That's not how a smart business runs and that's now how Nvidia operates.

Jensen is smart. He's gone through over 30 years of tech cycles.

Nvidia actively commoditizes the LLM models. Look at Nemotron. They've avoided making a SOTA model solely to keep the hyperscalers (aka crack addicts) coming back for more GPUs.

As soon as the bubble bursts, they can release some open weight NemoMambaDiffusiontron and keep folks buying GPUs to run the damn thing.

It still wouldn't be smart to do so, as this would fall into the common business pitfall of thinking you could easily do the next stack layer of work.