SpaceX just borrowed 40 billion dollars to rent chips it will never own. Behind that deal sits a quiet revolution in how the AI build out gets paid for, and private credit is now the lender of first resort. The risk is sliding off the balance sheets of the largest companies on earth and onto the pensions, insurers, and savers who may not know they hold it.
By Blumefield | October 27, 2026
A 40 Billion Dollar Rental Agreement
Elon Musk spent the weekend telling anyone who would listen that SpaceX had "closed" a 40 billion dollar financing. What he did not dwell on was the shape of the deal. An affiliate of Apollo Global Management will buy the Nvidia chips and servers SpaceX wants, then lease that hardware back to the rocket company over terms of a little more than five years. SpaceX gets the compute. Apollo gets the asset. The arrangement is private credit in its purest form, and it is the clearest signal yet that the AI build out has outgrown the balance sheets meant to carry it.
The mechanics matter. SpaceX is not buying the GPUs. It is renting them, with the chips expected to arrive through the first and second quarters of 2027, and the lease clock starting only once the hardware lands. Chief financial officer Bret Johnsen has guided fourth quarter capital spending to between 35 and 40 billion dollars, a staggering figure for a company that still describes its mission as getting to Mars. Musk called the funding "debt financing" in one breath and a lease in the next, and that ambiguity is the point. The obligation sits somewhere between a loan and a rental, which is exactly where private credit likes to operate.
For a privately held firm with no public shares to issue and a cash hungry satellite and compute roadmap, borrowing from a single alternative asset manager is faster and quieter than a syndicated bank loan or a bond roadshow. It is also a template that the largest technology companies on the planet have already embraced.
How Private Credit Became AI's Default Lender
A year earlier, Meta and Blue Owl Capital closed a 27 billion dollar financing for the Hyperion data center campus in Richland Parish, Louisiana, a four million square foot complex rated for several gigawatts of power. It was, at the time, the largest private credit transaction ever executed. The structure is the real story. Meta holds 20 percent of a joint venture. Blue Owl managed funds hold the other 80 percent. A special purpose vehicle arranged by Morgan Stanley issued 27 billion dollars of A plus rated debt, with PIMCO taking 18 billion and BlackRock another 3 billion, and Meta leases the finished campus back under a long term operating lease.
That design lets Meta keep operational control of a facility it does not fund and does not carry as debt. Its spending shifts from capital expenditure to operating expenditure, and the leverage lives inside the vehicle rather than on Meta's credit rating. The linchpin, as S&P described it, is a residual value guarantee that runs for 16 years and quietly hands the technology obsolescence risk back to Meta. The debt priced at roughly 225 basis points over Treasuries, about double the spread on Meta's own corporate bonds, which tells you how much extra yield investors demanded for exposure one step removed from the parent.
Apollo and Blue Owl are not dabbling. Apollo deployed more than 40 billion dollars into this corner of the market in 2025 and has been a front runner, alongside KKR and Blue Owl, for a loan to Meta reported at 29 to 35 billion dollars. Blue Owl arranged more than 50 billion dollars of data center financing in the same year. Blackstone is sitting on a data center portfolio worth more than 70 billion dollars and some 200 billion dollars of dry powder. Brookfield has stood up a 100 billion dollar AI infrastructure program with Nvidia and the Kuwait Investment Authority. Apollo Global Management and its peers have decided that lending against chips and concrete is the trade of the decade.
The 1.5 Trillion Dollar Gap
The reason this machine exists is arithmetic. Morgan Stanley projects that global data center spending will approach 2.9 trillion dollars through 2028, and that roughly 1.5 trillion dollars of that total has no obvious source of funding. Operating cash flow from the hyperscalers covers a large slice. Corporate bonds cover another. What remains is a gap wide enough to swallow several national economies, and private credit is the capital pool rushing to fill it. The full scope of the shortfall is laid out in Morgan Stanley Research.
The headline capital budgets explain the urgency. Amazon has guided to roughly 200 billion dollars of capital spending for 2026. Microsoft is pointing at 100 to 120 billion. Alphabet has signaled 175 to 185 billion. Meta sits between 115 and 135 billion. These are annual figures, and they are climbing faster than even the most profitable businesses in history can self fund without straining the credit ratings their treasurers have spent decades protecting.
That is why the pivot to private credit is structural rather than opportunistic. Issuing hundreds of billions in conventional debt would pressure the pristine balance sheets that let these firms borrow cheaply in the first place. Raising fresh equity dilutes founders and signals weakness to a market already nervous about returns. A private credit vehicle, by contrast, lets a company book the compute it needs while parking the borrowing where rating agencies treat it as someone else's problem.
Where the Risk Actually Sits
Follow the money to its resting place and the comfortable story turns uncomfortable. The 80 percent of Hyperion that Meta does not own is funded largely by insurers and pension funds hunting for long duration yield. Firms like Blue Owl Capital package that exposure and sell it to institutions whose ultimate beneficiaries are retirees and ordinary savers. They are, in effect, underwriting a bet on Nvidia's chip roadmap and the durability of AI demand, often without framing it that way to the people whose money is at work.
The specific danger is obsolescence. Analysts including Paul Kedrosky have labeled deals like these speculative finance, pointing to thin equity cushions of 8 to 10 percent and the awkward fact that the GPUs at the center of these leases may be commercially stale within five to seven years. A five year SpaceX lease on chips that depreciate on a similar timeline leaves little margin if the hardware loses value faster than planned. The closest historical rhyme is the dark fiber overbuild of the late 1990s, when telecom operators borrowed heavily to lay capacity that then sat unlit for years while the debt stayed very much alive.
Even the architects are nervous. Apollo chief executive Marc Rowan has publicly warned of a coming shakeout across private markets, and bond investors have started pushing back as AI linked debt marches past the half trillion dollar mark. Private credit grew up in an era of cheap money and vanishingly few defaults. It has never been tested at this scale, against assets this specialized, during a capital spending cycle this steep.
What Happens If The Demand Blinks
The next two days will tell part of the story. The largest technology companies report earnings this week, and investors who cheered the build out all year are suddenly asking whether the spending will ever convert into matching profit. If the hyperscalers reaffirm their budgets, the private credit machine keeps humming. If they hesitate, the vehicles built on an assumption of relentless demand will face their first real stress test.
What makes this moment different from earlier technology booms is where the leverage lives. In the dot com era, the pain landed on public equity holders who could watch the losses accumulate in real time. This cycle has routed a large share of the risk into private vehicles, insurance balance sheets, and pension allocations that mark to model rather than to market. That makes the system look calmer than it is, right until the moment it does not. As Blumefield has tracked across the AI infrastructure boom, the financial engineering has become as aggressive as the technology it funds.
For now, the deals keep closing. SpaceX will take delivery of its leased chips next year, Meta will keep leasing back campuses it does not own, and the asset managers will keep matching long dated liabilities to the hottest collateral in the world. Whether that collateral holds its value is the question nobody writing these checks can fully answer, and it is no longer a question for Silicon Valley alone. Follow it with Blumefield as the earnings land.