NVIDIA Wants to Turn AI Compute into a New Investable Asset Class

On August 11, NVIDIA introduced a model that could substantially change how artificial-intelligence data centers are financed. Together with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, it wants to create separate financing platforms capable of mobilizing more than $500 billion of third-party capital over time. According to the company, this is neither a single fund nor a promise of revenue for NVIDIA, but a framework for financing individual projects. The logic resembles the financing of power plants, telecommunications networks, or transport infrastructure. A compute center is expected to generate a measurable revenue stream, its capacity can be leased to different customers, and investors can assess individual projects based on utilization, contracts, cash flows, and the residual value of the equipment. NVIDIA is therefore moving away from describing its products merely as chips and instead talks about complete “AI factories” that include accelerators, networking, systems, and software. The company also points to the economic life of its hardware. NVIDIA says the A100, introduced in 2020, remains commercially useful for training, fine-tuning, inference, and high-performance computing. It also states that one-year rental prices for H100 capacity rose from roughly $1.70 per GPU hour in October 2025 to $2.35 in March 2026. For the newer B200, NVIDIA cites market rates of approximately $5.30 to $7.05 per hour. The assumption that equipment retains value is central to the financing model. If the same compute capacity can be redirected among customers and software updates gradually improve its performance, an investor may view the infrastructure as more comparable to an energy or telecommunications asset than to a rapidly obsolete enterprise server. That thesis, however, depends on several conditions: sustained demand for compute, high utilization, and the ability to operate the equipment economically even after newer chip generations arrive. NVIDIA acknowledges that for some projects it may provide residual-value support of up to 25 percent of a particular opportunity. Investment institutions are nevertheless expected to assess each project independently. The company is also responding to criticism of so-called circular AI financing, in which chip suppliers, data-center operators, and customers support each other financially and may thereby amplify reported demand. For companies, the practical consequence is clear. AI computing power may increasingly be acquired as a long-term financed service instead of through a one-off server purchase. For banks and investment funds, this creates a new field of credit risk: they must estimate not only borrower quality but also the future price of compute and the pace of technological obsolescence. Key terms Residual value: The estimated value of equipment after a period of operation, important for leasing and financing. AI factory: NVIDIA's term for a data center designed primarily for large-scale artificial-intelligence computing. GPU hour: One hour of use of one graphics accelerator; a common unit for billing compute capacity.

NVIDIA Wants to Turn AI Compute into a New Investable Asset Class

The full article is available to users registered at Hard Skills