AI models that cost $1 billion to train are underway, $100 billion models coming — largest current models take 'only' $100 million to train

- Over 3.8 million GPUs delivered to data centers last year - Nvidia's latest B200 AI chip costs $30,000 to $40,000 - Dario predicts billion-dollar AI models by 2024 - Anthropic predicts $100 billion models soon - Hardware innovations by manufacturers like Nvidia, AMD, and Intel are keeping up with AI advancements - The article contains a web address error message indicating HTTP ERROR 404, meaning the page cannot be found - The content of the article is missing or inaccessible due to this error - Mention of a null element within the web address citation - Image links associated with the error message - Overall, the page is unavailable due to the 404 error

🚨 Reddit sentiment: mixed/negative - While there are positive sentiments around Nvidia and Palantir, most comments reflect concerns about the feasibility and business rationale of such expensive AI models. Summarized comments: - The cost of training AI models is exorbitantly high, raising concerns about diminishing returns in model size - The necessity of substantial power infrastructure for these models - Nvidia has created a viable business model from AI developments - Some users are bullish on Palantir - There is skepticism about whether the results will justify the costs, pointing toward potential limits in current AI techniques Stock tickers discussed: $PLTR, $NVDA