Six of Wall Street’s biggest names. One announcement. And a chief executive asking investors to treat AI chips like toll roads — assets that keep earning long after the loan closes. The catch: nobody’s sure the chips will still be worth anything by then.
Nvidia announced Monday it’s partnering with six major financial institutions — Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR — to mobilise more than $500 billion in third-party capital for AI infrastructure. The arrangement lets Nvidia’s customers borrow against compute power itself, treating AI data centres the way lenders already treat commercial real estate or toll roads: durable assets that generate steady returns over years, not one-time purchases that lose value the moment they leave the warehouse. CEO Jensen Huang called the shift
on X: “We have moved from an era in which companies bought chips and built data centres project by project to one in which AI factories can be financed as productive infrastructure.”
What’s Happening & Why It Matters
How the Money Moves
Each of the six firms will build its own capital pool, at what Nvidia calls “attractive rates,” aimed at frontier AI labs, enterprises, and cloud providers who need to buy compute but don’t want to tie up their own balance sheets doing it. Financing runs through private offerings and bonds issued by special-purpose entities — vehicles capable of raising tens of billions at once. Goldman Sachs, the only bank in the group, is positioned to lead public debt deals while its asset-management arm distributes the returns.
Huang confirmed Nvidia itself might chip in financing support for up to 25% of any given deal, but the company’s real role is different: “help unlock a very large pool of independent capital while maintaining disciplined risk exposure.” Translation — Nvidia wants the capital, not the balance-sheet risk that comes with it.
The OpenAI Number Everyone’s Already Talking About
The $500 billion figure doesn’t exist in a vacuum. Nvidia was already
in talks to backstop as much as $250 billion to help OpenAI lease computing power from a 10-gigawatt Ohio data centre hub that SoftBank subsidiary SB Energy is building — a deal TF flagged as the trigger behind Sour Sentiment Triggers AI, Chip Stocks Sell Off just weeks ago. Axios reported it’s unclear whether that OpenAI backstop is inside the new $500 billion framework or runs alongside it as a separate commitment.
Either way, the scale keeps compounding. Nvidia’s stock has more than quadrupled since 2024, pushing its market valuation to $5.3 trillion. And the financing push is after the July market swoon TF covered, when investors first started questioning whether Big Tech’s AI spending would ever generate returns proportional to the money going in.
The Announcement’s Unanswered Question
Here’s the problem hiding underneath Huang’s toll-road analogy. Toll roads don’t become obsolete every 18 months. AI chips do — or close to it. Forbes raised the sharpest version of the question: if compute infrastructure ages faster than the loan terms backing it, who absorbs that loss? Amazon already shortened its own server depreciation periods, a quiet admission that hardware is ageing out of usefulness faster than companies planned for.
Strong rental prices today don’t guarantee strong resale value tomorrow, and that gap is what determines whether the financing model holds up or turns into the next place Wall Street discovers it overpaid. Moody’s has already warned, as TF covered in TF AI Round-up: 09 August 2026, that unprecedented AI capex is squeezing free cash flow and forcing tech giants into heavier debt loads across the board. The $500 billion doesn’t solve that pressure. It just moves a large piece of it off the hyperscalers’ own balance sheets and onto Wall Street’s.
TF Summary: What’s Next
The financing platforms are subject to final agreements, with deals expected to reach market within months, according to sourcing in The National. No confirmed structure exists yet for how the OpenAI Ohio backstop relates to the $500 billion commitment. Nvidia has not detailed specific loan terms, depreciation assumptions, or residual-value guarantees beyond Huang’s general 25% financing-support figure.
MY FORECAST: Expect the first concrete deals under the framework to close within six months, given how Huang called the announcement as ready to move rather than aspirational. The residual-value question Forbes raised is the one to watch closest over the next 18 months — if AI hardware keeps depreciating faster than the loan terms assume, expect the first real losses to surface at the moment tech commentators start asking whether the entire AI financing structure was built on an assumption that never held. Nvidia gets paid either way, selling chips regardless of who ends up holding the depreciation risk. That asymmetry is the actual story here, underneath the toll-road viewpoint.
Related Stories
- Sour Sentiment Triggers AI, Chip Stocks Sell Off
- TF AI Round-up: 09 August 2026
- Meta Stock Slides as AI Spending Outpaces Investor Patience