Their main growth drivers are data centers, when demand will dry within 2 years, a bubble will pop. Especially when theoretical architecture of Neural Network change, the need for high performance will decrease.
Well, there are a period after the last Bitcoin bubble burst when the best way to get a good Graphics card for cheap was buying a used one from on the Bitcoin miner operations that were closing down.
Their main growth drivers are data centers, when demand will dry within 2 years, a bubble will pop. Especially when theoretical architecture of Neural Network change, the need for high performance will decrease.
See also: Sun Microsystems, who made tons of servers that drove the dotcom boom. They didn’t fare so well afterword.
This is a “grab the pile of cash and be happy” situation.
Then we’ll all get cheaper GPUs! Oh no!
No not really, the GPUs that datacenters are buying doesn’t even have a display output so they are useless to the vast majority of home users.
If you run a home lab then maybe.
Well, there are a period after the last Bitcoin bubble burst when the best way to get a good Graphics card for cheap was buying a used one from on the Bitcoin miner operations that were closing down.
CUDA has a lot of applications outside of AI. They’ll just refocus on the next bubble and will continue hoarding wealth.