SpaceX borrows $40B for Nvidia chips — and bond markets are nervous

The space company buying as many Nvidia chips as Google

SpaceX is raising $40 billion to buy Nvidia GPUs. Not for satellites. Not for Starship fuel. The money is explicitly earmarked for artificial intelligence infrastructure — data centers, computing clusters, the physical bones of a training pipeline that will presumably feed whatever comes next after Starlink’s launch cadence.

That fact alone is the story. SpaceX is not a software company. It is not a cloud platform. It is a rocket manufacturer that also happens to own the largest satellite constellation in orbit. And yet it is now circling the same GPU pool as Broadcom, Oracle, Amazon, and the rest of the cloud giants. The bidding war for silicon has stopped distinguishing between industries. It only distinguishes between who can pay.

How the money is supposed to arrive

According to the Financial Times, the raise is structured in two parts. Ten billion dollars comes through bank loans. Thirty billion comes from bond issuance. Apollo Global Management is leading the bond side, and PIMCO is reportedly discussing participation. The target is completion sometime in 2027.

SpaceX’s credit rating is BBB — the second-lowest rung of investment grade. That matters because it keeps the door open to insurance companies and pension funds, the very buyers who have been quietly accumulating AI-exposed corporate debt over the past year. It also means SpaceX is one downgrade away from junk status, which would force a wave of institutional selling.

The broader pattern is what makes this notable. Oracle’s $65 billion data center build-out in Texas, Broadcom’s AI infrastructure push, Meta’s sprawling network of GPU clusters — they are all funded through debt markets that are already stretched. SpaceX is simply the latest entry in a line that now includes aerospace, telecom, energy, and sovereign-backed industrial firms. Every balance sheet is being pulled into the same capacity race.

Why bond investors are hesitating

SpaceX issued $25 billion in bonds in June, just two weeks after its Nasdaq listing. The market’s verdict has been unmistakable. The 2056-maturity bonds are trading at roughly 85 cents on the dollar. That implies a yield roughly 227 basis points above US Treasuries — a spread that sits uncomfortably close to high-yield territory despite the BBB rating. Investors are pricing in risk that the rating does not fully reflect.

The problem is not just the debt. It is the opacity. According to the FT, one investor who was approached about the new raise received a two-page deal memo. It described a data center being built somewhere in space. Literally. The memo included an arrow pointing at a vague location on a diagram with the note that the facility would be “somewhere out there.”

The investor’s response, relayed through the FT, was blunt: how do you take this to an investment committee?

That is not a joke. It is a structural friction. SpaceX has spent years operating with minimal financial disclosure by design. Its revenue streams — Starlink, launch contracts, government satellites — are real but fragmented. Adding a data center business with no cost estimates, no timeline, and no disclosed customer base makes the capital allocation even harder to model. Bond investors do not want vague promises. They want cash flow coverage ratios.

What changes if this closes

If SpaceX completes the raise and follows through on GPU purchases, three things happen.

First, Nvidia’s backlog tightens further. The company is already allocating capacity to the biggest cloud customers. Adding a $30-billion bond-funded aerospace company to that queue shifts the supply math. Smaller AI companies and well-funded startups face longer delivery timelines and higher negotiation leverage from Nvidia.

Second, the spread between SpaceX’s existing bonds and Treasuries likely widens before it narrows. The market will price in the new debt before it prices in the revenue it is supposed to generate. If the data centers produce usable AI infrastructure on schedule — a massive if — the bonds may recover. If not, they will trade closer to junk.

Third, and perhaps most important, the precedent normalizes non-tech balance sheets in the GPU auction. When a company whose primary business is rockets competes for the same chips as a semiconductor equipment maker, it signals that the shortage has moved past economics into speculation. Buyers are purchasing capacity they do not yet need, in facilities they have not yet planned, funded with debt that is already trading at distressed levels.

The larger picture nobody is naming

The Korean financial press picked up on this story and connected it to the broader Broadcom-Oracle-SpaceX triangle. That instinct is correct, even if the coverage stops short of the implication. The implication is that AI infrastructure is no longer a technology play. It is a credit play.

Capital is moving into physical AI capacity faster than revenue models exist to justify it. Companies that do not traditionally build data centers — SpaceX, Oracle, cloud providers pivoting into vertical AI — are issuing debt against future margins that have not been demonstrated. The bond market is absorbing the risk because it has nowhere else to deploy yield in sufficient volume. Pension funds and insurers are constrained by regulation to hold investment-grade paper. Space is expanding, literally and figuratively, into every corner of the market.

Musk’s net worth jumped back above $1.4 trillion after three days of stock gains. That is a headline number. The quieter story is whether SpaceX’s bonds will ever trade back above par.

The AI infrastructure build-out is real. The funding is real. The question now is whether the debt backing it is priced for a future that may not arrive on schedule.