Poolside

And can they subsidize with venture capital $5 to $10 per user per month? Sure.

But if they have to over-monetize it, what if it, they become Poolside? They could become the next Poolside.

Like, it's great, but literally I've got 10 million users at 10 bucks a month.

Now I've got a $1.2 billion nut a year to pay off.

It's like a human LLM.

You can be convincing on whatever you want me to...

Pulling them both together, right, I actually think, and I want to take the Poolside on that.

It's in, what you basically said is this is core to Meta, and it feels like something that they would, would want to do.

Like, oh, you know, lots of politics going on there.

But in all of that swizzle, a little lab, I don't know if little, but a lab that I had not yet heard of actually had some decent models before, launched a model called Poolside, or sorry, it's Poolside AI's lab, and they launched their Laguna small model, which allegedly started scoring around Sonnet 5 levels.

So Wow.

And what's kind of key is that one, it's an American lab.

maybe maybe some claw duct tape if that counts but uh

i think it definitely counts

yeah yeah well so just uh agentic agent uh engineering work uh so to speak but yeah uh i mean so i guess yeah they have a website i don't know i think it's a startup based in california somewhere but uh Cool name, Poolside.

Very fitting for the current summer vibes here.

I'm not...

There are some decent open weight models, and Nvidia, God bless them, is investing $10 billion with a company called Poolside to make strong open weight local models that you can run yourself.

And Nvidia is smart, because they sell the hardware, right? [laughs] So they don't care if you're, if you're buying the software from them.

They, they, they w- they, just like Apple, they want to sell you the hardware.

Our customers, which by the way, like 90% of our revenue comes from running our customers' custom models.

It's either our customers are labs who have pre-trained their own models, labs like Poolside and Inception and Cartesias and a bunch of other companies.

And then a bunch of companies who have post-trained their own models, which has gone through massive validation evals to ensure that they are behaving.

So, so far it has been mostly our customers taking care of that.

Um, and we're seeing a lot of open source growth, not just from Chinese models, but from all of the, the Western models as, as well.

Um, companies like Poolside with their Laguna models, companies like Cohere and Mistral that, that have models as well as all of the ones out of China, DeepSeek and Qwen and Kimi.

Um, those are getting a lot of traction because they have been price performant, and they have been able to get, um, get a lot of, a, a lot of usage, a lot of volume usage, uh, among s- a certain type of enterprise and a certain type of, type of workload.

That's, we think, is starting to shift because what you're, what we've seen in the most recent frontier model releases, and it's why these, these models that have come out in the last couple of weeks are, are so important, is these models are now not only the most performant models in terms of, of absolutely pushing what can be done with AI, but they're also economically efficient when you look at it on a cost per task basis.