The Chain Is Now Visible
Nvidia announced Monday a set of partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to create financing platforms intended to mobilize more than $500 billion for AI infrastructure. The capital will come largely from 'third-party investors,' allowing Nvidia customers to finance chips and data centers while keeping Nvidia's own risk limited and off its balance sheet.
The structure is straightforward. An independent financing vehicle raises money to purchase Nvidia GPUs and data-center infrastructure. An AI company then leases that compute or commits to using it, generating a payment stream against which the vehicle can borrow. Apollo, KKR, and their peers structure or manage that debt and place it with large pools of institutional capital — mostly insurance and retirement savings — that they oversee.
'We have moved from an era in which companies bought chips and built data centers project by project to one in which AI factories can be financed as productive infrastructure,' Nvidia CEO Jensen Huang wrote Tuesday. His premise: 'In AI, compute is revenue.'
From Cash Flows to Bonds to Pensions
A year ago, Big Tech could credibly claim it was funding AI from its own enormous cash flows. That era is closing. Goldman Sachs estimates AI-related financing now accounts for nearly one-quarter of all gross U.S. investment-grade issuance, while AI investment itself is approaching $600 billion this year.
The logic for drawing on pension and insurance capital is structural. Data centers are expensive, long-lived projects requiring financing over many years. Insurers and pension funds carry long-dated obligations — annuities and retirement benefits owed decades into the future — and therefore seek long-duration assets whose cash flows can be matched against those liabilities. Private-credit and infrastructure managers act as the middlemen.
Nvidia is not standing entirely aside. Huang said the company may provide residual-value support of up to 25% for some projects — effectively promising partial protection against the possibility that the chips backing a financing are worth less in the future than lenders expected. Technology strategist Ben Thompson, writing in Stratechery, called that 'in a certain sense, a price cut': Nvidia is using its own profits to reduce customers' cost of capital.
On Wall Street, the initial read was cautious optimism. Morgan Stanley's Joseph Moore said the arrangement alleviates concerns about circular financing because third-party investors would provide most of the capital while Nvidia participates only to a limited extent. Bank of America's Vivek Arya added that Nvidia's chips are 'unusually financeable' because GPUs can be moved among operators and CUDA software can extend their useful lives.
Thompson's concern, however, cuts to the core: the pools of capital now being recruited — pension funds, insurance floats — are fundamentally different from venture capital or tech stocks. They are designed, at least in part, to seek safety.
The Market Has Already Voted — But the Bill Comes Later
The numbers here are not small. Half a trillion dollars in mobilized capital, one-quarter of investment-grade issuance, $600 billion in annual AI investment — these are figures that reshape the credit landscape, not just a single company's balance sheet.
For free-market readers, the structure itself is not the problem. Private capital allocating to productive infrastructure is exactly how a market economy is supposed to work. The question worth asking is whether the risk profile of AI compute — chips that depreciate rapidly and face relentless generational replacement cycles — is being accurately priced before it lands inside instruments that retirees and policyholders depend on. Capital rewards clear rules, and the details of these arrangements, by Nvidia's own admission, remain largely unknown. When the fine print finally surfaces, the institutions holding that paper will want to know exactly what 'residual-value support of up to 25%' actually covers — and what it does not.
