Nvidia CEO Jensen Huang has a new message for anyone worried about AI chips aging like milk. He says Blackwell GPUs, along with the upcoming Vera Rubin architecture, could remain useful for seven or eight years.
That matters because a lot of financial models assume these chips last about four years. Double the lifespan, and the math behind trillion-dollar data center buildouts starts to look very different.
What Huang actually said
Huang made the remarks in an interview on October 21, 2026. His claim covers both Blackwell and Vera Rubin.
The headline number is simple. Huang puts the useful life of these chips at seven to eight years.
That is a sharp break from the roughly four-year lifespan that often gets baked into financial models.
Huang tied the lifespan argument directly to financing. His view is that longer-lived chips carry more collateral value. More collateral value, in turn, can support more favorable credit terms for the deals that pay for AI infrastructure.
The trillion-dollar backdrop
The lifespan comments land alongside some eye-catching demand figures. Nvidia projects $1 trillion in combined revenue from Blackwell and Vera Rubin through 2027.
That is up from a $500 billion projection previously. So the company has effectively doubled its visibility into demand for its two flagship architectures.
Blackwell entered volume production in 2025, and Vera Rubin is anticipated to ramp in the second half of 2026.
The obsolescence problem Nvidia is trying to solve
CoreWeave has extended contracts on Nvidia’s A100 GPUs, which launched in 2020, so they now run until 2029.
That would put the A100 at nearly a decade in service.
Operators are increasingly moving GPU depreciation schedules to five to six years, as older models continue to hold economic value. Demand for older GPUs remains strong even as newer models arrive.
Why depreciation is suddenly interesting
When a company buys a GPU, it spreads that cost over the chip’s expected useful life. A shorter assumed life means bigger annual expenses and thinner reported profits. A longer one spreads the hit out and makes the investment look healthier year to year.
The gap between four years and eight years is not a rounding error. On the scale of spending Nvidia is describing, it can reshape how lenders, investors and buyers view the entire buildout.
What this means
For operators like CoreWeave, their business models depend on squeezing revenue out of hardware for as long as possible. Contracts that stretch older GPUs into 2029 suggest customers are willing to keep paying for compute that is no longer cutting edge.
There is a reasonable caveat here. Huang leads the company that sells these chips, so his optimism about their longevity is not exactly a neutral assessment. A seven-to-eight-year lifespan is his estimate, and the industry’s depreciation schedules of five to six years still sit below it.
The research suggests that hardware supply constraints could soon be overshadowed by limits on power and data center space.
For investors watching Nvidia, watch whether the $1 trillion projection through 2027 holds as Vera Rubin ramps. Watch whether more operators follow CoreWeave in extending older GPU contracts. And watch whether depreciation schedules creep closer to the seven or eight years Huang is describing, or stay parked at five to six.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.