With limited compute capacity, how do AI companies balance inference vs training needs?

For me, when I look at it, I kind of look at it and go, they're going to have to have certain revenue milestones that they need to hit in order to continue to exist. Break down your split so that you can hit those revenue milestones, but while also, you know, leaving enough cash to be able to invest in the, you know, invest in the future. Yeah, that would be how I think about it because I promise you, if they go to the IPO and the numbers aren't right, yeah, people talk about a bubble, or, you see a bubble of epic, disproportionate levels. If they go into IPO and the numbers are not right and the margins are not right and people are starting to question the return, you can, you can forget about the whole thing. I, I think it makes sense why, like you said, this year they've probably increased inference, because now they're working towards IPO. Now it's more important that the books look good. Yeah. And then after the IPO, then they can now figure out, okay, you know what, they've got some money in, now maybe we can shift it up a bit.