$40bn of Debt for Nvidia Chips (SpaceX, October 2026): Why the AI Build-Out Is Moving From Cash Flow to the Bond Market — and What That Changes

SpaceX is in talks to raise about $40 billion to buy Nvidia AI chips — roughly $10 billion in bank loans and $30 billion in investment-grade debt, with Apollo expected to lead, according to the Financial Times and confirmed by Reuters' sources on 7 October 2026. Nothing is signed. The money matters less than the method: the AI build-out is now being paid for with borrowed money at a 10-year Treasury yield of 5.28%, so its cost now moves with the bond market.

What was reported, and what wasn't

All of the detail comes from people familiar with the talks, so start by separating what is reported from what is on the record.

The Reuters report published on 7 October said SpaceX is in talks with banks and asset managers to raise $40 billion to buy Nvidia artificial-intelligence chips, according to two sources. One source said the company is talking to PIMCO, and that the split is about $10 billion in bank loans and $30 billion in investment-grade debt — confirming the Financial Times, which first reported that Apollo is expected to lead and place the debt with a broad range of investors. Bloomberg's account of the FT report says the transaction is expected to close in 2027.

SpaceX and Nvidia did not respond to Reuters' request for comment, and Apollo and PIMCO declined to comment. Reuters reported SpaceX shares down 2% in early trade and Nvidia up 0.5%.

The reason for the chips is also on the record, via Reuters: Musk said last month that xAI's Colossus 2 data centre could more than double the number of Nvidia chips it uses by December, and that the company plans to use Nvidia hardware exclusively for its data centres.

Why chips are now bought with debt

A GPU cluster has the economics of an aircraft fleet, not of software. The money goes out up front, at delivery; it comes back over the several years the hardware stays useful. When the order is small next to a company's operating cash flow, the gap doesn't matter — the cash already coming in pays for it. When the order is large and urgent — doubling a data centre's chip count in a quarter — no company's operating cash flow arrives fast enough, and something else has to bridge the gap.

There are three options, and SpaceX has now used all three. It sold shares (the June IPO). It borrowed in the public bond market (the June $25 billion sale, which drew about $89 billion of demand, per Bloomberg, and repaid a $20 billion bridge loan from March). And now it is negotiating a mix of loans and bonds tied to a specific chip order.

The rate underneath moved 78 basis points

This is where the story reaches markets most people trade. A bond's yield is a benchmark Treasury yield plus a credit spread. Both halves have a price, and since June the benchmark half has risen a long way.

On 23 June 2026, the day the June bond priced, the 10-year Treasury yield was 4.50%, according to the US Treasury's daily par yield curve. On 7 October it was 5.28%, with the 30-year at 5.67%. The 2036 tranche of that bond came with a 5.875% coupon, about 1.4 percentage points over Treasuries and roughly 0.4 points wider than the average similarly rated BBB bond, per Bloomberg.

That last step is arithmetic, not a forecast: the spread on a new deal depends on demand on the day and could come in tighter or wider. But it shows the scale. On $30 billion, every percentage point of yield is about $300 million a year in interest. Once chips are bought with debt, the cost of the AI build-out rises and falls with Treasury yields — the factor the Fed, inflation data and Treasury supply drive, not anything a chip designer controls.

The loop between customer, lender and supplier

The second channel is less obvious and more important for index traders. Nvidia's revenue depends on its customers' ability to pay, and those customers are increasingly paying with borrowed money.

That link has become part of Nvidia's own strategy. Per Reuters, Nvidia partnered in August with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR on financing platforms intended to mobilise more than $500 billion for AI infrastructure. Apollo is now the reported lead on a customer's $40 billion chip financing. And this site has already covered how Nvidia's filings put a number on its guarantees around the OpenAI lease. Put together, the supplier, its financing partners and its largest customers are now connected through credit as well as orders.

None of this is unusual on its own: carmakers have finance arms, aircraft makers support customer financing, and equipment vendors have always helped buyers fund purchases. What it changes is the transmission. In a cash-funded build-out, AI demand slows when the buyers' profits slow. In a credit-funded one, it can also slow when lending slows — wider spreads, a failed deal, a rise in yields — even if the buyers' businesses look fine. Morgan Stanley's estimate, cited by Reuters, that AI infrastructure will need $1.5 trillion of external financing by 2028 gives a sense of how much of future demand now depends on lenders.

Where it reaches the indices

The concentration of the US indices is what makes a single company's financing a market story. A handful of mega-caps, Nvidia above all, account for an outsized share of NAS100 and US500 moves, and SpaceX has been in the Nasdaq-100 conversation since its listing — see the coverage of its first earnings and the $100bn ARR target.

Three variables to watch, none of them a call on direction:

• Credit spreads on the big AI issuers. If investors absorb heavy issuance at stable spreads, the financing channel stays open. Spreads widening on each new deal would be the first sign that lenders are pricing in more risk — Reuters noted lenders and investors are growing "more cautious" about funding the expansion.
• The 10-year yield. At 5.28%, the risk-free alternative to an equity earnings stream is already high — the same pressure covered in the S&P 500 earnings-yield note. Debt-funded capex adds a second, direct cost on top.
• Whether the deal actually closes. The reports describe talks, with a close expected in 2027. A smaller deal, a different mix or no deal at all would each tell you something about how willing lenders are.

It also connects to the power story from earlier this week: Google's 3,590 MW contract with Constellation was another case of AI spending turning into long-term, fixed obligations. Chips and power are the two largest physical inputs, and both are now being locked in with long contracts and borrowed money.

What it does — and doesn't — mean for the dollar

Be careful about forcing this into a currency story. A $30 billion dollar-denominated investment-grade deal that attracts foreign buyers is a capital inflow, but it is one of thousands of flows, and corporate issuance has never been a reliable driver of the US dollar on its own. The real link to currencies runs through the factors that already move the dollar: Treasury yields and risk sentiment. If credit-funded AI spending kept long-term yields high, that would show up in the dollar's rates picture; if a failed financing knocked risk appetite, it would show up in sentiment. Both are factors the meter already scores. For the background on how yields reach currencies, see bond yields and currencies, and for how this site approaches stories like this one, see about.

The narrow conclusion is the useful one. The AI build-out has reached the stage where its biggest buyers pay with borrowed money, so its pace now depends partly on the price of money. That doesn't say whether the spending pays off. It says that a rates trader and an AI-stock trader are now watching some of the same numbers.

Educational macro context only — not investment advice.