Jeff Gundlach has found something in the AI boom that bothers a bond investor more than an expensive stock: questionable collateral. The DoubleLine Capital CEO is warning that Nvidia’s plan with six financial giants to mobilize more than $500 billion for AI infrastructure could become a market-top signal because the long-term financing ultimately depends on rapidly evolving GPUs retaining enough value to support the debt.
Gundlach’s objection is unusually specific. Nvidia is working with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to bring institutional capital into AI data centers, while Nvidia may provide residual-value support on as much as 25% of individual opportunities. Gundlach compared financing long-term debt against chips of uncertain useful life to issuing 30-year securities against warehouses of newly engineered bananas.
Jensen Huang thinks the analogy misunderstands what Nvidia has built. He argues that GPUs aren't disposable technology tied to a single generation of AI models: they can move among customers and workloads, improve through CUDA software and remain economically useful long after newer chips arrive. Nvidia points to its A100, introduced in 2020 and still commercially active six years later, as evidence that older hardware doesn't automatically become worthless when the next architecture appears.
That disagreement over useful life is doing an extraordinary amount of financial work. If Nvidia compute behaves like durable infrastructure, institutional investors can finance it much as they finance other productive assets, collecting long-duration revenue while retaining collateral with meaningful resale value. If GPUs depreciate more like ordinary technology equipment, lenders may discover that the asset supporting yesterday’s loan is worth considerably less precisely when a borrower gets into trouble.
Nvidia has offered to absorb part of that uncertainty itself. Huang has said the company could backstop up to $125 billion, or 25% of the potential $500 billion program, and Nvidia says any residual-value support would be evaluated project by project. The company’s latest regulatory filing also shows how far this role has already expanded: Nvidia had $99 billion of equity investments, another $25 billion of equity commitments and $36 billion of cloud-service commitments as of July 26.
The balance sheet is reaching farther still. Nvidia disclosed this month that it could provide up to $105 billion of credit support for the land, power and buildings at an Ohio data-center campus where OpenAI will lease capacity, with Nvidia compute installed inside. Nvidia has also invested nearly $50 billion in frontier AI labs, arguing that these companies have enormous customer demand but don't yet possess the credit profiles necessary to finance all the computing capacity they need.
That puts Nvidia somewhere between supplier, investor and financial intermediary. It sells the GPUs, invests in companies that need them, helps secure the infrastructure where they operate and increasingly uses its own financial strength to attract outside capital into the system. Nvidia insists the resulting financing isn't circular because independent institutions are underwriting the infrastructure and its compute can be redeployed if one customer fails.
Gundlach is looking at the same structure from the other end. He has spent a career watching financial markets turn assets into collateral, collateral into securities and optimistic assumptions about future value into leverage. His concern isn't that Nvidia’s chips aren't useful today. It is that declarations of new asset classes, financial innovation and uncertain collateral values have an unfortunate habit of appearing when confidence is already abundant.
The question therefore isn't whether Nvidia can find $500 billion of capital. Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR suggest there is no shortage of institutions willing to examine the opportunity. The harder question is what happens years from now, when lenders discover what an older GPU is actually worth after several generations of better ones have arrived.
Huang believes he has made compute durable enough to become infrastructure. Gundlach has spent decades learning that every financing boom eventually discovers how durable its collateral really was.
