China-based startup DeepSeek became an AI standout this week by creating an AI model believed to be on par with leading models from U.S. startups — at a fraction of the cost. In a research paper released last month, DeepSeek said it developed its AI for under $6 million in only two months, a far cry from the $100 million it takes U.S. startups to train AI — and that’s on the lower end of the spectrum, according to Anthropic CEO Dario Amodei.
It quickly rose to the top of the app store charts, challenging the U.S.’s position as the world’s leader in AI. The release set off a race for AI dominance and shook Big Tech stocks, causing AI chipmaker Nvidia to lose almost $600 billion in market value one day and new competitor claims — from having an even better model to allegations of theft.
According to White House AI and Crypto Czar David Sacks, DeepSeek’s arrival shows that Chinese companies are “hot on our heels” but that the U.S. maintains its leadership in AI. He says DeepSeek’s AI is on par with OpenAI’s o1 model, which came out about four months ago.
“We basically have somewhere between a three and six-month lead on them [Chinese companies],” Sacks said. “But they are catching up very, very fast.”
DeepSeek. Photo Illustration by Justin Sullivan/Getty Images
ChatGPT-maker OpenAI says DeepSeek is copying it
OpenAI and Microsoft are investigating whether DeepSeek used large amounts of OpenAI training data without permission for its own AI. OpenAI told The Financial Times earlier this week that it had proof that DeepSeek used its large AI models to create its own through a process called distillation, in which one AI model learns from another like a student learning from a teacher.
Sacks backed up OpenAI’s claims in an interview with Fox Business on Tuesday.
“There’s substantial evidence that what DeepSeek did here is they distilled the knowledge out of OpenAI’s models,” Sacks said. “I think one of the things you’re going to see over the next few months is our leading AI companies taking steps to try and prevent distillation.”
Other industry leaders say DeepSeek’s success is due to the collaborative nature of open-source AI models.
DeepSeek “came up with new ideas and built them on top of other people’s work,” Meta’s chief AI scientist Yann LeCun stated in a Threads post on Saturday. “Because their work is published and open source, everyone can profit from it.”
Alibaba claims it has a better model
Chinese e-commerce company Alibaba is claiming that it has developed an even smarter model than DeepSeek’s.
Alibaba on Wednesday released a new AI model called Qwen 2.5 Max edition that the company says scored better than AI from Meta, OpenAI, and DeepSeek in leading benchmark tests, per Bloomberg.
“Qwen 2.5-Max outperforms … almost across the board [OpenAI’s] GPT-4o, DeepSeek-V3 and [Meta’s] Llama-3.1-405B,” Alibaba’s cloud division stated in an announcement on its official WeChat account, according to Reuters.
