The guarantee is not collateral
NVIDIA has agreed to backstop a quarter of its own chips’ residual value. What that promise is worth depends entirely on when you need it
On the 10th of August, NVIDIA announced financing platforms with Apollo, Blackstone, BlackRock, Brookfield, Goldman Sachs and KKR, intended to mobilise more than $500bn of third-party capital into AI infrastructure. The mechanism that makes it work is a residual value guarantee of up to 25%, assessed project by project, on NVIDIA’s own hardware.
Set aside the size for a moment, and the question of whether any of this is a bubble, which is being argued elsewhere by people with more time, and materially more expertise in the matter. The interesting thing for anyone who lends against equipment is that a manufacturer had to underwrite a quarter of the residual to make its own product financeable. That tells you what the market thinks a four-year-old GPU is worth, which is: nobody is confident enough to lend against it unaided. We have seen this before with EV’s, engines, or with new imported equipment.
It is an answer to a genuine problem. It is worth discussing what has been created with this type of backstop.
What a residual value guarantee is, structurally
A guarantee is not collateral. Collateral sits inside the structure: it is pledged, perfected, and it does not go anywhere if the counterparty has a bad year. A residual value guarantee is a contractual promise from a third party to top up a shortfall, and in the lender’s hands it is an unsecured claim on that third party. This risk transfer can be done with insurance or with the OEM, but It does not sit in the security package. It is not perfected against anything. It ranks with the guarantor’s other unsecured obligations.
In ordinary conditions this distinction is academic and slightly tedious, which is why it can help but can also just get waved through committee. But as some lessors have found it stops being academic on the day you call it.
Is the balance sheet big enough
The honest answer is that it depends on how far residuals fall, and the exposure behaves badly in exactly the direction you would not want. Is there cyclical contagion that increases the likelihood of the put? The contagion in EV’s or second hand vehicles caught the market out, albeit without global impact still caused huge stress in certain corners of the market.
But for this example and assuming one is dealing with a behemoth like NVIDA we can go with what is public. At its fiscal year end in January, NVIDIA held roughly $62.6bn in cash, cash equivalents and marketable securities, rising to about $80.6bn by the April quarter. Free cash flow for fiscal 2026 was around $96.6bn on operating cash flow of $102.7bn. By any normal standard that is an extraordinary balance sheet, and considerably stronger than the position of any vendor that has attempted something structurally similar before.
Now the arithmetic, with the caveat stated first. The $500bn is a platform target announced through memoranda of understanding, not committed capital, and the 25% is a cap applied project by project rather than an expectation. On those stated numbers, if the platforms reached their target and the backstop applied at the maximum across all of it, the covered notional would be on the order of $125bn. That is roughly one and a third times a record year of free cash flow, and something like one and a half to two times the balance sheet cash and securities.
The guarantee does not pay the notional, of course. It pays the shortfall to the floor, so the real exposure is the gap between the realised residual and the guaranteed one, capped at 25%. If residuals land at 20% against a 25% floor, the cost is five points. If the resale market for a superseded accelerator turns out to be thin, the cost approaches the cap. The loss is convex in the residual, and nobody knows the residual, which is the reason the guarantee exists.
|
THE PART THAT CONCERNS ME The contagion risk may be larger than the downside protection, and for a reason that has nothing to do with the size of the balance sheet. A residual value guarantee is called when residuals collapse. Residuals collapse when demand for AI compute disappoints. The guarantor’s revenue depends on demand for AI compute. The promise is drawn on precisely in the state of the world where the promisor’s capacity to pay is impaired, and it is drawn on by six of the largest alternative managers at the same time, against collateral that is correlated because it is the same collateral. Credit people have a name for this and it is not a compliment. The protection is real in the scenarios where you do not need it and thinnest in the one where you do. |
This has been tried, at a smaller scale
Vendor financing is not new and the precedent is documented. In the telecom build-out, Lucent committed around $8.1bn to financing its own customers, Nortel extended roughly $3.1bn with $1.4bn outstanding, and Cisco promised about $2.4bn. When traffic and pricing failed to arrive, the receivables went with them. Lucent’s bad loans ran from 2.6% of total loans at the end of 2000 to 60% a year later.
The comparison should be handled carefully, because it is a structural parallel rather than a prediction. Lucent was not generating $96bn of free cash flow, and the differences in scale and profitability are enormous, not cosmetic. What survives the comparison is narrower and more useful: when the vendor underwrites the demand for its own product, the reported strength of that demand becomes partly a function of the underwriting. That was the lesson then and it is a question worth asking now, respectfully and in public, of a company well placed to answer it.
The alternative is to not underwrite it
There is another way to lend against hardware, and it is duller. Decline to take a view on residual value at all.
If the asset generates a contracted payment stream, lend short against the contract rather than the box. Size the advance rate against what the equipment realises in a forced sale next quarter, not against what a guarantor says it will be worth in year four. Take the security over the receivable, put the cash through a pledged account, and let the facility amortise so that principal comes back on a schedule rather than on an exit. Under that construction the residual is upside you did not pay for, and the recovery does not depend on a resale market existing or a third party being solvent when you ring.
It finances less, and more slowly. That is the trade, and it is a real one, since $500bn of compute is not getting built on facilities structured like that. Both approaches are legitimate. They simply put the risk in different places, and an allocator should know which one it owns.
The question worth asking
For anyone holding hardware-backed paper, the diligence follows straightforwardly. Where does recovery come from if the guarantee is worth nothing. What rank does the guarantee hold. What is the advance rate assuming a residual of zero. Whether the facility amortises or relies on a refinancing. And whether the guarantor’s fortunes are correlated with the collateral, because if they are, you have bought one risk twice and paid for diversification you do not have.
It is possible that residuals hold, the compute gets used, the revenue shows up and none of this matters. That is the base case and it is a reasonable one. I am more interested in what the structure does if it is wrong, and I would be glad to hear from anyone who has seen the eligibility criteria and thinks I have this the wrong way round.
D3T Capital LTD is an Appointed Representative of Capricorn Fund Managers Limited, which is authorised and regulated by the Financial Conduct Authority. D3T Capital LTD is incorporated in England and the registered office is at Unit 17 Orbital 25 Business Park, Dwight Road, Watford, WD18 9DA, United Kingdom.
The investment products and services of D3T Capital LTD are only available to professional clients and eligible counterparties. They are not available to retail clients.
This article is general market commentary. It does not constitute an offer, invitation or inducement to invest in any product or service, and it is not directed at retail clients.
