🤖 DeepSeek proves frontier-model costs are collapsing

“But the point is that this model cost millions of dollars and was able to compete with a model that took billions of dollars to train. So my takeaway from that was that these costs are falling really precipitously. So that's sort of my one observation, is spending more money doesn't necessarily get you further along if the goal is progress.”

“And this is where I think, again, apologies to the NVIDIA bulls, but it's going to create a more heterogeneous environment. And the reason is because there's too much money and risk on the line to go through a single point of failure. A chip, a high-level framework to get to that chip, that's nuts. So I think like that, kind of like Emperor has no close moment is upon us.”

“Gavin Baker came on this podcast and said, this is the fastest deprecating asset in the world, was a large language model. He's been proven right. They're not worth anything. They're all going to be open source. They're all going to be commoditized, and that's for the best of humanity.”

“At the end of the day, if model performance continues to improve, get cheaper, and it's so competitive that it commoditizes much faster than anyone even thought, then the value is going to be created somewhere else in the value chain.”

“Here's the one thing on the NVIDIA thing that I would counter with a little bit of what's been said here, is like when AI gets cheap, you know what's going to happen, guys? There's going to be a lot more AI, right? I don't think... I think the price elasticity on this one is actually positive. So as the price goes down, the revenue usage, everything's going to go up.”

“people are waking up that return on invested capital and data centers will matter, that the models are basic commodities and super competitive. In the best case, it's kind of a land war in Asia, it's a melee. In the worst case, it's just total commodities.”

“What do they do? They just get more efficient, and they exported it back to us. They sent us back the best open source model when our guys were staying closed source for safety reasons.”

“So when Alibaba pushes Gwen this week, it was an exceptional model. It's probably one of the better open source, if not the best open source model out there. Deepseeks I think is also quite good. It doesn't get any press anymore. So I think we're starting to get to this place where there's such abundance, that actually people are just kind of like overwhelmed with all the choice and they don't know how to differentiate.”

“And I think the only reason Nvidia was able to grow through this product transition is because of reasoning models like DeepSea, which are just so compute hungry.”