SpaceX's $40 Billion Chip Loan Shows Who Carries AI's Balance Sheet Risk

The biggest number in AI infrastructure news this week is $40 billion. The more useful detail is who would be holding the paper.

On October 6, the Financial Times reported that SpaceX is seeking about $40 billion in financing, led by Apollo Global Management, to buy Nvidia AI chips. Reuters relayed the report and said, in a separate story, that SpaceX is in talks with banks and asset managers to raise the money, citing two sources. The reports rest on unnamed sources. SpaceX, Apollo and Nvidia had not confirmed the plan when the stories ran. Treat it as a credible report of talks, not a closed deal.

Even so, the structure tells builders something that a headline total does not. A chip buyer is borrowing to pay a chip vendor, and the debt stays with the buyer.

What Actually Happened

According to the FT, as summarized by The Next Web and 24/7 Wall St., the package would combine about $10 billion in bank loans with about $30 billion in investment-grade debt. Investment-grade debt is sold to investors who view the borrower as a relatively low default risk, which opens the door to pension funds and insurers. Apollo is expected to lead the deal and place pieces with a wide range of investors. The Next Web reports that Pimco is among the lenders in talks, citing both the FT and Bloomberg.

The two outlets disagree on how far along the talks are. The FT's sources point to a close in 2027. Bloomberg, per The Next Web, described the talks as early and said they could still end without a deal. That gap matters for anyone tempted to treat the figure as committed capital.

The chips are meant for SpaceX's AI data centers. At an August 4 earnings call, Elon Musk said, “So we’re exclusive to NVIDIA,” according to 24/7 Wall St. The Next Web adds that Musk said last month the Colossus 2 site could more than double its Nvidia chip count by December. SpaceX also rents out some of that hardware: The Next Web reports that Google agreed to pay SpaceX $920 million a month for access to about 110,000 Nvidia GPUs.

The market reaction was small. The Next Web and 24/7 Wall St. both report that SpaceX shares fell about 1% in after-hours trading on Tuesday while Nvidia rose about 0.5%. SpaceX closed October 6 at $171.92, per 24/7 Wall St., against a $135 IPO price in June.

The Buyer Holds the Debt

The financing structure is the story. In the arrangement 24/7 Wall St. describes, SpaceX takes on the debt, and Nvidia records the chip order as revenue when the hardware ships. Nvidia is paid. SpaceX owes lenders for years.

In the reported structure the vendor is not the lender. The risk sits with the buyer and the institutions that buy its bonds, not with the company shipping the chips.

Nvidia is not absent from the financing picture, though. The Next Web and Financial News both note that in August Nvidia set up financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to raise more than $500 billion in third-party capital for AI infrastructure. So the vendor helps assemble the lender network without necessarily lending itself. That arrangement lets Nvidia grow shipments while keeping the debt off its own balance sheet, at least as the reports describe it.

For builders, the practical read is that hardware demand now depends on credit markets in addition to customer demand. Morgan Stanley estimates AI infrastructure will need $1.5 trillion in outside financing by 2028, according to 24/7 Wall St. and Financial News, both citing the bank. If lenders slow down or reprice, the chips get bought more slowly, and that reaches capacity and pricing downstream.

Spending Has Outrun Revenue

SpaceX's own numbers explain why it needs outside money. According to 24/7 Wall St., in the second quarter of 2026 the company reported revenue of $7.81 billion against capital spending of $18.37 billion, of which $15.83 billion went to AI. It posted a net loss of $541 million. Financial News, citing the company's 10-Q, reports the same net loss, narrowed from $1.008 billion a year earlier.

Per 24/7 Wall St., management said on the August 4 call that the next two quarters should look “very similar” on spending. Spending that runs well ahead of revenue is a normal feature of an infrastructure build. It is also exactly the profile that makes the debt market a necessary participant.

There are other pressures on the same stock. 24/7 Wall St. reports that lockup expirations will add 328.4 million marketable shares on October 9 and another 328.4 million on October 24, with up to 1.3 billion more after third-quarter earnings. That is one reason borrowing is attractive: it avoids issuing new shares into a market that is already absorbing supply. 24/7 Wall St. frames it as debt avoiding dilution today.

The rate backdrop is not forgiving. Financial News, citing Federal Reserve Bank of St. Louis data, puts the 10-year Treasury yield at 5.31% on October 5. A $30 billion investment-grade tranche priced against that kind of benchmark carries a real carrying cost, which the borrower needs its AI revenue, including rental income like the Google arrangement, to cover.

None of this says the plan fails. It says the plan works only if utilization and pricing hold up long enough to service the debt. A rental contract at $920 million a month is the kind of cash flow that supports borrowing. A cluster that sits underused does not.

What This Does to Everyone Downstream

If you build on top of a model API or a GPU rental, you are several steps removed from this bond market. You are not insulated from it.

First, the cost of hardware capital is part of the price of compute. Providers financing chips with debt need contracts and utilization that cover interest. That tends to favor longer commitments, minimum spend levels and reserved capacity over casual on-demand pricing. Expect more sales motion aimed at locking customers in, because lenders want predictable revenue.

Second, concentration is rising. A buyer committing to one vendor's hardware, as Musk described, and financing that purchase with long-dated debt has little room to change course. If a better chip or a cheaper architecture appears, that buyer still owes the money. For customers, that can mean a provider that is slow to adopt alternatives or aggressive about filling existing capacity.

Third, the vendor is shielded, the customer is not. Financial News notes the muted share reaction fits a pattern: Apollo's role in financing Nvidia's supply chain was already public. 24/7 Wall St. makes the related point that Nvidia gets paid when chips ship, while SpaceX carries the debt for years, and that the power, cooling and networking suppliers behind the build-out are in a similar position. The people with the thinnest protection are the borrowers, and by extension the customers who depend on those borrowers staying solvent and well funded.

Fourth, this fits a pattern we have covered before. Nvidia's backing of an OpenAI data-center campus in Ohio is another case where compute sourcing is tied to who finances the hardware, not just who buys it. The financing layer is now part of the product.

What to Watch

Three markers will say whether this report turns into a real deal. The first is whether SpaceX confirms the financing in a filing, which 24/7 Wall St. flags as worth tracking. The second is whether rating agencies publicly engage with the investment-grade tranche. The third is how the stock behaves around the October 24 lockup, since a weak equity price makes every debt conversation harder.

A reasonable skeptic can also ask how much of the $40 billion would be drawn at once. The reports describe a total, not a draw schedule or terms, and no pricing has been disclosed in the material we reviewed. Until terms surface, the right posture is to hold the structure as informative and the amounts as provisional.

What Builders Should Take From It

• Ask providers how their capacity is funded. Debt-heavy providers need utilization and long contracts. That affects your negotiating position and what happens in a downturn.
• Price a second source into your plan. If your main compute or model vendor is tied to one chip supplier, keep a tested path to another provider or an open-weight option for your critical workloads.
• Watch commitment terms, not just list prices. Expect reserved-capacity and minimum-spend offers to get more attractive on paper. Read the exit terms before you sign.
• Track credit conditions as an input. The 10-year yield, bond-market appetite for AI paper, and lockup calendars are leading signals for how fast capacity expands.
• Separate announced from confirmed. This deal is a report of talks with a 2027 target close. Do not build a roadmap on a figure no principal has acknowledged.
• Model your own unit economics against a funding shock. If your gross margin depends on subsidized or cheap compute, test what happens when providers reprice to cover debt service.

Developer312 covers the AI business signals builders actually need to act on. Get the weekday briefing at developer312.com.