Nvidia’s $500 Billion Bet: Making AI Compute Wall Street’s Next Asset Class

Nvidia and Six Asset Managers Target $500 Billion for AI Factories

Nvidia is trying to convince Wall Street that AI computing power should be financed like infrastructure. On Monday, it said it had created financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, with a target of raising more than $500 billion in outside capital for 'AI factories', the data centres that train and run artificial intelligence models.

The agreements announced are memorandums of understanding rather than signed contracts, and Nvidia says each project still needs a definitive agreement. There is no timetable, no announced split of capital among the six firms and no first project. The $500 billion is not revenue for Nvidia and does not represent one fund or one customer. It is a capital target to be raised over time.

Under the model, the six firms would evaluate each financing opportunity separately, looking at customer demand, hardware utilisation, cash flow and likely second-hand value. Nvidia would supply the computing platform, while the investors decide what to fund. Jensen Huang has said Nvidia could offer a residual value backstop of up to 25% of a deal, case by case, but has not explained how it would work or who would absorb the first losses.

The core question is whether machines that lose value quickly can be financed as if they were long-lived infrastructure. If a new generation of chips arrives before a loan is repaid, the lender may be left with ageing hardware worth less than expected.

The Residual Value Problem Behind the AI Infrastructure Pitch

Why Nvidia Wants Outside Capital

By bringing in external financiers, Nvidia can expand its customers' purchasing power without loading each AI data centre onto its own balance sheet. The pitch rests on Huang's argument that AI compute is an 'investable asset class' because it keeps generating revenue and can be reassigned to new customers. That is a genuine commercial strength, but the announcement itself is still only a set of MOUs, not committed capital.

The Residual Value Question Lenders Must Price

Every GPU-backed loan depends on two figures: who has committed to pay for the computing capacity, and what the hardware is worth if that customer walks away. The second figure is the residual value. Huang's offer of up to a 25% backstop shows that outside capital may only arrive because Nvidia is willing to cover part of the downside. Until contracts are public, however, that is not a guarantee, and no one should assume Nvidia will absorb the first losses.

What Amazon's Depreciation Change Signals

Amazon provides a concrete warning. From 1 January 2025, the company cut the estimated useful life of some servers and networking equipment from six years to five years, citing faster technology development in AI and machine learning. It estimated the change added about $1.4 billion to 2025 depreciation expenses and reduced net income by roughly $1 billion, mostly in AWS. That was not specific to Nvidia chips, but it shows that one of the world's largest AI infrastructure operators concluded the equipment wears out faster than previously assumed.

Rental Income Is Not Resale Value

Nvidia points to rising rental prices as evidence that hardware is durable: the rental rate for an H100 chip rose from roughly $1.70 per hour in October 2025 to $2.35 in March 2026 by one measure. But rental income shows what a chip can earn today, not what it would sell for in three years. Different marketplaces report very different H100 prices, and a strong rental market does not by itself prove a strong resale market. Investor Michael Burry went further in November 2025, estimating that large cloud companies were under-depreciating AI infrastructure by roughly $176 billion between 2026 and 2028. That is an estimate, not a recorded loss, but it frames the scale of the debate.

The Cash Flow Versus Obsolescence Test

Nvidia also argues that its platform can move from one customer to another and that CUDA software keeps improving the performance of chips already installed. That may support cash flow, but it does not prove a liquid resale market. Airlines use aircraft as collateral because contracted revenue repays the loan before resale value becomes decisive. AI lenders need the same match between debt and hardware life. The $500 billion effort will only work if cash flow grows faster than the machines age.

What This Means for Lenders, AI Operators and Investors

• For lenders: before approving GPU-backed debt, require answers to four questions: who has committed to use the capacity, whether the loan amortises before the next hardware refresh, who can reallocate the machines to another customer, and who takes the first loss if resale value misses forecasts.
• For AI operators and cloud providers: negotiate multi-year capacity commitments that generate cash flow before the equipment becomes obsolete; this is the practical test the first deals will have to pass.
• For investors in the financing partners and Nvidia: treat the $500 billion figure as a target, not a completed transaction. The first definitive agreement, and its residual value backstop terms, will reveal whether the asset-class label is supported by contract economics.