SpaceX Bought a Cash Furnace and Turned It Into a Landlord. Now the Market Has to Price the Rent. | 07-10-2026
SpaceX Bought a Cash Furnace and Turned It Into a Landlord. Now the Market Has to Price the Rent.
Eight months after absorbing xAI, the merged company is the most heavily contracted compute lessor on earth, the most capital-hungry name in the Nasdaq-100, and up 16% in three sessions. A full reading of what was bought, what it has become, and how to trade it from tomorrow morning to next October.
View chart data (USD)
| Date (2026) | Daily close (USD) |
|---|---|
| Sep 8 | 153.47 |
| Sep 9 | 147.55 |
| Sep 10 | 148.18 |
| Sep 11 | 151.21 |
| Sep 14 | 148.15 |
| Sep 15 | 143.49 |
| Sep 16 | 150.88 |
| Sep 17 | 154.81 |
| Sep 18 | 152.71 |
| Sep 21 | 151.85 |
| Sep 22 | 154.72 |
| Sep 23 | 148.36 |
| Sep 24 | 148.03 |
| Sep 25 | 148.68 |
| Sep 28 | 145.47 |
| Sep 29 | 149.24 |
| Sep 30 | 150.86 |
| Oct 1 | 148.07 |
| Oct 2 | 158.96 |
| Oct 5 | 171.09 |
| Oct 6 | 171.94 |
SpaceX shares closed at $171.94 on Tuesday, sixteen percent above where they stood on Thursday. Three Falcon launches inside thirteen hours started the move on Friday. Monday added 7.6% as Morgan Stanley put a $300 target on the stock and reports circulated of talks to lease AI capacity to Microsoft. Then, after Tuesday’s bell, the Financial Times reported the company is seeking roughly $40 billion of debt, about $10 billion in bank loans and $30 billion in investment-grade bonds arranged by Apollo, to buy Nvidia chips. The stock slipped about 1% in late trading. Those two halves of one week, operational excellence followed by another enormous cheque, are the SpaceX story in miniature, and both trace back to a decision taken on 2 February.
That was the day SpaceX absorbed xAI in an all-stock deal that valued the rocket company at about $1 trillion. Reports of xAI’s price inside the transaction range from $125 billion to $250 billion, which is itself a comment on how hard the asset was to value. At the time it looked like a rescue. Eight months on it looks like something stranger and more interesting: a satellite operator that has become one of the largest sellers of AI computing power in the world, to the very companies its own chatbot competes with.
Part oneTwo companies, two very different problems
SpaceX before the deal
By the end of 2025 SpaceX had built two businesses that worked. Falcon 9 had made orbital launch routine, flying at a cadence of roughly 41 launches a quarter, more than the rest of the world combined. Starlink had turned that launch advantage into a consumer product: $11.4 billion of revenue in 2025 from a constellation nobody else could afford to put up. Group revenue for the year was $18.7 billion.
What SpaceX lacked was a way to pay for what came next. Starship, the fully reusable heavy lifter on which every long-range plan depends, was still in flight test. Employees and early backers had waited more than two decades for liquidity that tender offers only partly supplied. The company was large, admired and private, with a next phase that needed public-market money.
xAI before the deal
xAI was founded in 2023 as Elon Musk’s answer to OpenAI. It moved fast in one respect: Colossus, its Memphis supercomputer, went from empty factory to operating cluster in about four months in 2024 and kept expanding. In March 2025 it merged with X, the social network, which gave its Grok chatbot a distribution channel and a live data feed.
The economics were another matter. Bloomberg reported in mid-2025 that the company was burning around $1 billion a month. SpaceX’s own filings later showed the AI unit lost $3.2 billion in 2025 on $3.2 billion of revenue, then lost a further $2.5 billion in the first quarter of 2026 alone on $818 million of revenue. Grok remained behind ChatGPT, Gemini and Claude in usage. xAI had world-class hardware, a second-tier product and a funding need that grew every quarter.
Part twoWhy they merged, without the press-release language
The official framing was a “vertically integrated innovation engine”. The practical logic had three parts, and it is worth being plain about each.
First, xAI needed a balance sheet. A company losing $2.5 billion a quarter cannot keep raising private rounds indefinitely. SpaceX had Starlink’s cash flow and, more important, a path to the largest IPO ever attempted.
Second, SpaceX needed a bigger story. A launch and broadband company, however good, does not support a multi-trillion-dollar valuation. The prospectus claimed a total addressable market of $28.5 trillion, about 90% of it attributed to AI. Without xAI inside the perimeter, that number could not have been written.
Third, the two sets of assets do connect. Compute needs power, cooling and land, all of which are getting scarce on the ground. SpaceX has asked regulators for permission to fly up to a million satellites as orbital data centres, and it owns the only vehicle that could plausibly launch them at an acceptable cost. That plan is years from revenue. It is also a plan nobody else can even propose.
Part threeWhat the combined company actually is today
The second-quarter report, the first as a public company, is the cleanest picture available. It shows three businesses at three different stages of life.
| Segment · Q2 2026 | Revenue | Growth | Operating result | What it tells you |
|---|---|---|---|---|
| Connectivity (Starlink) | $4.29bn | +66% | +$1.66bn | The profit engine. 12.0 million subscribers, doubled in a year; a 39% operating margin. ARPU is $66 a month, down from $85, as growth comes from cheaper markets. |
| AI (compute, Grok, X) | $2.56bn | +247% | −$1.26bn | The growth engine and the cash drain. 1.4 GW of compute capacity. Most of the growth is capacity sold to others, not Grok. |
| Space (launch) | $962m | +29% | −$542m | The enabler. 38 launches in the quarter. Loss-making because it carries Starship development. |
| Group | $7.8bn | +92% | −$541m net | Capex of $18.4bn in the quarter and $23.6bn in the half. Cash of $93.5bn. Backlog of $47.5bn. |
The line that matters most is the one the market discovered slowly: the AI segment’s revenue is mostly rent. Anthropic agreed to pay $1.25 billion a month for the entire capacity of Colossus 1, about 220,000 Nvidia GPUs, on terms running to May 2029. Google agreed to pay $920 million a month for capacity equivalent to roughly 110,000 GPUs, at a reduced rate through September and the full rate from this month until 2029.
Put those together at full rate and SpaceX has about $26 billion a year of contracted compute revenue. Its entire group revenue in 2025 was $18.7 billion. In one stroke the merged company went from funding a chatbot to operating one of the most valuable pieces of digital real estate in existence, leased to two of the three labs that lead the field.
The third leg arrived in August, when the $60 billion acquisition of Cursor, the AI coding tool, closed. Grok never had a route into paying enterprises. Cursor is that route: a product developers already pay for, now attached to SpaceX’s own compute.
Part fourWhat becomes possible now
A compute lessor with no landlord of its own. Every other AI lab rents capacity from a cloud provider or builds slowly. SpaceX builds quickly, owns the site and sells what it does not use. The reported Microsoft talks and Tuesday’s $40 billion chip financing both point the same way: more capacity, pre-sold. One Bernstein estimate has AI revenue growing nearly fivefold over the coming year.
Orbital compute. If Starship reaches routine reuse, placing data centres in orbit, with continuous solar power and no grid connection queue, stops being a thought experiment. Two Starship test flights are pencilled in for this month. Each success shortens the distance between a slide and a product.
A full software stack. Cursor for developers, Grok for consumers through X, and owned compute underneath. The parts exist. Whether they become a coherent product is the open question of 2027.
Defence and government. Launch, a sovereign broadband network and domestic AI capacity under one roof is a combination governments buy. SpaceX already has the launch relationships; the AI capacity gives it a second product to sell into the same customers.
Starlink as the distribution layer. Twelve million subscribers and a per-user price that has been falling is a base onto which AI services can be bundled, in markets where no competitor has infrastructure at all.
Part fiveThe drawbacks, in order of seriousness
The rent can stop. Both the Anthropic and Google contracts are reported to carry 90-day termination clauses. A lease to 2029 that either tenant can exit in a quarter is not the same asset as a firm eight-year take-or-pay contract. The Google deal also reportedly hinged on a hardware delivery deadline of 30 September, with reduced payments possible if it was missed. The third-quarter report will show whether it was met.
The spending is ahead of the income. Capex in the second quarter was 2.4 times revenue. The $93.5 billion cash pile looks large until it is set against $23.6 billion spent in six months and a further $40 billion being borrowed for chips. This is a company that must keep raising to keep growing.
The customers are the competitors. SpaceX is selling its best capacity to Anthropic and Google while Grok tries to catch them. Every GPU leased out is a GPU Grok is not trained on. Management has chosen cash over product parity, sensibly, but it means the “frontier AI lab” part of the valuation rests on a product that is being deprioritised in practice.
The valuation leaves no room. At $2.25 trillion the company trades on roughly 72 times annualised second-quarter revenue. Morningstar’s fair value estimate at the time of listing was $780 billion; Aswath Damodaran’s was about $1.3 trillion. The spread between the lowest and highest published price targets, $75 to $800, is the widest of any mega-cap, and it says analysts do not agree on what kind of company this is.
Governance is one person. Mr Musk holds 82–85% of the votes through Class B shares. Public shareholders own an economic interest with no practical say.
The supply of stock is still arriving. Only about 5% of the company floated at the IPO. The rest unlocks on a published schedule, covered in the next section, and the largest pieces are still ahead.
Part sixWhy traders should have this on the screen every day
It moves like a small-cap and weighs like a giant. On Monday the Nasdaq Composite rose 1.05% and SPCX rose 7.63%. On 5 August it fell more than 10% after its first earnings call. A $2 trillion stock that travels that far in a session pulls the tech indices with it, and it entered the Nasdaq-100 under a fast-entry rule after only fifteen trading days.
It tells you about AI demand before the chipmakers do. SpaceX has said it will build exclusively on Nvidia’s platform. Its capex line and its financing are a direct read on next year’s Nvidia orders.
Its calendar is unusually precise. Most stocks give you earnings dates. This one gives you dated supply.
| Date | Event | Shares eligible | At $172 | Read |
|---|---|---|---|---|
| Fri 9 Oct | Day-120 lock-up release | 328.4m | ~$56bn | Lands two days after a 16% run. First real test of the rally. |
| October (TBC) | Starship flights 15 and 16 | — | — | Binary headline risk in both directions. |
| Sat 24 Oct | Day-135 release (effective Mon 26) | 328.4m | ~$56bn | Second tranche inside three weeks. |
| November (TBC) | Q3 results. Calendars show 3 or 17 Nov; unconfirmed | — | — | First full quarter of Anthropic at the full rate. |
| Results + 2 days | Earnings-triggered release | ~1.31bn | ~$225bn | The largest single tranche of the year, unconditional. |
| Tue 8 Dec | 180-day lock-up expiry | 797.6m | ~$137bn | The end of the standard lock-up. |
| 12 Jun 2027 | Founder block eligible | 6.4bn | — | Eligibility is not selling, but it is the largest overhang on the tape. |
Eligible shares are a ceiling on supply, not a forecast of sales. The first tranche, on 6 August, arrived a day after the low of the year. The September tranches passed with the stock flat between $143 and $155. What matters is the price at which holders are being asked to wait, and at $172 they are being offered 27% more than the IPO price to sell.
Part sevenGood investment or bad one
The case for owning it
- Starlink is a 39%-margin business growing 66% a year with no real competitor.
- About $26bn a year of compute revenue is contracted with two of the best credits in technology.
- Launch is a monopoly in all but name, and Starship would widen it.
- Q3 should show the first full quarter of Anthropic’s rate: $3.75bn from one customer.
- Thirty of 44 analysts rate it a buy; the average target is 27% above the price.
The case against
- 72 times sales, on a company that lost $541m last quarter.
- The marquee contracts can reportedly be exited on 90 days’ notice.
- Capex exceeds revenue by a wide margin and is now debt-funded.
- More than two billion additional shares become eligible to trade by 8 December.
- Independent fair-value estimates sit 40–65% below the market price.
The desk’s reading is that both cases are correct and they apply to different holding periods. As a business, SpaceX after the merger is stronger than SpaceX before it: the AI unit that was a liability now has tenants. As a stock at $172, after a 16% week and ahead of roughly $475 billion of newly eligible shares by early December, it is priced for the good version of every open question. That makes it a name to own on weakness created by the supply calendar and to avoid chasing on strength created by headlines. The setups below are built on that distinction.
SPCX projections
Desk scenario ranges from $171.94 · not forecasts of certaintyEach range is where the desk expects the bulk of outcomes to fall given the dated events inside the window. The stock has moved 7% or more in a session several times since listing, so the edges are reachable.
The debt report is not bad news for the business, but it reminds holders how much capital the plan needs, and it arrived with the stock extended 23% above its 50-day average. $175.50, the level that would have triggered a bonus share release in August, is the first ceiling. Friday’s 328m-share unlock discourages fresh buying a day early.
Friday’s tranche is the first to arrive with the stock well above the IPO price. If it is absorbed above $165, momentum buyers return and $180 comes into view. If not, the natural target is the $159–163 area the stock leapt from on Monday. A Starship flight inside the window adds a headline in either direction.
A second 328m-share release on 24 October, then the run-in to third-quarter results, which some calendars place on 3 November. The quarter itself should be strong on AI revenue. The problem is that a good print is followed two trading days later by the 1.31bn-share release, so strength into results is likely to be sold.
The earnings-triggered tranche and the 8 December expiry together make about 2.1bn shares eligible. The August precedent was a low on the eve of the first release, then a recovery of more than a third within a week. The desk expects a similar shape: a drawdown into the supply, followed by a recovery once the standard lock-up is finished and Q3 has confirmed the compute revenue.
By April the market will have seen Q3 and Q4 with both leases at full rate, the first Cursor quarters, and whether new capacity was pre-sold. If the contracts hold, the consensus target near $219 becomes the centre of the range. If either tenant gives notice, the floor of the range is where the argument restarts.
The top of the range is Morgan Stanley’s target and assumes AI revenue compounds as Bernstein projects. The bottom is close to HSBC’s $115 and assumes a lost tenant or a funding squeeze. In between sits 12 June 2027, when the 6.4bn-share founder block becomes eligible. The desk’s central case is $200–230, reached unevenly.
Trade ideas and setups
SPCX, plus the two instruments it moves directlyEvery horizon has one SPCX setup and one setup in a directly linked instrument: the Nasdaq-100 (US Tech 100 on most CFD platforms), which SPCX now sits inside, or Nvidia, which SpaceX has named as its sole AI hardware supplier. Levels are conditional: if price does not come to the entry, there is no trade. Given SPCX’s daily range, the desk would risk no more than 0.5–1% of account equity on any single idea, which at these stop distances means small notional size.
Only if the stock pushes into the $175.50 area in the first half of the session. Extended, facing a financing headline and a Friday unlock, a first test of that level is more likely to fail than to break.
Invalid if a confirmed Microsoft agreement or a Starship success lands; close flat by the end of the session either way.
The index is at a record and SPCX was among the largest gainers in Monday’s advance. A one-day view on the index is a view on Fed minutes, not on SpaceX. Stand aside and use SPCX’s reaction to $165 as the tell for whether tech leadership is intact.
The zone is where Monday’s 7.6% jump began. A retreat there after Friday’s release would be supply being absorbed, with the Morgan Stanley call and the compute story still fresh. The stop sits under the top of September’s $143–155 range.
No entry before Friday’s open. Invalid on a daily close below $155.
A reported $40bn of financing earmarked for Nvidia hardware from a single buyer is a demand signal for 2027. Nvidia set a record at $238.90 on Monday; the idea is to buy a 3–4% retreat, not the high.
Invalid if the financing is reported as scaled back or delayed.
If the stock extends into the $180s ahead of the 24 October release and results, it will be offering locked-up holders a third more than the IPO price just as 1.6bn further shares approach eligibility. That is the best-defined short window on the calendar.
Cover before the Q3 print; do not hold a short through results. Invalid on a daily close above $193.
For traders long SPCX who do not want to sell it. An index at a record with a member facing this much supply is a cheaper way to carry protection than shorting a stock that can jump 8% on a launch. Size it against the SPCX position, not as a standalone view.
Index levels are estimates; set them from your platform’s price. Remove when the SPCX earnings release is past.
The earnings-triggered release and 8 December are known in advance, which means sellers and front-runners act early and the stock tends to bottom around the event, as it did the day before the 6 August release. Scaling in across the $138–158 band, around the 50-day average and the IPO price, buys the business at the point of maximum supply. Target 2 is just under consensus.
Invalid if either Anthropic or Google is reported to have served termination notice, at any price.
The same thesis with fewer moving parts. SpaceX’s Q3 capex figure, due with results, will be read directly across to Nvidia’s order book. Nvidia carries no lock-up calendar and no single-tenant risk.
Invalid if SpaceX guides capex lower.
A weekly close above this month’s likely high, with the standard lock-up finished, would show the new supply has found owners. That is when the consensus target becomes a reasonable objective and CLSA’s $250 a reasonable stretch. Trail the stop on the earlier tranches to their entry price.
Invalid if Q4 AI revenue falls quarter on quarter.
For traders who believe the SpaceX story but not the level of the market. The short index leg removes the part of SPCX’s move that is simply tech beta, leaving the company-specific return: contracts, capacity, Starship. Equal notional under-hedges a stock this volatile, so treat it as a partial hedge.
Open only after 8 December. Close both legs together.
The position is a claim on Starlink’s cash flow plus an option on compute. It is worth holding for a year only while the leases stand, so the real stop is contractual, not technical. Reduce ahead of 12 June 2027, when the founder block becomes eligible, and reassess after.
Exit regardless of price if a major compute tenant leaves and is not replaced within a quarter.
If SpaceX’s build-out continues, Nvidia is paid first and in cash. If SpaceX’s stock disappoints because it overpaid for growth, Nvidia has still booked the sale. For a twelve-month holder it is the more forgiving way to express the same view, and the desk would carry it larger than the SPCX position.
Invalid if SpaceX diversifies away from Nvidia hardware.
A better company than it was in January, at a price that already knows it
The merger was sold as a vision and was, at the time, closer to a bailout. What has happened since is more impressive than either description. xAI’s hardware has been turned into roughly $26 billion a year of contracted revenue, Starlink has doubled its subscriber base, and Cursor has given the AI unit a product enterprises pay for.
None of that makes $172 cheap. The contracts can reportedly be ended on a quarter’s notice, the spending runs well ahead of the income, and about 2.8 billion more shares become eligible to trade by 8 December. Traders should respect the business and distrust the price in the short run.
The plan that follows from this is simple. Do not buy this week’s strength. Use the October and November releases to build a position between $138 and $158. Hold it through the end of the lock-up into the spring, keep it small, and treat a lost tenant as the exit signal whatever the chart says.
Risk disclosure. This article is editorial analysis and market commentary. It is not investment advice or a recommendation to buy or sell any instrument. Projections and trade setups are the desk’s scenarios, not predictions, and can be wrong. CFDs are leveraged products; losses can exceed deposits and most retail accounts lose money trading them. SPCX has moved more than 10% in single sessions. Index levels marked “est.” are derived from earlier published prices and must be checked against a live quote. Lock-up figures are shares becoming eligible to trade, not expected sales. The Q3 results date is unconfirmed by the company. Contract terms are as reported in public filings and press coverage and may differ from the executed agreements.
Sources. SpaceX Q2 2026 results release; Investing.com (SPCX price history); MarketBeat (analyst consensus); Parameter (Morgan Stanley target); Yahoo Finance (lock-up coverage, $40bn financing report citing the Financial Times, market wrap); PurePowerPicks and StockAlarm (lock-up schedule); The Motley Fool (Q3 preview); Wikipedia (SpaceX IPO); Benzinga and Kalkine (index and Nvidia levels).
