Can't wait for this bubble to burst..

It doesn't really make much sense to go faster than silicon node changes unless there is a lot of optimisation on architecture that needs doing. Historically all these refreshes between nodes were largely pointless with small benefits and preparing them took development effort away from the big changes. It's progress in silicon that matters and brings the performance improvements and moving to a faster cadence hasn't historically worked out well.

This is the best summary I could come up with:

Until now, Nvidia’s produced a new architecture roughly once every two years — revealing Ampere in 2020, Hopper in 2022, and Blackwell in 2024, for example.

(The industry darling H100 AI chip was Hopper, and the B200 is Blackwell, though those same architectures are used in gaming and creator GPUs as well.)

Huang says Nvidia will accelerate every other kind of chip it makes to match that cadence, too.

“New CPUs, new GPUs, new networking NICs, new switches... a mountain of chips are coming,” he says.

Huang also shared a couple of his sales pitches on the call by way of explaining the incredible demand for Nvidia’s AI GPUs:

Nvidia’s CFO interestingly says that automotive will be its “largest enterprise vertical within data center this year,” pointing to how Tesla purchased 35,000 H100 GPUs to train its “full-self driving” system, while “consumer internet companies” like Meta will continue to be a “strong growth vertical,” too.

The original article contains 377 words, the summary contains 155 words. Saved 59%. I'm a bot and I'm open source!