an AI business losing billions. The business is mixed, because the AI side's new revenue is computing rented to customers who can leave on 90 days' notice. The price assumes those rentals become lasting profit, so it is stretched, priced for perfection. As a bet it is fragile: if the AI business works, today's money is worth about the same; if it fails, under a fifth is left. Starlink is excellent, but the price asks for far more.
The long caseAI companies need more computing than data centres can supply, and satellites bring internet where cables never reach; SpaceX sells both, if rentals last and Starlink keeps growing.
The evidence under this - 8 claims
- The vendor's value of $1.99T uses the July share count without the Cursor shares, which lands in the same place by a different route.
- 1.30B of Musk's Class B shares vest only on a million-person Mars colony or space computing goals; leaving them out cuts the equity value to about $1.85T.
- Cash falls by about $18B a quarter on the CFO's own capex outlook, so net cash will not stay at $60.6B.
- If December 2026 revenue really runs at $100B a year, the value is about 20 times that pace, not 86 (C14).
- V3 satellites carry about ten times the capacity of today's, so Starlink's growth could accelerate rather than slow (C21).
- Demand for AI computing still exceeds supply: a further $6.7B of six-month contracts was signed in July (C14, C28).
- The pace is December's revenue times twelve, so one month's cloud fees, cancellable on 90 days, carry much of it (C42).
- Q2 2026 revenue of $7.81B is a $31B yearly pace; reaching $100B needs it to triple in two quarters.
- Management gives no formal guidance in any filing; the figures were said on a call.
- The first-quarter 2027 guide of $33B-$37B is a quarterly figure, equal to a $132B-$148B yearly pace, against $14.6B in Q3; the $500B-$1T figure for end-2027 is a stated possibility with no stated build.
- 2026-10-21editedThe Q3 call added a first-quarter 2027 revenue guide and a 2027 recurring revenue range to the management targets
The company
SpaceX builds rockets that land and fly again, and uses them mostly to launch its own Starlink satellites. Starlink sells internet from space to 12.0 million homes and small businesses, plus airlines, ships and governments, often where no cable reaches.
Since buying xAI in February 2026 it also owns the Grok AI models, the X social network, and huge data centres whose chips it rents to other AI companies such as Google.
The evidence under this - 6 claims
- 2026-10-20editedQ3 2026 release: 13.6M consumer subscribers and $69 a month, up from 12.0M and $66
How it makes money
Most of the money comes from Starlink subscriptions, $4.29B of Q2 2026's $7.81B. Launches for outside customers added $619M to $962M a quarter this year, and the AI business sells X advertising, Grok subscriptions and rented computing.
Only Starlink makes a profit. The rocket business spends its earnings developing Starship, and the AI business spends far more than it earns.
The evidence under this - 4 claims
Where the money comes from, stream by stream
| Stream | Share of sales (Q2 2026) | How it earns it |
|---|---|---|
| Starlink consumer | 31.8% | Monthly broadband subscriptions, mostly month to month, plus the dishes households buy. |
| AI solutions and infrastructure | 28.1% | GPU computing rented to Anthropic, Google and others, plus Grok and X subscriptions, data licences and Grok API access. |
| Starlink enterprise and government | 23.1% | Negotiated contracts for airlines, ships, armies and governments, including Starshield, and the Starlink Mobile service through phone carriers. |
| Launch services | 8.3% | Falcon 9, Falcon Heavy and Dragon flights sold to companies, NASA and the military. |
| X advertising | 4.7% | Ads sold on the X social network; down on a year earlier during a platform change. |
| Launch and development | 4.0% | Paid development work such as NASA's Moon lander and crew vehicles, recognised as the work is done. |
Who runs it
Elon Musk founded SpaceX in 2002 and still runs it while also running Tesla; Gwynne Shotwell has run operations since 2008 and Bret Johnsen the money since 2011. Musk controls about 84% of the votes, so outside shareholders cannot outvote him. Nobody gets a yearly bonus. Musk's big share award pays only if the company becomes worth far more and puts a colony on Mars; the finance chief's pays on cash profit before spending.
The evidence under this - 3 claims
Why it wins, and who could beat it
Why customers come back
Its rockets land and fly again, one booster 34 times, so each launch costs SpaceX far less than a rival pays to buy one. Most launches carry its own satellites, so Starlink gets built at cost.
An outside count by BryceTech shows the result: about half of all launches in 2025 and 85% of the satellites. Amazon, the nearest rival network, had about 400 satellites to Starlink's 9,600.
The evidence under this - 4 claims
- BryceTech's independent count agrees on launches (51% in 2025) but no outside body publishes mass to orbit by provider for the whole year.
- China landed a reusable rocket stage for the first time in July 2026, so landing boosters is no longer unique to SpaceX and Blue Origin.
The threat
Amazon's network is years behind, yet Delta has already chosen it over Starlink for about 500 planes from 2028, a buyer wanting a second supplier.
In AI, both big renters can supply themselves: Google sells its own AI chips, and Anthropic has over a gigawatt of Google's arriving in 2026. Google calls its SpaceX deal short-term bridge capacity.
The evidence under this - 4 claims
The rivals: AMZN, GOOGL, CRWV, RKLB, VSAT.
Who else does this, rival by rival
| Ticker | Rival | Where it competes | How far along | Priced against it |
|---|---|---|---|---|
| AMZN | Amazon Leo, Amazon's low-orbit network, with Delta and JetBlue signed for in-flight Wi-FiPriced against it: One segment only | Connectivity | Announced | One segment only |
| GOOGL | Google's own TPU systems and data centres, which also supply AnthropicPriced against it: One segment only | AI | On sale | One segment only |
| CRWV | CoreWeave and the hyperscalers renting GPUs on committed contractsPriced against it: One segment only | AI | On sale | One segment only |
| RKLB | Blue Origin's New Glenn and Rocket Lab's NeutronPriced against it: One segment only | Space | On sale | One segment only |
| VSAT | ViasatPriced against it: One segment only | - | On sale | One segment only |
How big it could get
How big this could get
The ITU, the UN's telecoms agency, counts 2.2 billion people still offline, though most live where a $66 monthly bill is out of reach.
AI is the other market. The International Energy Agency expects data centres to use about 945 terawatt-hours of electricity by 2030, more than double 2024's 415. SpaceX's computers drew 1.4 gigawatts in June 2026.
The evidence under this - 4 claims
- 2026-10-20editedQ3 2026 release restated the series as nameplate compute: 1.7 GW against 1.4 in June
- 2026-10-20editedQ3 2026 release: 13.6M consumer subscribers and $69 a month, up from 12.0M and $66
Who has signed
Anthropic pays $1.25B a month and Google will pay $920M a month, both into 2029, but either can leave on 90 days' notice once an early period ends. Firmer is a $47.5B backlog, including over $6B of Space Force work.
The evidence under this - 4 claims
- The July contracts' terms are not disclosed and may be firmer than the Anthropic and Google deals.
- A 90-day exit is a right, not a forecast: both customers say demand exceeds what they can build.
The results
What growth costs
Growing is making SpaceX less profitable for now. Spending on research and on new computers and rockets is rising much faster than sales, because the AI build and Starship are paid for up front. Only office and selling costs are shrinking as a share of sales.
The evidence under this - 2 claims
The last report (4 Aug)
Quarter ended June 2026, filed figures.
In Q3 2026, SpaceX's revenue rose 161% to $14.6B. Starlink revenue grew 22% on the quarter to $5.2B, with 13.6 million consumer subscribers, up 1.6 million. The AI business, now called Super Intelligence, made $7.6B and a $0.4B operating profit, after a $1.2B pay charge for the Cursor purchase.
Net income was $4.4B, helped by $1.0B of other income, including a $0.56B mark-up of SpaceX's Boring Company stake, and a tax credit. Equipment spending was $19.6B in the quarter, against $13.1B of nine-month cash from operations. The filing shows no computing customer gave notice, but one was 34.5% of sales.
What it did to the thesisIt met the first two parts of our test: Starlink kept growing and AI stayed profitable at the operating line. The filing, now read, shows no cancellation, yet over a third of sales came from one customer who can leave on 90 days' notice. The view stays Unattractive.
The quarter in detail
Sales growth from the previous quarter, filed figures.
Earnings per share $-0.09 against $-0.23 expected; $-0.09 as filed, with no adjustments.
The evidence under this - 7 claims
- 2026-10-20editedQ3 2026 release: 13.6M consumer subscribers and $69 a month, up from 12.0M and $66
- 2026-10-20editedQ3 2026 release restated the series as nameplate compute: 1.7 GW against 1.4 in June
- Grok and X subscriptions also grew $258M in the quarter, so not all of the improvement is rentals (S02).
- Demand for computing exceeds supply, so a customer that leaves may be replaced at the same or a higher price (C14).
- On the Q3 call Musk said SpaceX rents at roughly the spot rate on short contracts by choice, expects to rent about half its compute, says short contracts are renewing at higher dollars per watt and that demand would take the compute several times over; the CFO said new compute pays back in under a year. This is management's account, with no customer named and no renewal figure.
- 2026-10-21editedThe Q3 call gave management's reason for short contracts and said renewals come at higher prices; counter-evidence added, inference unchanged
What management did with the money
It raised about $85.7B selling shares and $25.0B in bonds, then spent $60B of its own shares buying Cursor, a coding-software company. It ended June holding $100.0B of cash and securities.
Insiders in 2026Q3: 3 purchases and 9 sales, from their Form 4 filings.
The evidence under this - 3 claims
The price, and where we disagree
In short. Priced for perfection. An exceptional rocket and internet business, priced as though renting computing power for AI were already a lasting profit. It is not yet.
On 2026-09-28 Starship reached orbit for the first time and released 26 of Starlink's next, much larger satellites, the step the growth plan waits on.
The shares, listed at $135 in June, swung from $201.80 to $108.27 and back to $150.86. Management now promises a $100B yearly revenue pace by December, while spending about $18B a quarter.
The evidence under this - 4 claims
- The pace is December's revenue times twelve, so one month's cloud fees, cancellable on 90 days, carry much of it (C42).
- Q2 2026 revenue of $7.81B is a $31B yearly pace; reaching $100B needs it to triple in two quarters.
- Management gives no formal guidance in any filing; the figures were said on a call.
- The first-quarter 2027 guide of $33B-$37B is a quarterly figure, equal to a $132B-$148B yearly pace, against $14.6B in Q3; the $500B-$1T figure for end-2027 is a stated possibility with no stated build.
- 2026-10-21editedThe Q3 call added a first-quarter 2027 revenue guide and a 2027 recurring revenue range to the management targets
- The 10-Q says spending can be cut if data-centre needs ramp more slowly, so the plan is flexible by the company's own account.
- Nameplate compute was 1.4 GW in June 2026; 15 GW within 18 months has no precedent in the company's own record.
- The CFO said funding comes from operating cash, investment-grade debt markets and partnering with vendors and customers, and that the payback is short; no amount or lender was named.
- 2026-10-21editedThe Q3 call raised the fourth-quarter capital spending guide from about $18B to $35B-$40B; the earlier 'similar to Q2's $18.4B' and '15 to 20 gigawatts by end-2027' statements are superseded
How we read the multiple
Net of its cash, the business is valued at about $1.99T. Starlink, its only profitable part, earns $5.31B a year from operations, under 0.3% of that value.
So the price needs the AI business to turn a $7.62B yearly loss into lasting profit, mostly by renting computers to customers who can cancel on 90 days' notice, while Starlink keeps growing for years.
Against its own history. Its shares have traded only since June, so the older normal shown beside the price comes from before listing and says little.
The evidence under this - 4 claims
- The vendor's value of $1.99T uses the July share count without the Cursor shares, which lands in the same place by a different route.
- 1.30B of Musk's Class B shares vest only on a million-person Mars colony or space computing goals; leaving them out cuts the equity value to about $1.85T.
- Cash falls by about $18B a quarter on the CFO's own capex outlook, so net cash will not stay at $60.6B.
- If December 2026 revenue really runs at $100B a year, the value is about 20 times that pace, not 86 (C14).
- V3 satellites carry about ten times the capacity of today's, so Starlink's growth could accelerate rather than slow (C21).
- Demand for AI computing still exceeds supply: a further $6.7B of six-month contracts was signed in July (C14, C28).
- Grok and X subscriptions also grew $258M in the quarter, so not all of the improvement is rentals (S02).
- Demand for computing exceeds supply, so a customer that leaves may be replaced at the same or a higher price (C14).
- On the Q3 call Musk said SpaceX rents at roughly the spot rate on short contracts by choice, expects to rent about half its compute, says short contracts are renewing at higher dollars per watt and that demand would take the compute several times over; the CFO said new compute pays back in under a year. This is management's account, with no customer named and no renewal figure.
- 2026-10-21editedThe Q3 call gave management's reason for short contracts and said renewals come at higher prices; counter-evidence added, inference unchanged
Priced against its rivals
None of the listed rivals above sells the same thing to the same buyers as a large part of what it does. A price comparison needs at least two, so the price is judged against its own history instead.
Private or foreign rivals that cannot be priced: Blue Origin (New Glenn, TeraWave), United Launch Alliance, Arianespace, OpenAI and Anthropic.
The bet: what it pays if it works
This is a bet that SpaceX's AI computing business keeps its renters and grows. Even if revenue reaches the yearly pace management promised for December and stays there, the company would be worth only about today's value if buyers paid the price per dollar of sales the thesis would accept, so most of the success is already in the price.
If it fails, shareholders keep Starlink's profitable broadband and whatever cash is left, while the computers bought for renters who leave lose money and new shares pay for the rest.
| Revenue in two to three years | Computing chips rented out needed | At the stated multiple |
|---|---|---|
| $60.00B | about 835.5K at 100% kept | 0.6× |
| $100.00B (the big win) | about 1.7M at 100% kept | 1.0× |
| $124.00B (market forecast) | about 2.2M at 100% kept | 1.3× |
| $150.00B | about 2.8M at 100% kept | 1.6× |
Forecasters put world AI capacity near 130 GW by 2029; renting 3% of it is 1.7M chips, as many as the big win, and Starlink at 32M subscribers adds the rest, about $124B. On that forecast, $124.00B pays 1.3× from today's value: an Expensive bet on the market case.
To pay 2× (a Fair bet) it would need about $192.80B of revenue (about 3.7M computing chips rented out); to pay 5× (a Cheap bet), about $481.99B of revenue (about 10.0M computing chips rented out), at the same multiple.
The steps it has to take
- Done Cloud agreements with Anthropic and Google signed
- Done Starship reaches orbit and releases V3 Starlink satellites
- In progress Q3 2026 results: no cloud customer gives notice and AI's operating loss narrows below Q2's due 2026-11
- In progress Revenue running at a $100B yearly pace due 2026-12
- In progress About 1,000 V3 satellites in orbit due 2027-06
- Not started A cloud customer signs a firm multi-year commitment with no 90-day exit due 2027-06
- Not started Connectivity revenue grows at least 40% over four quarters with V3 in service due 2027-12
- In progress 15 to 20 gigawatts of data-centre power installed due 2027-12
- Not started The AI segment earns an operating profit due 2028-06
The detail behind the label
The evidence under this - 8 claims
- The pace is December's revenue times twelve, so one month's cloud fees, cancellable on 90 days, carry much of it (C42).
- Q2 2026 revenue of $7.81B is a $31B yearly pace; reaching $100B needs it to triple in two quarters.
- Management gives no formal guidance in any filing; the figures were said on a call.
- The first-quarter 2027 guide of $33B-$37B is a quarterly figure, equal to a $132B-$148B yearly pace, against $14.6B in Q3; the $500B-$1T figure for end-2027 is a stated possibility with no stated build.
- 2026-10-21editedThe Q3 call added a first-quarter 2027 revenue guide and a 2027 recurring revenue range to the management targets
- The July contracts' terms are not disclosed and may be firmer than the Anthropic and Google deals.
- A 90-day exit is a right, not a forecast: both customers say demand exceeds what they can build.
- The 10-Q says spending can be cut if data-centre needs ramp more slowly, so the plan is flexible by the company's own account.
- Nameplate compute was 1.4 GW in June 2026; 15 GW within 18 months has no precedent in the company's own record.
- The CFO said funding comes from operating cash, investment-grade debt markets and partnering with vendors and customers, and that the payback is short; no amount or lender was named.
- 2026-10-21editedThe Q3 call raised the fourth-quarter capital spending guide from about $18B to $35B-$40B; the earlier 'similar to Q2's $18.4B' and '15 to 20 gigawatts by end-2027' statements are superseded
Where we disagree with the market
One company owning rockets, Starlink and AI computing, growing 92% and heading for $1T of revenue.
Starlink is excellent, but most new AI revenue is rented computing customers can leave on 90 days' notice.
Most analysts are positive, and management talks of $1T of revenue by 2030. The common view treats the computing deals as signed growth.
We read SpaceX's own quarterly filing, which says the deals can end on 90 days' notice, and Google's own words calling its deal short-term. We would be wrong if customers turn them into firm multi-year contracts.
What analysts say, and the evidence
The evidence under this - 6 claims
- The pace is December's revenue times twelve, so one month's cloud fees, cancellable on 90 days, carry much of it (C42).
- Q2 2026 revenue of $7.81B is a $31B yearly pace; reaching $100B needs it to triple in two quarters.
- Management gives no formal guidance in any filing; the figures were said on a call.
- The first-quarter 2027 guide of $33B-$37B is a quarterly figure, equal to a $132B-$148B yearly pace, against $14.6B in Q3; the $500B-$1T figure for end-2027 is a stated possibility with no stated build.
- 2026-10-21editedThe Q3 call added a first-quarter 2027 revenue guide and a 2027 recurring revenue range to the management targets
- Grok and X subscriptions also grew $258M in the quarter, so not all of the improvement is rentals (S02).
- Demand for computing exceeds supply, so a customer that leaves may be replaced at the same or a higher price (C14).
- On the Q3 call Musk said SpaceX rents at roughly the spot rate on short contracts by choice, expects to rent about half its compute, says short contracts are renewing at higher dollars per watt and that demand would take the compute several times over; the CFO said new compute pays back in under a year. This is management's account, with no customer named and no renewal figure.
- 2026-10-21editedThe Q3 call gave management's reason for short contracts and said renewals come at higher prices; counter-evidence added, inference unchanged
- If December 2026 revenue really runs at $100B a year, the value is about 20 times that pace, not 86 (C14).
- V3 satellites carry about ten times the capacity of today's, so Starlink's growth could accelerate rather than slow (C21).
- Demand for AI computing still exceeds supply: a further $6.7B of six-month contracts was signed in July (C14, C28).
What could go wrong, and how we would know
The risks
- Renters can walk away. One AI customer was 19.5% of Q2 2026 revenue and can end its contract on 90 days' notice.
- Spending outruns the cash. About $18B a quarter of equipment spending would use the $60.6B of net cash within about a year.
- One person decides. Musk holds 82.4% of the votes, and paid for xAI and Cursor with new shares.
- Starship has flown once to orbit. Starlink's next satellites need it to fly often, and an engine failed on the first orbital flight.
- A rival network arrives. Amazon's network starts service this year and has already won Delta.
The evidence under this - 8 claims
- The 10-Q says spending can be cut if data-centre needs ramp more slowly, so the plan is flexible by the company's own account.
- Nameplate compute was 1.4 GW in June 2026; 15 GW within 18 months has no precedent in the company's own record.
- The CFO said funding comes from operating cash, investment-grade debt markets and partnering with vendors and customers, and that the payback is short; no amount or lender was named.
- 2026-10-21editedThe Q3 call raised the fourth-quarter capital spending guide from about $18B to $35B-$40B; the earlier 'similar to Q2's $18.4B' and '15 to 20 gigawatts by end-2027' statements are superseded
- Flight 14 carried 26 V3s, not 60, and landed nothing; Starship has not yet been caught or reflown in service (C20).
- SpaceX has said Starship would carry payloads to orbit 'in the second half' of earlier years before; the V3 plan rests on a vehicle with one orbital flight.
- On the Q3 call Shotwell said 'a few hundred' V3 satellites are needed before users feel the difference and that this can be reached by mid-2027, below the roughly 1,000 by Q2 2027 the claim carries.
- 2026-10-21editedThe Q3 call put V3 satellites at a few hundred by mid-2027, below the roughly 1,000 the claim records
What we still do not know
- Whether Anthropic and Google keep renting after they are free to leave; neither has said.
- How many Starlink customers leave each year, and what a dish costs SpaceX to supply; it does not publish either.
The evidence under this - 3 claims
- 2026-10-20editedQ3 2026 release: 13.6M consumer subscribers and $69 a month, up from 12.0M and $66
What would change our view
- Unattractive→NeutralRevenue catches up with the price
It needs:
- ○The value of the business falls to about 20 times last year's revenue, because revenue catches up or the price falls.about ≈$35.00 today, 77% below the last close
- Unattractive→NeutralThe AI revenue becomes firm and Starlink delivers V3
It needs all of:
- ○A computing customer signs a firm contract for several years with no 90-day way out.watched as it happens
- ○Starlink revenue grows at least 40% over a year, with the new satellites working.checked at the next results, due in Feb 2027
We would also be wrong if
- A computing customer gives notice to end or shrink its contract.
- Fewer than 500 of the new Starlink satellites are in orbit by June 2027.
What is coming
- 9 October 2026 Lock-up releases of up to 328.4M shares each: How much early holders sell into the market on 2026-10-09 and 2026-10-24, a test of demand at today's price rather than of the business.
Every dated event ahead
- 9 October 2026 Lock-up releases of up to 328.4M shares each: How much early holders sell into the market on 2026-10-09 and 2026-10-24, a test of demand at today's price rather than of the business.
- November 2026 Q3 2026 results, with up to 1.3B shares released two trading days later: Whether Starlink keeps growing, whether any cloud customer gave notice, and whether AI's operating loss narrows below Q2's $1.26B.
- 12 November 2026 OpenAI ends Cursor's model access: Whether Cursor, bought for 391.0M shares, keeps its users on Grok and other models once OpenAI's models are withdrawn.
- December 2026 Revenue pace target and Google's exit window: Whether revenue runs at the promised $100B yearly pace by December, and whether Google stays once it may leave on 90 days' notice after 2026-12-31.
- June 2027 About 1,000 V3 Starlink satellites in orbit: Whether Starship flies often enough to deliver V3 capacity on management's schedule, which the Starlink growth case needs.
Since we wrote this
- 26 OctNewsMusk says SpaceX raised $40 billion of debt for AI chips; no terms given. View heldMusk says SpaceX secured $40 billion of debt to buy AI chips.
- 23 OctEarningsThe 10-Q showed no cloud notice; one customer is 34.5% of salesThe filing confirms the backlog and the 90-day exit on cloud contracts. The $561M Boring Company gain is a mark-up, not business profit. The view holds.
- 21 OctEarningsThe call doubled the spending plan and confirmed Tesla merger committeesSpaceX guided fourth-quarter spending to $35B-$40B and first-quarter 2027 sales to $33B-$37B. Both fit the view; the 10-Q still has to show what the backlog can cancel.
- 20 OctEarningsQ3: revenue up 161%, first AI operating profit; view holdsStarlink grew 22% on Q2 and AI turned a profit on rented computing, but no contract became firm and the price was not re-struck; the 10-Q is still to read.
- 9 OctNewsSpaceX agrees to buy low-band wireless spectrum from Grain, price not disclosed. View heldSpaceX agreed to buy 14 MHz of nationwide 800 MHz spectrum from Grain Management, terms undisclosed.
- 8 OctNewsSpaceX's reported $40B chip borrowing is still unconfirmed; FCC cleared 15,000 more satellites. View heldReported $40B Nvidia-chip financing, and an FCC order for 15,000 direct-to-device Starlink satellites.
- 5 OctThesis reviewThe bet now names what it ridesThe owner asked that every bet name the macro trend behind it, after BWXT's bet was missing its chief executive's case for nuclear over gas.
- 4 OctThesis reviewOwner's checklist: revenue streams, cost of growth, management, catalystsThe owner's checklist found the method silent on who runs the company, what growth costs, where the money comes from and what is dated ahead; this pass fills those four blocks from the prospectus and filings.
- 2 OctNewsSpaceX launched Google's test AI chip as one of 100-plus outside payloads. View heldGoogle's first AI chip reaches orbit on a SpaceX rideshare rocket.
- 1 OctThesis writtenUnattractive, medium convictionAn excellent rocket and Starlink business priced as if rented AI computing were permanent profit.
Also checked 6 times with nothing new to act on.
Sources 33
Sources - 33
Every claim above cites one of these, or states the reasoning that derived it. The quote is frozen at the date it was read: a link can rot, and an agent fetch returns a reading of a page rather than the page. Backs counts the whole claim ledger, so it can name a claim this page does not surface.
- S01primarycompany filing2026-06-12the three segments, FY2023-FY2025 and Q1 2026 results, Starlink subscribers and ARPU by year, the Anthropic cloud agreement and its 90-day exit, launch record, the competition section, Musk's voting control and restricted Class B shares, the Terafab frameworkbacksC02C04C06C07C08C09C10C21C24C30C40C47C48
- S02primarycompany filing2026-08-04segment results, customer concentration (Customer A 18.3%, Customer B 19.5%), backlog of $47.5B, debt and the Valor related-party leases, the EchoStar spectrum terms, the new cloud-services risk factor, cash flow, share counts on the coverbacksC01C05C07C08C09C12C15C17C22C25C29C31C32C37C38C39C48C53C54C58
- S03primarycompany release2026-08-04segment capex, Starlink subscribers and ARPU by quarter, $14.1B of contracted cloud sales and its definition, $6B of Starshield awards, launch countsbacksC01C05C06C07C08C09C13C15C29C37C38
- S04primarytranscript2026-08-04Q2 2026 earnings call transcript (FMP)data/cache/transcripts/SPCX-2026Q2.jsonmanagement's $100B December run-rate, $1T revenue by 2030, capex of about $18B a quarter for two more quarters, 2 GW of compute by year end, V3 critical mass by Q2 2027; eight analysts' questions are a separate origin
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
backsC13C14C16C21C33 - S05primarycompany filing2026-08-14389,289,254 plus 1,752,426 Class A shares issued for Cursor, and 29.1M RSUs and 44.4M options assumedbacksC02C31
- S06primaryindustry data2026-01launches by provider and upmass by provider, counted by an analytics firm outside the company
In Q4, SpaceX had the most orbital launches, followed by CASC
backsC03 - S07secondarynews2026-04-10BryceTech's full-year 2025 count: 165 SpaceX launches of 325, and 85% of satellites launched
SpaceX dominates here, completing 165 orbital launches in 2025 - accounting for nearly 51% of the global total of launches.
backsC03 - S08secondaryvendor data2026-10-01Research pack, market tier (FMP: last close 2026-09-30, daily closes since listing, analyst split)content/SPCX/research-pack.jsonthe $150.86 close, the daily closes since the 2026-06-12 listing, the analyst split and its driftbacksC32C33C34
- S09secondarynews2026-09-28Flight 14 reached orbit, deployed 26 V3 satellites, lost one of six ship engines and was cut from 10 hours to 3
During its flight, SpaceX deployed 26 Starlink V3 next-gen satellites into its constellation of thousands of internet satellites that provide broadband access worldwide.
backsC20 - S10primarycompany release2026-07-30the size of Starlink's nearest rival constellation, and that its service begins this year
Completed four additional launches for Amazon Leo, its low Earth orbit satellite network, bringing the total constellation to nearly 400 satellites in orbit
backsC22 - S11secondarynews2026-03-31a large airline choosing Amazon Leo rather than Starlink, for about 500 aircraft from 2028
The initial Delta installation will see around half its fleet upgraded with advanced satellite connectivity starting in 2028.
backsC23 - S12primarycompany filing2026-08-12how a listed GPU-rental rival contracts: committed take-or-pay termsbacksC25
- S13primarycompany filing2026-07-23Google sells its own TPU computing systems to customers, and holds $94.1B of SpaceX sharesbacksC26
- S14primarycompany release2025-10-23SpaceX's largest cloud customer has another, larger supplier of computing coming online the same year
Anthropic will have access to well over a gigawatt of capacity coming online in 2026.
backsC26 - S15primarygovernment data2025-11-17people offline worldwide, and the urban-rural gap (85% against 58% online), from the UN agency
However, 2.2 billion people remain offline, down from a revised estimate of 2.3 billion in 2024.
backsC27 - S16primarygovernment data2025-04-10data centres used 415 TWh in 2024, growing about 12% a year since 2017, rising to about 945 TWh by 2030
Data centre electricity consumption is set to more than double to around 945 TWh by 2030.
backsC28 - S17secondarynews2026-06-05Google's own description of the contract, $920M a month from October 2026 to June 2029, about 110,000 GPUs, and a 90-day exit for either side after 2026-12-31
This is a short-term, timely agreement to ensure we have bridge capacity to meet surging customer demand for our agent platform, Gemini Enterprise, which has been even higher than we expected.
backsC11 - S18primarygovernment data2026-09-15FINRA consolidated short interest (Rule 4560), via scripts/api/finra.pyFINRA Consolidated Short Interest datasetshort interest and its history since listing, days to cover, share of floatbacksC35
- S19primarycompany filing2026-10-01Form 4 filings, open-market rows read for the Rule 10b5-1 flag (scripts/insider_fetch.py) SpaceX insiderssix open-market sales in the year, five of them (342,170 shares) Gwynne Shotwell's under a 10b5-1 planbacksC36
- S20primarycompany filing2026-08-10Neutron, the nearest listed medium-lift rival, has not flown; pad delivery targeted for Q4 2026backsC24
- S21secondarynews2026-08-25the $100B Starbase Louisiana plan, announced with the state; no SEC filing carries it
Construction is expected to begin in 2027, with the first launch targeted for as early as 2029, according to state officials.
backsC40 - S22secondarynews2026-09-29a Tesla merger is talked about and prepared for, and not agreed
Tesla still hasn't scheduled its 2026 annual meeting, and Elon Musk keeps teasing a SpaceX merger
read, but no claim rests on it - S23secondarynews2026-08-29the end date of 2026-11-12 and OpenAI's stated reason
OpenAI notified SpaceX on August 28 that it plans to wind down its contract with Cursor
backsC43 - S24primaryindustry data2026-04-02Synergy Research Group press release - Neocloud Market Forecast to Approach $400B by 2031, Driven by Surging AI Infrastructure Demand Synergy Research Groupan independent market tracker's count of 2025 GPU-rental (neocloud) revenue and its forecast to 2031; read through ir_fetch
New data from Synergy Research Group reveals that neocloud revenues are scaling at an unprecedented pace, reaching $9 billion in Q4-up 223% year-over-year-and exceeding $25 billion for full-year 2025. Driven by surging demand for AI infrastructure, Synergy forecasts the market will approach $400 billion by 2031, representing a sustained 58% compound annual growth rate.
backsC44 - S25primaryindustry data2025-12a second analyst firm's 2025 base and 2030 forecast for GPU rental by neoclouds, on a different definition from Synergy's; read through ir_fetch
ABI Research expects that by 2030, global revenue from GPUaaS offerings by neocloud providers are projected to reach nearly US$250 billion, at a Compound Annual Growth Rate (CAGR) of 43% (from US$42 in 2025).
backsC45 - S26primaryindustry data2025-01-28RAND Corporation - AI's Power Requirements Under Exponential Growth (RR-A3572-1), summary page RAND Corporationan independent research institute's extrapolation of global AI data-centre power demand in gigawatts, the unit SpaceX reports its build in; read through ir_fetch
Global AI data center power demand could reach 68 GW by 2027 and 327 GW by 2030, compared with total global data center capacity of just 88 GW in 2022.
backsC46 - S27primarycompany filing2026-08-11the largest listed GPU-rental rival's quarterly revenue, revenue backlog, active power and contracted powerbacksC49
- S28secondarynews2026-06-07an independent satellite analyst's bull and bear Starlink subscriber counts for 2030, and the article's estimate that today's network supports about 20 million broadband subscribers; read through ir_fetch
Farrar outlines a “Bull case” scenario in which Starlink could capture more than 45 million subscribers by the year 2030. Even the more pessimistic “Bear case” projection estimates that the satellite internet provider will secure a massive customer count exceeding 35 million by the end of the decade.
backsC50 - S29tertiarynews2026-06-04the IPO's lead underwriter's segment forecasts to 2030 as reported second-hand: AI $34.5B in 2027 and $322B in 2030, Starlink $144B and rockets $8.3B in 2030; the bank's note itself was not readable; read through ir_fetch
According to a report by The Financial Times, Goldman expects SpaceX's total revenue to rise 25-fold to $474 billion by 2030, up from $18.7 billion in 2025.
backsC51 - S30primarycompany filing2026-06-12executive officers and since-dates, the CD&A (no annual bonus, Musk's market-value and Mars or space-computing awards, the CFO's adjusted EBITDA options), beneficial ownership as of May 1, 2026, the lock-up release schedule, the FY2023-FY2025 income and cash flow statementsbacksC54C55C56C57C58C59
- S31primarycompany filing2026-10-20Q3 2026 segment revenue and operating income, Starlink consumer subscribers and revenue per subscriber, nameplate compute, capital spending by segment, the unaudited income statement, balance sheet and nine-month cash flow, the Cursor purchase terms and the 800 MHz spectrum agreement; unaudited, and the 10-Q and the call were not yet availablebacksC06C29
- S32primarytranscript2026-10-20fourth-quarter capital spending guide, the first-quarter 2027 revenue guide, backlog, the spot-rate rental approach, 2027 power targets, V3 satellite timing, Terafab dates, the Tesla special committees, and five analysts' questions
we expect fourth quarter capital expenditures to now be in the range of $35 billion to $40 billion for this quarter.
backsC14C16C21C60C61 - S33primarycompany filing2026-10-22customer concentration, backlog and its timing, the 90-day termination term on cloud contracts, segment results, the Boring Company gain, non-cancelable commitments, cash and debtbacksC62C63