What do you do when your financial capacity to help your customers buy your products becomes stretched? You turn to Wall Street, of course.
That’s what Nvidia has done, enlisting some of the biggest names in private capital – Goldman Sachs, Apollo Global, KKR, Brookfield and Black Rock – to arrange a pool of $US500 billion ($708 billion) or more of capital to help fund purchases of Nvidia chips.
The firms haven’t actually committed that much capital – they’ve signed memorandums of understanding, not lines of credit – and every project brought before them will be individually evaluated.
Nevertheless, Wall Street has provided a signal of intent.
With Nvidia’s own ability to continue to effectively provide vendor financing for purchases of its chips – the “circular financing” arrangements that have been increasingly scrutinised and criticised – perhaps constrained, it wants to enlarge the circle – and it’s Wall Street’s access to vast pools of private capital that will enable that.
Not that Nvidia won’t contribute. It says it will guarantee 25 per cent of the residual value of the collateral for any project.
For the Wall Street firms, the allure is that it could create a new class of asset-backed securities that they can on-sell to pension funds, insurers, sovereign wealth funds and wealthy investors, clipping the tickets on the way through.
It refers to those projects – the chips and data centres and other infrastructure that support the training of AI models – as “AI factories,” but two-thirds or more of the value of those factories will lie in the value of its own chips. That’s potentially a $US125 billion exposure for Nvidia, its shareholders and its creditors.
For Nvidia, the concept is designed to increase the capacity of its non-hyperscaler customers to fund the purchases of its chips.
Sharemarkets and conventional lenders have become warier, and more discerning, about the AI companies they are prepared to back now that the hyperscalers are out-spending their cashflows and competing for equity and debt. The cost of equity and debt for the smaller entities is becoming prohibitive.
For the Wall Street firms, the allure of the concept is that it could create a new class of asset-backed securities that they can on-sell to pension funds, insurers, sovereign wealth funds and wealthy investors, clipping the tickets, naturally, on the way through.
The “AI factories,” are, however, rather different to the asset classes that are usually securitised.
The cash flows from securitised residential and commercial mortgages, or infrastructure assets, or corporate loans are high quality and supported by cashflows and collateral that tend to increase in value over time.
The core assets of the AI factories, the Nvidia chips, have more limited and uncertain lifespans. The income streams they might generate and their value as collateral is less certain, but almost certainly diminishes over time.
There’s a debate about the useful lives of chips, given that Nvidia (and others) produces a “next generation” chip with superior capabilities every two years or so.
Some argue that their value should be fully depreciated over only three to five years. Nvidia says their useful life can be up to a decade.
There’s no doubt, however, that an outdated chip whose use is now to apply the knowledge gained from the training of AI models is less valuable than the latest generation chips being used to train those models.
The decline in value over time – and the time period is unknowable today – would erode the value of the collateral, with Nvidia’s guarantee providing the backstop for investors.
What Nvidia and its Wall Street are depicting is a form of collateralised debt, with Nvidia holding the “first loss” tranche that will be called on if a loan defaults and the rate of chip obsolescence is greater than anticipated.
Wall Street is good at financial engineering a solution to difficult financial challenges and, faced with funding the biggest investment boom in history, there’s no bigger challenge than supplying the capital to acquire the chips and infrastructure for artificial intelligence.
It’s not just the scale of the demand for capital – the hyperscalers ( Google, Meta, Amazon, Microsoft, Oracle) alone are investing about $US750 billion, largely in AI, this year and will spend more than $US1 trillion next year — but the current reality is that AI revenue growth is being far outstripped by the growth rate of the spending.
It appears near certain that those hyperscalers, despite their vast legacy cashflows, will all be cash-negative, and more highly leveraged, next year.
Google has raised new equity and most of the big tech companies pursuing AI leadership ambitions have, perhaps for the first time, begun tapping debt markets, with about $US350 billion of AI-related bond issues so far this year. Despite the cash from their non-AI businesses, the spreads on those bonds have been blowing out.
Wall Street firms can be ingenious but, as the global financial crisis demonstrated, they sometimes push the risk envelope too far.
Independent AI promoters like OpenAI and Anthropic have raised their capital from private investors and are, like Elon Musk’s SpaceX, planning initial public offerings to access the bigger pool of equity capital in the listed market. Any debt they have raised – and SpaceX has raised debt – would be priced as “junk” and be costly.
At those companies’ rates of investment, however – and the market’s satiation with all things AI – the planned listings might provide a bandaid, not a cure, for their financial stress.
At face value, gaining access to the estimated $US22 trillion of private capital could provide the extra funding required, and provide it on more attractive/less expensive terms than are currently available to individual AI borrowers.
Securitising debts – packaging up parcels of loans and therefore diversifying the risks – would create securities with much stronger credit quality than that of an individual AI borrower.
As was discovered in 2008 and the sub-prime mortgage crisis, however, bundling poor credits together doesn’t necessarily alter the reality that they are risky loans that could all go south together.
As the Nvidia chips age, the value of the collateral that the end-investors holding those bundles rely on would be diminishing.
And even that diminished value would be vulnerable as inference chips become increasingly commoditised. That’s the segment of the sector – the applications of AI – that China is strategically targeting, with chips whose prices already significantly undercut their US counterparts. The collateral, in the event of a default, could be near-worthless.
The Wall Street firms are no doubt comforted by Nvidia’s guarantee, but the tech giant is at the epicentre of the boom in AI and AI valuations and has – due to its myriad of circular financing deals in a sector where there are now layers of circularity to financial relationships – a leveraged exposure to the fate of that boom.
The degree of financial engineering already pervasive in the sector and the extent to which the off-balance-sheet liabilities of AI’s blue-chip companies, estimated at more than $US1.6 trillion, now surpass their debt on the balance sheet, say that all the conventional funding options for AI in the US have been stretched to their limits.
In a sense, the agreement Nvidia unveiled with the Wall Street players underscores how fragile the AI ecosystem now is.
Unable to raise sufficient funds conventionally, the sector is being forced to turn to a bespoke, unconventional, financially engineered structure to try to sustain its insatiable appetite for risk capital.
Wall Street firms can be ingenious but, as the global financial crisis demonstrated, they sometimes push the risk envelope too far, with unpleasant consequences.
For an industry already laden with risk, is the creative structure they envisage to fund Nvidia’s customers a step too far?
The Market Recap newsletter is a wrap of the day’s trading. Get it each weekday afternoon.
From our partners
About the Author
