Nvidia's Jensen Huang defends GPU lifespans, CoreWeave financing and AI demand
Nvidia's CEO answered three doubts about AI spending: new cloud debt, how long chips last, and DeepSeek's inference pricing. He sells the chips, so weigh his answers accordingly.
The short version
- Huang said Nvidia GPUs can last seven to eight years, versus the four-year lifespan markets usually assumed.
- He argued that longer life and resale value make the chips usable as loan collateral.
- Nvidia has guaranteed the resale value of chips in some financing deals to lower costs for CoreWeave and similar firms.
- Investor Gavin Baker cited DeepSeek's reported GPU inference pricing as a sign of strong demand.
- Huang's summary of the argument: with demand, all the other problems can be solved.
Why it matters
Whether AI hardware spending pays off depends on how long chips stay useful and whether customers keep paying for them. Huang's answers are a position from the biggest chip seller, not proof. The real test is whether lenders and investors accept them.
Three worries, one interview
As Wall Street CN reported, Huang spoke to Yahoo Finance alongside Silicon Valley investor Gavin Baker after GTC Berlin. They took on the questions that have unsettled AI investors lately.
The first is the debt carried by "neoclouds," newer firms like CoreWeave that rent out GPU computing. The second is how long a GPU really lasts. The third is DeepSeek's reported inference economics, meaning the money earned from running a model for users.
How long does a chip last?
Companies spread a chip's cost over its expected life, a process called depreciation. Markets have commonly assumed four years. Huang said that is outdated and that the real figure is seven to eight years, possibly longer, because Nvidia keeps updating its CUDA software, which keeps older chips productive. He said its next chip, Vera Rubin, should last as long.
His financing argument follows from this. A chip that holds value and runs any AI model makes better collateral for a loan. He also said Nvidia has guaranteed resale value in some deals to cut borrowing costs for firms like CoreWeave.
Demand as the answer
Baker pointed to DeepSeek's disclosure that it earns far more per GPU hour than current market rental rates. He suggested this may explain why rents for Blackwell chips have been climbing.
Huang listed demand from OpenAI, Anthropic, Meta, xAI, Google and others. His line, in effect: with demand, all problems can be solved.
What to keep in mind
Huang also described a modular, container-based data center design meant to shorten the months between buying equipment and earning revenue, when interest is already accruing.
The claims come from a company that benefits from them, and the DeepSeek figures were relayed by an investor, not checked here. Treat the interview as the bull case.
What to watch
- Will lenders and credit markets accept the longer chip lifespan as collateral value?
- Do Blackwell rental prices keep rising, and does DeepSeek's pricing hold up under scrutiny?
Try it
When you read about AI companies' profits, look for the assumed chip lifespan. A longer assumption makes yearly costs look smaller.
Sources and links
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by Wall Street CN · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.