The most important SpaceX number going into tonight's print is not $125 billion. It is 83.345, the price at which the company's 6.65% notes due 2056 traded on Monday. That is a lifetime low for paper that came to market at 98.83 in June.
A quarter century ago the fiber builders had the bandwidth story, the equity multiples and sell-side diagrams of traffic doubling every few months. Global Crossing laid glass under oceans while its admirers talked about the terabits to come. In that cohort, the bonds often started trading like a question while the stocks were still being discussed as an answer. Once again a builder sits at the center of a capacity race, pouring gigawatts this time instead of glass, with a 2027 forecast to match the ambition. Its name is SpaceX. Its creditors are already asking the question.
Here is the read the equity debate is missing. Deutsche Bank's Edison Yu has 2027 revenue at $125 billion and EBITDA at $84 billion, well above consensus of $108 billion and $65 billion. Every one of those dollars depends on a financing assumption the credit market is actively repricing. Equity analysts treat compute deployment as the swing factor. It is not. The swing factor is what the next $40 billion costs, and where it sits in the capital structure relative to the bonds already outstanding.
The equity model holds the cost of capital constant. The bond market has stopped holding it constant.
What Does the Quarter Actually Tell Us?
Start with what consensus has right, because it has a good deal right. The revenue is real and it is accelerating. Deutsche models Q3 revenue of $12.6 billion, up 127% year over year, with AI solutions & infrastructure jumping from $2.2 billion in Q2 to $6.0 billion on a full quarter of cloud hosting. That makes AI just over half of total revenue at a company still named after rockets. Connectivity adds nearly $5 billion, up 60%, with broadband subscribers climbing to 13.5 million even as ARPU slides another $2 to $64. The Space segment, meanwhile, is expected to shrink 22% on twelve external Falcon 9 launches.
None of that is fiction. A business that triples a segment sequentially has customers.
Now read the other column of the same quarter. Yu forecasts $19 billion of capex "mainly directed at AI compute," taking capacity to 1.8GW, and free cash flow burn of $14 billion. That is one quarter. Capex is running at roughly one and a half times revenue, and the gap is not a rounding issue. The quarter's revenue would cover about two thirds of the quarter's equipment bill before anyone gets paid a salary.
Q2 is the precedent. SpaceX beat on revenue, and the stock closed down 14% the next day, erasing some $225 billion of market value. Options now price a move of about 7.8% either way, a little more than half of what actually happened last time. The market has decided that a revenue beat is no longer news. The part consensus misses is that it has not decided what is news, and the last print suggested the answer lives below the EBITDA line.
The beat is priced. The burn is not.
Who Pays for 2027?
Deutsche's own estimate changes answer this, if you read to the bottom of each block. Yu raised 2027 revenue by $10 billion and 2027 capex by $22 billion, to $207 billion. He raised 2028 capex by $40 billion, to $230 billion. Against $84 billion of 2027 EBITDA, $207 billion of capex means about $2.46 of spending for every dollar of operating earnings. The bull model does not shrink the funding problem. It enlarges it, and it does so faster than it enlarges the earnings.
Line DB's EBITDA up against DB's capex and the shortfall runs about $47 billion this year, $124 billion in 2027 and $81 billion in 2028. That is roughly $250 billion across three years. SpaceX ended Q2 with $100 billion in cash and marketable securities, and Deutsche expects about $13.5 billion of it to have gone in Q3 alone. The 2027 gap by itself is larger than the entire remaining cash pile.
On the debt side, SpaceX reported $38.4 billion of borrowings as of June 30. It then sold $25 billion of investment-grade bonds less than two weeks after its $86 billion IPO. On October 6 the FT reported a further $40 billion ask: about $10 billion of bank loans and $30 billion of IG debt in a deal led by Apollo, to pay for Nvidia chips. If all of it lands, that is $65 billion raised or sought since June, stacked on top of the June 30 balance. Against a three-year gap of a quarter of a trillion dollars, $65 billion is a down payment.
Hyman Minsky, in Stabilizing an Unstable Economy, sorted borrowers by whether their cash flows could service their debts or whether they depended on rolling it. The distinction should be applied carefully here. SpaceX is not a borrower that cannot cover its interest; on Deutsche's numbers EBITDA covers coupons many times over. What Minsky's frame does capture is dependence. The growth case, as distinct from the survival case, now requires continuous, uninterrupted access to markets at a price that does not erode the return on what is being built. A firm that needs the capital markets every quarter to hit its forecast has handed its forecast to the capital markets.
The 2028 line is where the arithmetic turns from demanding to uncomfortable. Revenue barely moves, $200 billion against $198 billion previously, and 2028 EPS actually dips. Yu says plainly why: "there is concern by investors that pricing could normalize in 2028-29," and the bank now assumes $/GW pricing "declines by 25% to about $30bn in 2028." The capex peaks in the same year the rental rate on the capacity falls by a quarter.
The bull case is a financing forecast wearing a revenue forecast's clothes.
Why Would a Bondholder Accept 83 Cents?
At 83.345, the 2056s yield roughly 8.1%. That is a BBB-rated issuer, supposedly about to produce $84 billion of EBITDA, borrowing for thirty years at a yield usually seen in high yield. For orientation, the 10-year Treasury sat at 5.09% on October 16, with 2s10s at a positive 0.35%. The tenors differ, so treat this as orientation rather than a precise spread, but SpaceX's long paper is clearing about three percentage points above the government's 10-year. The shorter paper says the same thing. The 5.875% notes due 2036 traded at a 193 basis point spread earlier this month against 140 at issue in June. Five-year CDS jumped as much as 16.7 basis points to about 197.6 the day after the Apollo story, the highest since the contracts began actively trading in June near 110.
Credit is not pricing default. Three narrower problems, each specific, are getting priced.
First, the duration mismatch is extreme. A thirty-year liability is funding an asset class whose economic life is measured in product cycles, and whose revenue is contracted on terms that do not resemble thirty years. Goldman models $85.5 billion of 2027 AI solutions & infrastructure revenue, $45.5 billion of it from hosting deals on roughly 1GW of contracted capacity. Most of those deals are "90-day cancellable on both sides." A bondholder in the 2056s is lending money until the middle of the century against cash flows that can walk out the door in a quarter. Equity investors can extrapolate a 90-day contract to 2029 because their upside is open-ended. A creditor with a fixed coupon gets none of the upside and all of the cancellation risk.
Second, supply. The $25 billion in June was described, depending on who you asked, as "4x oversubscribed." Bondholders have been underwater nearly since the break. Barclays' credit team, led by Andrew Keches, warned that the coming Apollo supply could pressure existing bonds in the near term; within two weeks the 2056s were 1.4 points below their October 7 level. Deutsche's revised capex makes clear $40 billion is not the last ask. Sal Naro of Coherence Credit put the general case bluntly to Bloomberg: "This is unprecedented debt supply with no real ending in sight." That supply lands in a duration market where long Treasuries already yield around 5% and the government is not shy about issuing either. A buyer who can get 5% from the Treasury needs a reason to take thirty years of Musk risk at anything less than a wide spread.
Third, and least discussed, is structure. The Apollo financing is reported as chip-collateralized, and Barclays raised the possibility of an SPV. Their long-run argument, that an SPV reduces debt at the parent, is reasonable from the company's side of the table. Look at it from the unsecured parent bondholder's chair. The most liquid, most financeable assets SpaceX is buying, the Nvidia chips, may be pledged to someone else. How much of the $40 billion actually sits in a separate vehicle has not been disclosed, and that link remains unverified. The direction of the risk is not ambiguous. Each secured or ring-fenced tranche pushes the existing unsecured paper further back in line for the assets that wear out first.
The 2056s are pricing a position at the back of the queue for depreciating collateral, funded by contracts that can lapse in ninety days.
What Happens If the Bond Market Is Right?
As of Friday's close the stock sat at $146.03, about 8% above its $135 IPO price. The bonds are almost 16% below theirs. Deutsche holds its Buy and a $235 target; Evercore's Kutgun Maral sees a $104 billion run rate by year-end and raised his target to $270; 30 of 39 sell-side analysts rate the stock a Buy. Musk himself has floated $1 trillion of revenue by 2030 and a "non zero chance" of 2029, against an average analyst 2029 estimate of $208 billion. Can we have $1 trillion? $1 trillion! A non-zero chance of a year early!
The equity crowd is not uniformly euphoric, and the bull case deserves its strongest form. Morgan Stanley's Adam Jonas recounted asking 40 clients who owned the stock; "not a single hand went up." An under-owned stock with a tripling segment and a Buy-rated sell side is the kind of setup that can rip on a clean print, especially if management pairs it with a financing plan that calms creditors. That alternative history is real. The question is what it requires.
Follow the escape hatches, because each one runs through the 2027 equity number. If spreads keep widening, the next tranche costs more, and higher interest expense eats into the earnings the multiple is applied to. If SpaceX substitutes equity for debt, the share count rises and the per-share value of that $84 billion of EBITDA falls. If management slows capex to protect the balance sheet, compute deployment slips, and compute deployment is precisely the variable Yu calls "the largest swing factor" for revenue. There is a fourth door, where financing arrives cheaply and in volume, but that is the door the bond market is currently pricing as harder to open. Every route except the fourth reduces the equity forecast. The equity model assumes the fourth.
I don't know what management will say tonight, and the scenario weighting here rests on public consensus figures rather than live positioning data. What can be said is where the asymmetry sits. The outcome the math supports most comfortably is a repeat of Q2: a solid revenue print overshadowed by the cost of producing it, with credit setting the tone the next morning. A bullish resolution exists but runs one way only, through a capex outlook that does not begin with a "2" and specific clarity on how much of the Apollo money sits ahead of existing holders. The bull case is one outcome among several, and not the most likely of them.
So the checklist for the call is shorter than the one the sell side is circulating. Neocloud pricing matters, Cursor matters, Flight 15 timing matters at the margin. The items that move the bonds are three: the data center capex outlook for 2027, the status of the $40 billion deal, and the share of it housed in an SPV. Any hint that more borrowing follows belongs on the list too. The cleanest test of the print will not be the stock in after-hours trading. It will be where the 2056s and the five-year CDS open on Wednesday.
There is a wider implication for anyone holding the investment-grade index through vehicles like LQD. If AI builders keep landing $30 billion tranches into the IG market, the pressure is not confined to one issuer's spread. Whether that becomes an index-level problem is unproven, and SpaceX alone does not make it one. The pattern of a single sector monopolizing duration demand is a signal of how the funding of this build is being shared out, and with whom.
Equity holders own the gigawatts. Bondholders own the invoice, and the invoice comes due first.
