“There is a sense of the models are going to be commoditized. And at the top end, that hasn't really happened, but there is a real sense that almost nobody actually needs these frontier models to do the things that they're trying to do. The sixth best model at this point is good enough for most of the things people want to do with AI.”

And you have to convince all of those people that your tool on top of your model is better than the vast quantity of specific enterprise software they already have, which is hard. Anybody who has ever tried to sell to an IT manager at a company knows that is very, very, very hard work. And they have to do that at such pace and such scale that just seems virtually impossible. All of this has moved faster than anybody expected. So like, who knows? But all of that is like, that is different from we just have to keep building bigger models. So we need to spend a theoretically infinite amount of money on compute. And to that point, actually, there is a sense in a lot of conversations I'm having with AI people and to some extent in this S1 that actually maybe we are going to fairly quickly hit a phase where just having a leading model is not how you win. That having better products or more tuned products or other things that a company like Google is much better set up to do, like vast quantities of customer support is just the thing that's going to be the actual differentiator rather than we have a slightly better model on…