Here's something I just read about the subject:
DeepSeek's claim is that they developed this great language model for only $5.5 million and that they did not use high-tech chips such as the latest Nvidia chips, the H100s, while doing this, they achieved this with old generation Nvidia GPUs.
Why is that? Because, as you know, the US is imposing a technological embargo on China. That's why DeepSeekers found our own new solutions, made innovations and thanks to this, there is no need for Nvidia chips, AMD chips, Arm chips. At least that's how the market reads it, they don't actually say it and that's why there is a big sale
So basically NVIDIA is seen overvalued if true that their latest products aren't needed to create these LLMs.