Originally posted on X: after the CoreWeave CEO discussed the useful life of GPUs on a podcast, I publicly raised several questions and comments for Mr. Intrador.
Please watch starting at 46:27.
Hello Mr. Intrator,
A few questions for you, sir.
What is the difference between your GPUs and customers compared to Nebius, such that Nebius’ GPUs have a useful life of 4 years while CoreWeave’s are 6 years? To the best of my knowledge, you both share Microsoft and Meta as clients and use Nvidia GPUs, the very same GPUs.
Let’s say CoreWeave indeed found an innovative and secret way to extend the useful life of Nvidia’s GPUs to 6 years or even more, as it seems you are going to use the same data center equipment for the next 12 years, until 2037, even though the cycle is now annual and not every two years, while Nebius is still lagging behind. Do you think Microsoft and Meta will be happy to use H100s in 2032 like a reliable 2019 Toyota rented from the neighbor next door, when they could terminate your contract and rent 4-year-old Feynman GPUs instead from your competitor?
Oh, and one more thing. I know it is very trendy now to define and create new accounting definitions and workarounds, but let me tell you something: it never worked.
Mar 18
at
10:52 AM
Relevant people
Log in or sign up
Join the most interesting and insightful discussions.