The day after Christmas, a small Chinese startup called DeepSec unveiled a new AI system that matches the capabilities of advanced chatbots from companies like OpenAI and Google.
This alone would have been a milestone. But the team behind the system, called DeepSeek-V3, described an even bigger step. In one Research paper Explaining how they built the technology, DeepSec engineers said they used only a fraction of the highly specialized computer chips that leading AI companies rely on to train their systems.
These chips are at the center of a fierce technological competition between the US and China. As the U.S. government works to maintain the country’s lead in the global AI race, it is trying to limit the number of powerful chips from the likes of Silicon Valley firm Nvidia that can be sold to China and other rivals. can
But the performance of the DeepSec model raises questions about the unintended consequences of the US government’s trade restrictions. The controls have forced researchers in China to get creative with a wide range of tools freely available on the Internet.
The DeepSec chatbot answered questions, solved logic problems and wrote its own computer programs as competently as anything already on the market, using US AI companies.
And it was built on the cheap, challenging the prevailing notion that only the tech industry’s biggest companies — all of which are based in the US — can afford to build sophisticated AI systems. Chinese engineers said they needed just $6 million in raw computing power to build their new system. That’s 10 times less than what tech giant Meta spent building its state-of-the-art AI technology.
“The number of companies that have $6 million to spend is far greater than the number of companies that have $100 million or $1 billion to spend,” said Chris V. Nicholson, an investor at venture capital firm Page One Ventures. ” AI technologies.
Since OpenAI sparked the AI ​​boom with the launch of ChatGPT in 2022, many experts and investors concluded that no company could compete with the market leaders without spending hundreds of millions of dollars on specialized chips.
The world’s leading AI companies train their chatbots using supercomputers that use as many as 16,000 chips, if not more. Deepsec engineers, on the other hand, said they only needed 2,000 specialized computer chips from Nvidia.
Geoffrey Ding, an assistant professor at George Washington University, said the chip bottlenecks in China forced DeepSec engineers to “train it more efficiently so that it’s still competitive.”
Earlier this month, the Biden administration issued new rules aimed at preventing China from acquiring advanced AI chips through other countries. The rules build on earlier rounds of restrictions that prevented Chinese companies from being able to buy or manufacture advanced computer chips. President Trump has not yet indicated whether he will keep the rules in place or repeal them.
The US government has tried to keep advanced chips out of the hands of Chinese companies over concerns that they could be used for military purposes. In response, some firms in China have stockpiled thousands of chips, while others source them from a thriving underground market of smugglers.
DeepSec is operated by a quantitative stock trading firm called HighFlyer. By 2021, he had invested his profits in acquiring thousands of Nvidia chips, which he used to train his first models. The company, which did not respond to requests for comment, has become known for promising high salaries from top universities in China and its ability to pursue research questions that most interest them.
Zihan Wang, a computer engineer who previously worked on the DeepSec model, said the company hires people with no computer science background to help them understand the technology, and that they are notoriously difficult at Chinese colleges. Can create poetry and twenty one questions in entrance exam.
DeepSec does not build any products for consumers, leaving its engineers to focus entirely on research. That means its technology doesn’t comply with the strictest aspects of China’s regulations on AI, which require consumer-facing technology to comply with government controls on information.
Leading U.S. companies are pushing the state of the art in AI In December, OpenAI unveiled a new “reasoning” system called o3 that outperforms existing technologies, though it’s still widely available outside the company. is not But Deepak continued to show he wasn’t far behind. This month, he released an impressive reasoning model of his own.
(The New York Times has filed a lawsuit against OpenAI and its partner Microsoft, alleging copyright infringement of news content about AI systems. OpenAI and Microsoft have denied the claims. )
An important part of this rapidly changing global market is an old idea: open source software. Like many other companies, DeepSeek has open-sourced its state-of-the-art AI system, meaning it has shared the underlying code with other businesses and researchers. It allows others to create and distribute their own products using the same technologies.
While employees at large Chinese technology companies are limited to collaborating with their peers, “if you work on open source, you work with talent from all over the world,” says Yanning, lead software engineer at Beeston in San Francisco. said Zhang, who works on the open source SGLang. The project helps other people and companies build products using DeepSeek’s systems.
The open source ecosystem for AI gathered steam in 2023 when Meta freely shared an AI system called LLama. Many believed that this community would only flourish if companies like Meta — tech giants with massive data centers filled with specialized chips — continued to open-source their technologies. But DeepSec and others have shown that they, too, can leverage the strengths of open-source technologies.
Many executives and pundits have argued that large American companies should not open source their technologies because they could be used to spread disinformation or cause other serious harm. Some US lawmakers have explored the possibility of blocking or throttling the process.
But others say that if regulators stifle the development of open-source technology in the United States, China will gain a significant lead. If the best open-source technologies come from China, he says, American developers will build their own systems on top of those technologies. In the long run, this could place China at the center of AI research and development.
“The center of gravity of the open source community is shifting to China,” said Einstein Stoica, a computer science professor at the University of California, Berkeley. “This could be a huge threat to the United States,” because it allows China to accelerate the development of new technologies.
Hours after his inauguration, President Trump rescinded an executive order from the Biden administration that threatened to curb open-source technologies.
Dr. Stoica and his students recently built an AI system called Sky-T1 that rivals the performance of OpenAI’s latest system, called OpenAI o1, on certain benchmark tests. All they needed was $450 in computing power.
They did this on top of two open-source technologies released by Chinese tech giant Alibaba.
Their $450 system isn’t as powerful as OpenAI’s technology or DeepSeek’s new system. And the techniques they used are unlikely to yield systems that outperform known technologies. But the project showed that even operations with modest resources can create competitive systems.
Raven Cohen, a technology consultant in Toronto, has been using DeepSeek-V3 since late December. They say it’s comparable to state-of-the-art systems from OpenAI, Google and San Francisco startup Anthropic — and much cheaper to use.
“DeepSac is a way for me to save money,” he said. “This is the kind of technology that someone like me wants to use.”
