GPUs as a new asset class
NVIDIA (NVDA) is helping unlock more than $500 billion in financing for AI infrastructure. This is another wave of hyperscaler-type capital spending flowing into the AI boom. It does not come only from NVIDIA (NVDA). Some of the world's biggest credit firms are involved: BlackRock (BLK), Blackstone (BX), Goldman Sachs (GS), and Apollo (APO).
They are taking NVIDIA (NVDA) GPUs, which the company calls "NVIDIA AI factories," and turning them into a brand-new asset class that can be financed. The main effect is that smaller players - the Neoclouds and other companies building AI outside the hyperscalers - can now get the money they need. That widens the pool of clients buying GPUs at scale.
Why chips can be financed like cars or planes
This changes the economics of the AI industry. GPUs last much longer than most people assumed. Jensen Huang of NVIDIA (NVDA) has said this for a while, but the market did not grasp it. CoreWeave (CRWV) reported earnings this week and said it made a deal for A100 chips that runs until 2029 - chips first made in 2020.
The pattern of GPU life:
- About two to three years used for training frontier models.
- Another three years on inference.
- Another three, four, or five years on niche training and niche work for enterprises.
Because their useful life stretches this long, GPUs can be financed like a car or an airplane. That makes the Neoclouds and other GPU buyers more profitable on the capital they spend.
Who benefits
NVIDIA (NVDA) gains directly. Financing makes chips cheaper to buy, so demand rises. NVIDIA (NVDA) is also part of the financing itself, taking roughly 25% of some of this funding, so it earns money that way too. The credit firms earn revenue from the deals.
The hyperscalers benefit as well. They spend hundreds of billions of dollars a year, a large share on GPUs. They are now learning these assets last longer than the five or six years they currently use for depreciation. Expect this to show up in their earnings over the coming quarters as longer depreciation schedules.
Power companies benefit too. Longer-lasting data centers and GPUs mean more sustained power demand, and older GPUs actually use more power as they age. Power companies, data center companies, and the wider infrastructure trade all gain from this shift.
The chip-lifespan debate
Some skeptics argue chips have a short useful life. The main voice here is Michael Burry, who thinks GPUs last only three years. That is why he is short NVIDIA (NVDA), and also short Micron (MU) and Nebius (NBIS) - a position that is not going well for him right now.
A chart from the past couple of weeks looked at GPU rental prices over the last year. Every GPU class is up over the year, even the A100s, which are five to six years old. The market got this wrong and is only now understanding how much these GPUs can do.
Market and Fed backdrop
The current setup is a "Goldilocks" market. The economy is booming and earnings are strong, especially in Q2. Inflation has come down over the last two months, largely because oil fell since the start of the Iran-US war. Jobs are staying roughly flat.
A big part of this ties back to AI. The labor force participation rate is falling. Boomers are finally retiring, and their jobs can be filled by younger workers using AI who can do the work of one, two, or three older-generation workers. That is good for earnings and keeps inflation down. If that trend holds - and how long it lasts is unclear - it is a very good moment for the stock market.
