Everything About AI
NextFin News - Elon Musk's SpaceX reported third-quarter revenue of $14.6 billion on Tuesday, a 161% jump from a year earlier and a decisive beat of the roughly $12.7 billion Wall Street had expected, as the rocket maker's bet on artificial-intelligence compute rental moved from a cash-burning experiment to the company's largest and now-profitable revenue pillar. The quarter marks the first full period in which the data centre contracts signed around SpaceX's June initial public offering — including agreements with Anthropic and Alphabet's Google — flowed through the income statement, and it flips the script on the scepticism that had followed the company's record-breaking public debut.
Revenue came in well ahead of the $12.72 billion consensus compiled by LSEG, with adjusted earnings per share of $0.35 more than double the $0.15 estimate and net income reaching $4.4 billion against an expected $1.88 billion. Adjusted EBITDA climbed 306% to $8.8 billion, and operating income swung to a $2.8 billion profit from a $273 million loss in the year-ago quarter. The market's verdict was measured: shares rose 1.6% in late trading and traded between $147.31 and $151.01 during the session, still well below the roughly $226 record set days after the June IPO and only modestly above the $135 offering price.
The Quarter in Numbers: AI Compute Overtakes Rockets and Satellites
The composition of that revenue is the real story. SpaceX's Super Intelligence division — the AI and compute business — generated $7.6 billion, ahead of the $6.3 billion forecast, making it larger than the Connectivity segment's $5.2 billion and the Space segment's $1.8 billion combined. Connectivity, which houses the Starlink satellite internet service, still grew, beating its $4.9 billion estimate, and the legacy Space business exceeded its $1.43 billion target. Connectivity operating income rose 99% year over year to $2.4 billion.
To appreciate the velocity of this pivot, consider the sequence. In the second quarter, reported in August, total revenue was $7.8 billion, up 92% year over year, and the AI division contributed $2.6 billion. Three months later, AI revenue nearly tripled to $7.6 billion while total revenue almost doubled again. The acceleration is exactly what the contract calendar implied: the Anthropic agreement, signed in May at $1.25 billion a month, was only partially recognised in the second quarter but ran for the full third quarter, and the Google deal, at $920 million a month, began ramping in late September.
This was not a gradual evolution; it was a step-change delivered by two contracts. In May, ahead of its public listing, SpaceX agreed to rent all of the GPU capacity at its Colossus 1 data centre to Anthropic for $1.25 billion a month — a deal covering more than 220,000 Nvidia chips. Weeks later it added a second agreement with Google at $920 million a month, securing 110,000 Nvidia GPUs from October 2026 through June 2029, with capacity ramping from late September and a termination right for Google if the hardware is not delivered. Together, the two deals put more than $2 billion a month of compute-related revenue on the books once fully ramped — an annualised run rate exceeding $26 billion from just two customers.
On the second-quarter earnings call in August, finance chief Bret Johnsen told investors the company had contracted an additional $6.7 billion of cloud services revenue "over a six month period that begins ramping starting in October of this year." The accounting consequence is mechanical and front-loaded: a $1.25 billion monthly contract annualises to $15 billion, and even recognising revenue as capacity is delivered, the third quarter was always going to show a discontinuity once those deals crossed from signed paper into live capacity. What surprised was not that AI revenue grew, but that it grew profitably — the Super Intelligence division's contribution pushed the whole company into a $2.8 billion operating profit, a roughly $3.1 billion swing from the year-ago loss.
The Mechanism: From Launch Provider to Vertically Integrated Compute Landlord
SpaceX is not selling a commodity. It is selling a scarce, vertically integrated bundle — Nvidia chips, power, cooling, and the data centre real estate to house them — at a moment when frontier AI labs and hyperscalers are competing for exactly that package. The company's advantage is structural: it builds its own rockets, launches its own satellites, and now owns its own data centres, giving it a cost position and deployment speed that few competitors can match. The $6.7 billion in contracted cloud revenue that begins ramping in October is not a forecast; it is a backlog, and backlogs do not mean-revert.
That is the hinge point of this quarter. Through the first half of 2026, SpaceX's operations produced about $3.5 billion of cash while capital spending ran $28.5 billion. The fear was that the AI build-out was a bottomless pit: capital expenditure of $18.4 billion in the second quarter, more than double the first quarter's $10.1 billion and far above the roughly $13 billion analysts had pencilled in, of which $15.83 billion went to AI. Tuesday's report suggests the spending is buying revenue that carries margin, not just megawatts.
On the August earnings call, the company said it expected a $100 billion revenue run-rate by December and promised a less than one-year payback on new capital deployments for AI compute.
That payback claim is the crux of the bull case. If new data centre capacity really earns back its cost in under a year, then today's capital intensity is a temporary bridge to a high-return franchise rather than a permanent drag. It also reframes the financing problem: a company that can demonstrate sub-12-month paybacks on AI infrastructure can raise debt on materially better terms than one spending into the void. The market will test that claim directly: SpaceX has already raised more than $150 billion across its IPO and debt markets in a matter of months, and reports indicate it is seeking another $40 billion of debt — about $10 billion in bank loans and $30 billion of investment-grade paper — to buy more Nvidia chips. Total debt stood at $38.4 billion as of June 30, up from $22 billion at the end of 2025, while cash and equivalents reached $93.5 billion at the end of the second quarter, up from $24.7 billion at year-end 2025.
Cyclical Wave on Top of a Structural Shift
The durable part of this story is structural — a genuine regime change in SpaceX's business model, from a two-legged stool of launch services and satellite subscriptions to a three-legged one that includes compute. But there is a cyclical layer on top, and it matters for how investors should read the multiple. The pricing power SpaceX enjoys today rests on an acute shortage of Nvidia GPU capacity, and that shortage is being addressed at pace. Every major cloud provider is adding gigawatt-scale clusters, and Anthropic itself has locked down roughly $517 billion of compute commitments ahead of its own listing, according to a September report. If supply catches up with demand faster than expected, the rental rates SpaceX can command on new contracts will compress even as its contracted backlog delivers.
So the correct framing is a structural shift in the business model overlaid with a cyclical pricing wave in AI infrastructure that will eventually flatten. The company's valuation should reflect the first; the risk discount should reflect the second. Blending the two into one verdict — either "AI is the future" or "this is a bubble" — misses the point. Both are true, on different clocks.
The cyclical overlay also shows up in the starkest possible form in the company's own guidance arithmetic. Management's $100 billion revenue run-rate target for December would require roughly doubling from the current pace in a single quarter. That is achievable only if at least one more hyperscaler-sized contract lands before year-end and ramps immediately — a big ask, but one consistent with the cadence of the past six months, in which two such deals were signed in a matter of weeks.
The Counter-Thesis: You Are Buying Capital Spending, Not Free Cash Flow
The strongest argument against the bull case is simple and quantitative. In the second quarter, SpaceX spent about $2.35 on capital projects for every $1 of revenue. Free cash flow remains deeply negative, and it will stay negative for as long as capex runs at many times revenue. Even with Tuesday's profitability turn, the company is not yet self-funding its ambition, and the gap is wide enough that it cannot be closed by operating improvements alone within the current investment cycle.
Sceptics also argue that the AI division's profitability may partly reflect accounting timing rather than unit economics. Much of the data centre capacity was already built and depreciated within the xAI business before it was folded into SpaceX; renting it out recognises revenue against sunk cost. The true test is whether new capacity — the billions being spent this quarter — earns an acceptable return once it comes online. If the answer is no, then this quarter's operating profit is the peak of a cyclical pricing wave, not the base of a structural one.
There is also concentration risk. A large share of the AI revenue traces to a small number of counterparties — Anthropic and Google — signed in a compressed window before the IPO. If either renegotiates, delays, or walks when contracts mature, the Super Intelligence division's $7.6 billion quarter becomes far less durable than the headline suggests. Google's own framing of its deal underscores the point: a company spokesperson described it as a "short-term, timely agreement to ensure we have bridge capacity" for its Gemini Enterprise platform, language that suggests the arrangement may be a stopgap until Google's own capacity comes online rather than a permanent tenancy.
The specific signal that would break the structural thesis is observable in the next report: if AI revenue fails to grow sequentially from $7.6 billion, or if capital expenditure remains above $18 billion a quarter while revenue growth decelerates below 50% year over year, the "profit engine" narrative gives way to the cyclical-overhang view. A secondary tell is the cost of new debt: if SpaceX's next financing comes at materially wider spreads despite the profitability turn, the credit market is voting that the payback claim is not yet proven.
What Comes Next: Beneficiaries, the Exposed, and Three Scenarios
The immediate beneficiaries of this quarter are SpaceX's existing shareholders and the creditors it will soon approach: a profitable, backlog-backed AI division makes the next debt raise cheaper than it would have been a quarter ago. The exposed parties are the private neocloud competitors — CoreWeave, Lambda and similar infrastructure players — who now face a public competitor with a $93.5 billion cash balance, in-house manufacturing, and a founder willing to price aggressively for share. Nvidia, as the supplier of the GPUs underpinning both major contracts, remains a second-order beneficiary of the build-out regardless of which tenant wins, and the same logic applies to the power and cooling equipment vendors that sit upstream of every new data centre.
In the short term, the muted +1.6% reaction reflects a market that has heard Musk's ambition before — including his August statement to employees that AI revenue would overtake all other SpaceX revenue "probably in September," a crossover this quarter's $7.6 billion AI figure against $7.0 billion from the other two segments appears to have achieved. Pivotal Research, forecasting $13.2 billion of revenue and $0.21 of earnings, raised its price target to $230 from $220, while the options market had been pricing in a 7.8% move in either direction — suggesting the beat was largely anticipated and that the real repricing will come from the guidance and the capex line, not the top-line surprise.
The base case for the medium term is continued sequential growth as the Google GPU cluster comes fully online in October and the $6.7 billion backlog ramps, with operating margins expanding as fixed costs spread across more contracted revenue. The upside case is that rental pricing holds and SpaceX reaches the $100 billion annualised run-rate targeted for year-end — a figure that would require roughly doubling from the current pace but is not implausible with two more hyperscaler-sized contracts. The downside case is that capex stays above $18 billion a quarter while GPU oversupply compresses pricing, forcing SpaceX to tap debt markets on deteriorating terms and turning the AI division from a profit engine into a funding sink.
Long term, the structural call stands: if AI compute demand proves durable through the late 2020s, SpaceX's vertically integrated model gives it a cost position few competitors can match, and the three-segment structure — space, connectivity, compute — is more resilient than any single line of business. The Connectivity segment, with more than 12 million Starlink subscribers and an average revenue per user of $66 a month, provides a recurring-revenue floor that pure-play compute landlords do not have, while the Space segment's $1.8 billion quarter shows the launch business remains a meaningful contributor even as it recedes in relative importance.
But that is a five-year judgment, and the next two quarters will decide whether the market treats SPCX as an infrastructure compounder or as a capital-intensive bet on the AI cycle. The fourth-quarter report will be the first real test of whether the Google ramp delivers as contracted and whether management can hold the line on the $100 billion run-rate promise without sacrificing pricing.
SpaceX did not just beat earnings on Tuesday; it proved that a rocket company can rent out its data centres at a profit. The harder proof comes next, when every new dollar of capital spending has to earn its keep in a market that is building GPU supply faster than ever.
Explore more exclusive insights at nextfin.ai.
