Deepseek’s dirt-cheap AI puts American tech companies’ frenzied AI spending in perspective
Perhaps what’s most unsettling the markets about China’s powerful new AI model DeepSeek is its price tag — its very small price tag. The company said it spent just $5.6 million to train its latest model. That’s a small fraction of what its US competitors are spending.
OpenAI, for example, spent more than $100 million training GPT-4, and it appears that DeepSeek beats OpenAI’s leading models.
America’s biggest tech companies are tripping over themselves to shell out more and more money for their AI ambitions. Last quarter, Amazon, Alphabet, Microsoft, and Meta spent nearly $60 billion in capex, much of it earmarked for AI projects, and that number will likely go up when those companies report earnings in coming weeks. This year alone, Meta has committed to spending up to $65 billion in capital expenditures, thanks in part to a miles-long AI data center.