Nvidia's financial engineering
Building AI models requires tremendous upfront investment.
We are seeing unprecedented spending by the hyperscalers on compute and other related infrastructure. Tomasz Tunguz wrote on Aug 5, 2026:
SpaceXAI reported $18.37b of capex in its first public quarter, $15.83b of it AI infrastructure, against a $13.2b consensus. Amazon spent $53.1b in the same quarter, Alphabet $44.9b, Microsoft $41.0b & Meta $31.1b, & sequential dollar additions were comparable across all five. The difference is funding : operating cash flow covers 155% of capex at Microsoft & 106% at Meta but only 12% at SpaceXAI, with Oracle at 89%, Alphabet 87%, Amazon 84% & CoreWeave 39%.
[Spending Like a Hyperscaler Tomasz Tunguz](https://tomtunguz.com/the-newest-hyperscaler/)
Hyperscalers possess free cash flows to redirect to capex spending. They still have to face investor scrutiny on earnings calls about how they see these investments panning out.
The frontier labs, neoclouds and startups, on the other hand, have to raise capital for every chip they buy.
The traditional financing methods are punitive.
At this order of magnitude, issuing equity dilutes founders heavily. Raising corporate debt, on the other hand, is incredibly hard and expensive, with adverse effects on credit rating.
The consequence of this financing difficulty lands on the compute providers, since that is where the money was headed. Among them, Nvidia leads the charge.
To maintain its extraordinary sales growth, Nvidia needs the next wave of buyers (who lack hyperscaler balance sheets) to be able to afford its chips.
In view of this, Nvidia has gone further than any tech firm previously has, and partnered with leading asset managers on Wall Street to create a new financing instrument. From CNBC:
Nvidia signed memorandums of understanding with Apollo Global Management, Blackstone, BlackRock, Brookfield Asset Management, Goldman Sachs and KKR to establish financing platforms for Nvidia’s customers.
The effort aims to mobilize more than $500 billion in third-party capital for hyperscalers, frontier AI labs and enterprises to build out data centers and acquire Nvidia hardware.
By using institutional credit, insurance funds and private capital to underwrite GPUs and data centers, Nvidia is helping its end users secure financing without tapping their own balance sheets.
Think of it as buying a car or a home. The bank gives you a loan where the underlying asset (car or home) serves as collateral.
This is what Nvidia is trying to engineer with the asset managers.
The chips serve as collateral for the loans that buy them.
The underlying assumption here is that Nvidia’s chips will retain their economic value over time and can be repurposed and redeployed if the borrower defaults.
The following are some interesting observations to note:
- In the most recent earnings calls of the hyperscalers, there has been a new trend of increasing depreciation timelines for chips to 5-6 years (from 2-3 years). This is primarily being done to lower annual non-cash expenses, which artificially increases operating income and EPS.
- Nvidia releases new architectures every 18-24 months. For training frontier models, chips become technologically obsolete every 1-3 years for two reasons:
- Advent of newer, more powerful and efficient chips
- Extreme thermal and electrical stress that the chips are subject to in high utilisation datacenters
- The chips have a secondary cascade life. Once they are too slow for cutting-edge model training, they are repurposed for less demanding tasks like real-time inference, fine-tuning, or software development.
Two of these observations cut against the collateral thesis and the third, the cascade life, supports it. The chips do hold value, but only at a discount and on a shorter clock than the depreciation timelines imply.
This can very well turn out to be a masterstroke by Jensen. But financial engineering has traditionally also led to catastrophic downsides when the assumptions have not played out as expected.
In the financial crisis of 2008, the US housing bubble collapsed leading to the crumbling of two fundamental assumptions backing mortgage-backed securities: 1. housing prices would always rise, 2. not everyone will default at the same time.
At the outset, Nvidia’s financial engineering seems smart, but the assumptions are the same shape here. Every one of these loans is backed by the same chips and betting on the same demand. If that demand softens, the borrowers and the collateral go down together. While the risk remains, as a techno-optimist, I hope nothing of this sort happens here.