China's chatbot juggernaut DeepSeek could mark the end of US supremacy in AI

If you believe the hype surrounding China’s new chatbot, DeepSeek, building a top-tier AI model might cost about the same time and money as your dream home, complete with a tennis court, pool, and water feature—roughly two months of searching and $5.5 million. In some areas, developing new AI might even be cheaper and faster than entering the real estate market.

There’s reason to believe DeepSeek's creators, a Chinese company of the same name, have exaggerated their frugality—more on that later—but it would be a mistake to ignore the global panic it has sparked. DeepSeek rapidly ascended to the top of app stores over the weekend and wiped over half a trillion dollars off the market capitalization of the world's largest companies and Nvidia, a leading AI chip supplier, in Monday's trading.

This has openly shaken the world’s previous steadfast faith in Silicon Valley’s top labs, which have insisted that the only path to a glorious AI future is paved with as many expensive computer chips as possible. Now, it appears DeepSeek has overturned one of the tech world's most sacred tenets, achieving similar success with fewer and older chips. "This marks the end of US dominance in the AI race," said Toby Walsh, chief scientist at the University of New South Wales' AI Institute.

“It’s like running a four-minute mile,” he said. “When Roger Bannister first ran a four-minute mile, it proved it was physically possible.” Unlike Bannister, DeepSeek has conveniently published much of its how-to manual on the internet. Some, but not all, of DeepSeek’s code and technical explanations are open-source, meaning anyone can view, download, and use them.

“This isn’t just going to stop with this Chinese company DeepSeek, which has managed to match the best in the West,” Professor Walsh said. “It shows that it can be done by all of us… not just in China, possibly even in Australia.” So, how much of DeepSeek’s story is actually overblown?

As for what DeepSeek has actually done, some of the claims don’t hold up, or haven’t been thoroughly tested. Many experts are deeply skeptical of the claimed $5.5 million (A$8.8 million) price tag and two-month timeline. “I think that needs to be questioned,” said Jenny Wong Leung, a senior analyst at the Australian Strategic Policy Institute (ASPI). The AI model in question, DeepSeek-R1, is not the company's first attempt. "The $5.5 million over the past two months was specifically for the latest version and built on previous investments since 2021," she said.

The US government imposed export restrictions on the most advanced AI chips, known as GPUs, to Chinese companies in 2022, but she notes DeepSeek’s founders were reportedly stockpiling these chips well before the ban. In other words, be wary of DeepSeek’s hype. “It’s very likely a framing to get maximum publicity,” said George Buchanan from RMIT. “[It] is very likely heavily supported by the Chinese government to highlight Chinese innovation.”

So, DeepSeek’s engineers didn’t just cobble together their chatbot genius over a lunch break, but this doesn’t diminish Professor Walsh’s assessment of the achievement. “It might be $12 million or $24 million, but whatever it is, just matching [OpenAI’s latest model] is impressive,” he said. In terms of the AI race, he believes DeepSeek’s model has “put China on a par with the US.” This is an unwelcome shock to US AI leaders, and analysts believe this will provoke a response.

The losses in tech stocks have been dramatic, but don't expect the AI bubble to burst anytime soon, says ASPI's Wong-Leung. “Whether justified or not, the fundamental belief in the power of AI to change our economies and societies remains very strong,” she said. “This disruption will likely lead to a recalibration of investment globally in the sector, but not a collapse.”

If anything, experts believe the US government is likely to double down. “From a geopolitical perspective, the US is very likely going to want to stay ahead of China,” said Jenny Paterson from the University of Melbourne. “I don’t think this is going to quell AI investment,” she said, although she warns the competition itself presents risks. “What we should be talking more about is whether this is the end of responsible and safe AI.” Even before DeepSeek's breakthrough, the brakes were being applied by US President Donald Trump, who scrapped Biden's executive order on AI safety in his first week in office.

“My concern is that a possible AI race means that the guardrails on the AI models that are being developed in the US [will] fall away,” Professor Paterson said. In short, be prepared for AI development to move faster—and sometimes in the hands of unfamiliar players. “Like the early days of the internet, we should expect the early leaders to be overtaken by new entrants,” Professor Buchanan said. “This is a warning to all the current AI innovators who have been able to progress with very few new challengers.”

“I think it’s actually good news for the rest of us,” Professor Walsh added. “It's not just going to be the tech giants in the US that are going to dominate the AI race”.