SpaceX Seeks $40 Billion for Nvidia Chips

Elon Musk put the Colossus 2 numbers on X in plain sequence. The Memphis-area AI cluster already held 110,000 Nvidia GB200 chips and 440,000 GB300s. Another 220,000 GB300s were due to become operational the following week. A further 220,000 were expected in November. One more block of 220,000 might come online in late December, he wrote, if they got lucky. He said the cluster may more than double its Nvidia chip count by the end of 2026. It was the most detailed timetable he had yet given for the buildout.

The GB200 and GB300 are Nvidia’s specialized processors for training and running large artificial-intelligence models; stacking them by the hundreds of thousands is how a data center becomes a machine of that scale. The post fixed the near-term arithmetic in public: three planned tranches of 220,000 chips apiece, timed through December, aimed at more than doubling the installed base before the year closed. The chips were named. The volumes were named. What the post left open was how hardware on that scale gets paid for.

Earlier approaches to pay for chips at multi-billion-dollar scale had already failed. When the company pitched the order to some institutional investors, the materials amounted to a two-page transaction memo. The pages carried space imagery and an arrow marking a data center to be built somewhere in the universe. “How are we supposed to take this to the investment committee?” That line, from an unnamed source, landed as the practical verdict. A sparse memo with celestial photographs and a directional arrow gave committees nothing they could underwrite.

SpaceX is seeking $40 billion to buy Nvidia chips—about $10 billion in bank loans and $30 billion in investment-grade debt. Apollo Global Management is leading the financing and helping place the debt with a broad investor base. Pimco is among a small group of lenders in talks to provide the money.

The order rests on a single architecture decision. Musk has said the company will build its data centers entirely on Nvidia hardware. “We have decided to build entirely on Nvidia because we believe the Vera Rubin architecture is the optimal architecture. We believe this is the best AI computer, and we highly value our close cooperation and partnership with Nvidia on multiple levels.” Vera Rubin is Nvidia’s latest AI computing platform, the processors built to train and run the largest models at scale. Committing the centers exclusively to that design is the technical rationale for the purchase: one supplier, one architecture, financed through bank loans and investment-grade paper that institutional buyers already know how to underwrite.

The standing to attempt a raise this large arrived in June 2026. SpaceX completed an $86 billion initial public offering, the largest in history. Shortly after the listing it received a BBB credit rating, the low end of investment grade. Less than two weeks after the shares began trading, the company finished a $25 billion investment-grade bond issuance. A BBB rating is the threshold that opens a company’s debt to insurers, pension funds, and other institutions that cannot hold junk. The June sequence—record IPO, investment-grade stamp, then a completed $25 billion bond sale—supplied the rated paper that earlier outreach never did.

Apollo’s role in the present raise grew directly from a partnership Nvidia struck in August. Alongside BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, Apollo joined financing platforms meant to mobilize more than $500 billion for AI infrastructure, creating dedicated pools of capital at scale for Nvidia customers. That distribution network was the channel through which an exclusive-Nvidia order of this size could finally clear real investment committees.

Amazon, Microsoft, Alphabet, Meta, and Oracle together issued about $200 billion of investment-grade debt in the first half of 2026—nearly double their issuance for all of 2025, according to iShares. The five companies were using it to fund data centers, chips, and power. JPMorgan Chase put the next stage higher. AI spending across the hyperscaler ecosystem could rise from roughly $700 billion this year to $1 trillion in 2027.

SpaceX bonds maturing in 2056 were trading around 85 cents on the dollar, MarketAxess data showed. A buyer at that level paid eighty-five cents for every dollar of face value due three decades out. Yields ran about 2.27 percentage points above U.S. Treasurys and were approaching junk-bond levels—the higher yields attached to debt rated below investment grade. Limited financial disclosure was cited as a reason some investors stayed cautious.

After the Financial Times report broke, SpaceX shares erased earlier postmarket gains. On Tuesday the regular session closed 0.49 percent lower at $171.92. In the after-hours and postmarket trading that followed the official close, the stock fell as much as 1.8 percent. Prices in that window printed around $169.79 and $169.94. Nvidia shares rose as much as 0.6 percent on the same news.

The transaction is expected to close in 2027. Fundraising discussions remain in an early stage and could still end without a completed deal. People familiar with the matter described the talks as private and unfinished. The only firm date so far attached to the raise is that distant close.