Capital expenditure for AI infrastructure is shifting from "giants paying out of pocket" to "borrowing money across the entire market for leverage."

SpaceX is currently planning to raise about $40 billion for large-scale purchases of Nvidia chips. Of this, about $30 billion is planned to be raised through investment-grade bond issuance, and about $10 billion comes from bank loans. The deal is expected to be led by Apollo Global Management, with participation from institutions such as Pimco, aiming for completion in 2027.

After the news broke, SpaceX's stock price fell about 1.8% in after-hours trading, while Nvidia saw a slight increase. The market response was direct: demand for computing power remains strong, but the pace of debt expansion is starting to unsettle investors.

| Musk's Choice: Fully Betting on Nvidia's Vera Rubin

This financing is not simply about "buying graphics cards," but SpaceX's clear stance on the AI computing power path.

At the August earnings call, Musk stated that SpaceX decided to build entirely on the NVIDIA platform, mainly because he was optimistic about Nvidia's latest Vera Rubin architecture.

This means that SpaceX's future AI data centers, cloud service expansion, and space computing workloads in cooperation with NVIDIA will all be deeply tied to the NVIDIA ecosystem.

For NVIDIA, this is not just a new order, but also a strengthening of ecosystem barriers. Even as competitors like Broadcom and AMD continue to catch up, leading customers are still willing to pay a premium for NVIDIA's latest architecture.

|Why was so much money borrowed? BBB ratings are key

The core support behind SpaceX's ability to advance such large-scale debt financing is its investment-grade credit rating.

After completing its IPO in June this year, SpaceX received a BBB rating and issued $25 billion in investment-grade bonds in less than two weeks. Although BBB is a lower tier among investment-grade funds, it is sufficient for long-term funds such as insurance companies and pension funds to buy in compliantly. Compared to junk bonds, these funds offer greater allocation space and lower costs.

However, the market is not without concerns. SpaceX's existing 2056 maturity bonds are currently trading at about 85% of face value, with yields about 2.27 percentage points higher than U.S. Treasuries, approaching junk bond levels.

Some investors remain cautious about SpaceX's limited financial disclosures, excessive capital expenditures, and rapid debt expansion.

| From "two-page PPT" to Apollo's endorsement: upgrading the professionalization of financing

This deal also exposes the awkwardness of SpaceX's previous fundraising attempts.

According to reports, SpaceX had previously pitched chip procurement financing to some investors, but only provided a two-page memorandum, even labeling it with space images and arrows stating "a data center will be built somewhere in space." Some investors bluntly said, "How did we take this to the Investment Committee?" ”

The introduction of Apollo as the main driver this time is precisely to address this shortcoming. Apollo's credit business manages assets totaling about $800 billion, having led major financings for Intel, Bayer, and others. In June this year, it even led a $35 billion chip financing round to purchase Broadcom processors. It organizes distribution and makes it easier to persuade conservative institutions like insurance and pension funds to enter the market.

Broader context: Nvidia is building a $500 billion computing power financing platform, and SpaceX is just one example of this AI infrastructure financing wave.

In August this year, NVIDIA announced partnerships with Apollo, BlackRock, BlackRock, Brookfield, Goldman Sachs, and KKR to build a $500 billion financing platform to help customers purchase chips and build AI data centers at lower costs. Nvidia itself may endorse up to 25% of the chip residual value, but the main investors remain third-party long-term capital.

This shows that AI computing power is no longer just "hardware procurement," but has been redefined by Wall Street as a collateralizable, securitizable infrastructure asset that generates cash flow.

But this also raises new questions: if terminal commercialization lags behind, will high-leverage computing power investments become the next risk point?

Kingtech Perspective | The main theme of computing power remains, but it's time to start looking at "who can afford to pay back"

From an investment perspective, this news further confirms three trends:

First, demand for AI computing power remains strong

Whether it's SpaceX, cloud providers, or cutting-edge AI labs, all are accelerating their capacity expansion. Nvidia's Vera Rubin architecture has received endorsement from leading customers and remains dominant in short-term competition.

Second, financing methods are shifting from on-balance-sheet CapEx to off-balance-sheet debt financing

This means the upper limit on AI capital expenditure has been raised, but it also means that future demands for cash flow, computing power utilization, and commercial monetization will be higher.

Third, the market will begin to scrutinize the "quality of computing power assets" more strictly.

Not all purchases of GPUs are the same. Companies that generate stable computing revenue, have clear customer structures, and reasonable debt structures are more likely to receive valuation premiums.

Considering the current market, it is recommended to pay attention to three clues:

SpaceX's $40 billion purchase of NVIDIA chips may seem like a bold bet by Musk, but in fact, it signals that the entire AI industry is entering a "capital-intensive + financialized" phase. Computing power remains the main theme, but investment logic is shifting from "buying chips" to "looking at cash flow."

In the next phase, whoever can turn computing power into sustainable income will truly be worth long-term allocation.