The logic of the move is straightforward: Nvidia sells more chips, customers
get the money elsewhere to buy them, and much of the financial risk stays off
Nvidia's balance sheet. What is not settled is what happens when those chips
stop being the fastest on the market.
What exactly was announced
| What | Detail |
|---|
| Counterparties | Apollo, Blackstone, BlackRock, Brookfield, Goldman Sachs and KKR |
| Amount | More than $500 billion |
| Purpose | Finance customer purchases of Nvidia chips |
| Status | Memorandums of understanding, not financing already executed |
That last point is the one most often lost in the headlines, and it is worth
underlining: a memorandum of understanding is not debt taken on.
It is a statement of intent about how much capital these firms would be willing
to mobilize. The real figure will depend on how many of those deals actually get
signed.
Nvidia's thesis: the GPU as an asset class
Jensen Huang, Nvidia's chief executive, has argued that GPUs can become
a new asset class, comparable to real estate or
infrastructure. In other words: something you can lend against, because it holds
its value and generates income predictably.
Given how important they are to technology today, the idea is not far-fetched.
The problem shows up when you follow the analogy all the way through.
A highway can generate income for decades. A GPU can be
technologically obsolete in four or five years.
That is the whole difference. Physical infrastructure depreciates slowly
and predictably. A chip depreciates when the next chip arrives, and nobody
controls that.
The precedent: Lucent and Nortel in the late 1990s
This structure — the manufacturer helping finance its own customers to drive
its sales — is not new. Companies like Lucent and Nortel did
exactly the same thing with telecom equipment in the late nineties.
For a while, the system seemed to work perfectly. Sales were growing,
customers were buying, and everyone in the chain seemed to be winning at the
same time.
Until the cycle turned.
It is worth being clear about what this parallel does and does not mean.
It does not imply that the same thing will happen to Nvidia:
it is a different company, with different margins and a different market
position. What the precedent contributes is the right question: this kind of
structure has been tried before, and what determined the outcome was not the
growth phase but the contraction.
The question left open
If a GPU financed today loses much of its value in five years,
who absorbs that loss?
- Nvidia, which sold the chip and got paid for it?
- The customers, left with a depreciated asset and an intact debt?
- The banks and the large funds that put up the capital?
- The insurers that covered part of those deals?
That answer does not exist yet, and it is exactly the thing to watch. While
the chips are new and demand is maxed out, the question does not come due. It
comes due when they stop being new.
What to watch from here
- How much of that $500 billion actually gets deployed.
The gap between the memorandum and the money disbursed is the first real
measure of whether this is a financing structure or an announcement.
- The repayment terms. If the financing is structured over
terms longer than the chip's useful life, someone will keep paying for
hardware that no longer competes.
- What the quarterly earnings say
about these agreements. How they are accounted for — and whether or not
they show up on the balance sheet — is information in itself.
- Whether a secondary market for used GPUs appears. For an
asset to work as collateral, it has to be sellable. If nobody buys
second-hand GPUs at a reasonable price, the collateralizable-asset thesis
gets a lot weaker.
None of this is a recommendation to buy or sell. It is a description of a
financial structure and of the questions it leaves open. See the
legal notice.
Frequently asked questions
What deal did Nvidia sign with Apollo, Blackstone and BlackRock?
Nvidia signed memorandums of understanding with six financial giants — Apollo, Blackstone, BlackRock, Brookfield, Goldman Sachs and KKR — to mobilize more than $500 billion to finance purchases of its own chips. These are statements of intent, not debt already taken on: the real figure depends on how many deals actually get signed.
Why is using a GPU as loan collateral risky?
Because the collateral can lose value before the loan is repaid. A highway or a building depreciates slowly and predictably; a GPU can be technologically obsolete in four or five years, whenever the next chip arrives, and nobody controls that timing. If the asset falls faster than the debt, someone has to absorb the difference.
What happened to Lucent and Nortel in the late 1990s?
Both financed their own customers to sell telecom equipment. While the cycle held, sales grew and every link in the chain seemed to win at once. When the cycle turned, customers could not pay and the loss came back to the manufacturer. It does not mean Nvidia will follow the same path; it points at what to watch.
