Nvidia (NVDA +0.22%) signed memorandums of understanding Monday with six of Wall Street's largest investment firms to establish independent financing platforms -- vehicles meant to mobilize more than $500 billion of third-party capital for artificial intelligence (AI) infrastructure. The six are Apollo Global Management, BlackRock, Blackstone, Brookfield Asset Management, Goldman Sachs, and KKR.
The idea is to let lenders treat computing hardware the way they treat buildings or toll roads (an asset that produces revenue and can therefore secure debt).
Markets have run a version of this experiment before. In the late 1990s, telecom equipment makers lent their customers the money to buy their gear, and by the end of 2000, nine suppliers (including Lucent, Nortel, and Cisco) had extended about $25.6 billion, according to a McKinsey estimate.
Nvidia's new target is about 20 times that figure. The structure is different this time, and the difference matters. But the announcement leaves something open that may matter more.
Image source: Nvidia.
Compute as collateral
The platforms are designed to create dedicated pools of outside capital for Nvidia's customers (the companies building AI data centers) so the next wave of hardware isn't paid for out of the buyers' own cash.
"This is really the first time that technology chips have become an investable asset class," CEO Jensen Huang told CNBC on Monday.
The market's response was cool. Shares slipped about 3% Monday and trade near $218 as of this writing.
Note what the plan is not: a response to weak demand. Nvidia's trailing-12-month revenue rose about 71% to $253 billion, and its net income roughly doubled from the year before, to about $160 billion. In the fiscal first quarter of 2027 (the period ended April 26, 2026), revenue grew 85% year over year, accelerating from the quarter before, and management guided for about $91 billion in the quarter it reports next. The sequence is still climbing.
The financing exists because the bill is starting to outgrow the buyers. Rating agencies have warned that record capital spending has begun squeezing the big AI spenders' free cash flow and pushing them toward heavier debt loads. In other words, the next $500 billion of hardware needs more money than the customers' own operations throw off.
The telecom precedent
The last time an industry's suppliers arranged their customers' financing at anything like this scale, it ended badly. Network operators in the late 1990s demanded financing from equipment vendors as a condition of awarding contracts, and the vendors obliged with their own money.
When spending collapsed, 24 of the 30 largest publicly traded telecom carriers went bankrupt, and an estimated one-third to 80% of the suppliers' loan portfolios were lost. Lucent and Nortel were pushed to the brink of insolvency.
That history, to me, is the right measuring stick. Against McKinsey's count of what the era's nine biggest lenders had extended, more than $500 billion is roughly 20 times as much. Broader measures of the period's vendor lending run higher, which would shrink the multiple. On any measure, though, the new pool is far larger.
The difference is who carries the loans. Lucent's sat on Lucent's balance sheet, which is why its customers' failures nearly became its own. Nvidia isn't the lender here. The platforms are independent, and the money is Wall Street's.
NASDAQ: NVDA
Key Data Points
Who takes the loss?
That is the part Monday's announcement doesn't finish explaining. The release describes the platforms as independent and the capital as third-party. It says nothing about who absorbs the damage when a borrower defaults.
The collateral case comes from Huang himself. Chips can be moved from one customer to another, he argues, and they keep improving through Nvidia's CUDA software. A lender could, in principle, take the hardware back and lease it to someone new.
Maybe so. But a repossessed chip is only worth what someone will pay for computing at that moment, and each new chip generation can cut the price of the one before it. If a borrower failed because AI demand cooled, the collateral would be cooling with it.
And financed demand is still demand. After all, hardware bought with borrowed money shows up in Nvidia's revenue exactly the way hardware bought with cash does. In the telecom collapse, that was arguably the demand that vanished first.
Ultimately, I think the platforms fix the failure that defined 2000 -- the vendor betting its own solvency on its customers' success. Nvidia's balance sheet stays out of the lending, as the release describes it, and the business it is financing demand for keeps posting extraordinary numbers.
What the structure can't change is what the money is for -- hardware that gets paid back out of AI revenue that mostly doesn't exist yet. The announcement doesn't say who takes the loss if that revenue falls short.
