A YouTube video by Justin R. Lester introducing DeepSeek (also written “Deepseek”), a new AI model from China that claims to match top OpenAI/Google models at much lower training cost (~$5.6 million). The presenter frames the claim’s potential market impact—cheaper training could disrupt chip demand and investor sentiment (e.g., pressure on Nvidia)—and signals skepticism, promising follow-up analysis.
Chapters
Okay, so let's talk about Deep Seek. If you've been anywhere near Tech Twitter or Tech X or the AI subreddit, or even financial news or any podcast, you've seen all this hype about Deep Seek. Some are calling it the biggest AI leaps in chatgpt other than smoke and mirrors. I'm here to help you guys out. Before I get into it, be sure you like and subscribe. Deepseek R1 is this new AI model out of China claiming to match the best AI models from OpenAI and Google at a fraction of the cost. So what's the big deal? Because training AI models cost hundreds of millions, sometimes billions. Deep Seek said they did for 5.6 million. Now if that's true, it changes everything. So investors panic and that's why Nvidia stock tanked because of AI training is cheaper if your companies need to buy expensive chips. But is Deep Seek actually a game changer or marketing stunt? Come back for the next video like and subscribe. I'm here to give you the business for your business and figure out if Deep Seek stuff really is going to hold up.