SpaceX's $40 Billion Debt Deal for Nvidia AI Chips

The structural oddity here is duration mismatch, not headline size. The package is shaped as roughly $10 billion in bank loans plus $30 billion of investment-grade debt ^[1], with reporting pointing to maturities stretching out as far as 30 years ^[4], and the Nvidia GPUs themselves expected to serve as the collateral ^[3].

AI data center hardware typically loses about 30% of its replacement value per year, which means the chips securing those notes could be worth a fraction of the outstanding principal well before the debt matures ^[4]. That asymmetry is exactly what one bond strategist objected to: "The equity market owns the upside, bondholders don't. So you have to get paid for the risk." ^[4]The deal is not expected to close until 2027 ^[1], which leaves a long runway for the collateral question to get repriced before a single dollar is drawn.

This is not an improvised financing. In August 2026 Nvidia announced AI compute infrastructure financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, aiming to mobilize over $500 billion of third-party capital explicitly so buildout costs sit on outside investors' balance sheets rather than Nvidia's, with Jensen Huang holding an option to backstop up to $125 billion of potential deals ^[6]. Apollo is now the expected lead on SpaceX's package and is expected to place the debt broadly ^[2], with Pimco among the lenders in talks ^[3]. Layered on top is Musk's own commitment on the August earnings call that SpaceX would build "exclusively on Nvidia because we think the Vera Rubin architecture is the best architecture" ^[3].

Trading and technology communities seized on the loop this creates - a chip vendor helping assemble the capital that funds purchases of its own chips from a customer that has sworn off alternatives - and the critique surfaced across all three forum threads that discussed the deal, from trading-focused forums to general technology forums. The pushback was not unanimous: a meaningful contrarian strand argued the arrangement is closer to ordinary vendor financing than to anything circular, and noted SpaceX's investment-grade credit ratings as a counterpoint.

Morgan Stanley's framing is the most useful lens: the pricing and investor take-up of this deal is "a useful test of how much more AI borrowing investors can absorb," set against its estimate that AI infrastructure will need $1.5 trillion of external financing by 2028 ^[8]. The run-rate context is stark - investment-grade issuance from hyperscalers and AI-focused issuers hit $218.0 billion through July 8, 2026 versus $80.5 billion over the same stretch of 2025 ^[11].

The stress indicators are already visible on SpaceX's existing paper: the 2056 notes trade near 85 cents on the dollar at roughly 2.27 percentage points over Treasuries, while investors approached earlier about the chip purchase reported receiving only a two-page memo ^[3]. A thin diligence package on a $40 billion ask is the kind of detail that looks like froth in hindsight. An independent commentary channel picked up the same thread, framing the raise within a broader AI-debt-bubble narrative rather than as routine corporate financing.

The practice SpaceX is about to scale started small and recently. CoreWeave's 2023 $2.3 billion facility backed by Nvidia H100s was the first known case of that hardware being pledged as loan collateral ^[10]. By March 2026 the same playbook had produced an $8.5 billion delayed draw term loan for CoreWeave, alongside a $7.5 billion Blackstone and Magnetar facility, a $10 billion Fluidstack deal and a $500 million Lambda loan ^[9].

SpaceX's $40 billion ask is roughly four times the size of the largest of those facilities, while the underlying valuation question - what a depreciating GPU fleet is actually worth in a default - has never been tested through a down cycle. The precedent chain explains why lenders are comfortable with the structure, and also why that comfort is largely untested.

Everything supportive about this deal depends on compute demand, not on credit metrics. SpaceX put roughly $15.8 billion of about $18.4 billion of Q2 2026 capital spending into AI compute, ended the quarter at 1.4 gigawatts of capacity, targets more than 2 gigawatts by year-end, and its CFO has guided to over $100 billion in annualized revenue by December ^[1]. Wolfe Research thinks those compute targets could eventually mean around $350 billion of Nvidia hardware sales ^[4], and Motley Fool's read leans on SpaceX's 247% year-over-year AI segment revenue growth with Nvidia still trading at "a discount of just 25 times forward earnings" ^[5]. Dan Ives argues the backlog and Starlink cash flow carry the leverage, while conceding "the debate is whether a company this early in its public life should carry that much leverage" ^[7].

On the other side of the ledger, Oppenheimer models more than $400 billion of net debt by 2031 ^[4]. The market itself barely blinked - SpaceX down about 1-2% and Nvidia up roughly 0.5% on the report ^[1]- and community analysis split along the same seam, with some pointing to SpaceX's cash position against post-raise total debt to argue the headline overstates the risk, while others questioned whether renting out idle compute inflates the revenue run-rate the whole thesis rests on.