Palantir CTO Shyam Sankar said DeepSeek's latest models highlight how models are commoditizing and "the price of inference is dropping like a rock."
Speaking on Palantir's blowout fourth quarter earnings call, Sankar also said that DeepSeek also shouldn't be underestimated. Sankar said:
"One of the things I want to make sure we all do realize is that the engineering in R1 is exquisite. The optimizations that they've done are really impressive. And I don't think you can get away with the facile explanation that the Chinese just copy and we're the only innovators, we have to wake up with the respect for our adversary and realize that we are competing."
All of that said, Sankar went on to say that DeepSeek likely stole a good bit of that innovation.
"They absolutely did steal a lot of that through distillation of the models and perhaps they stole even more.
And then you can look at the GPU sales growth in Singapore. It's a tiny island nation. I'm pretty sure there's some sanction invasion going on there, and we have to realize that the AI race is winner-take-all and it's going to be a whole of nation effort that extends well beyond the DoD in order for us as a nation to win."
Another theme from the DeepSeek rise is that models are commoditizing and that'll have benefits for enterprises. However, enterprises should note that DeepSeek usage on the web goes to China for processing. If you download the model open source and work on it via a hyperscaler, the data stays put. We saw DeepSeek put through its paces side-by-side with other models and it appeared to need a good bit of tuning.
Sankar said:
"I think one of the obvious lessons of DeepSeekR1 is something that we've been saying for the last two years, which is that the models are commoditizing. Yes, they're getting better across both closed and open, but they're also getting more similar and the price of inference is dropping like a rock. But I think the real lesson, the more profound one is that we are at war with China. We are in an AI arms race."
