It was already going to be a huge week for AI news: OpenAI released its eagerly anticipated Operator AI agent for ChatGPT on Jan. 23, and just two days before that, U.S. President Donald Trump announced the Stargate initiative, a $500 million joint venture with OpenAI, Softbank and Oracle intended to ensure “the future of technology” in the U.S.
But it was the meteoric rise of a new player in the AI industry that turned the tech world on its head.
“It certainly was a splash. Lots of news events in terms of it being cheaper, less capacity, open source … and, probably most notoriously, it’s from China,” Jitterbit CEO Bill Conner said during a Fox LiveNOW segment on Jan. 27. Speaking with host Andy Mac about DeepSeek, Conner explained what the success of this startup could mean for the AI industry, for cybersecurity, and for the two countries now competing for AI supremacy.
The Rapid Rise of DeepSeek
While DeepSeek got off to a modest start following its Jan. 10 launch, it has reportedly added 3 million users since then — with as many as 80% of those within the last week alone. This roughly 2.4 million users easily outpaced the then-unprecedented launch of ChatGPT, which garnered 1 million users in five days, pushing DeepSeek to the top of App Store download rankings in 57 countries.
With a majority of AI development originating in the U.S., the emergence of a serious competitor coming from China wasn’t on many people’s radars — particularly since U.S. sanctions prevent the export of high-end GPUs and chips like the ones used to develop existing AI products. The stock market reacted to this Sputnik moment accordingly: U.S. chip giant Nvidia lost nearly 20% of its value after DeepSeek hit the headlines, resulting in the single-biggest one-day loss in stock market history. Google, Microsoft, ASML and Broadcom saw shares tumble as well.
While much of this value has since recovered, according to Conner the lesson is clear: “No country is going to control this technology. It’s in the wild. It knows no borders.”
How the Startup Got Its Start
Despite the deep pockets of company founder Liang Wenfeng, a billionaire hedge fund manager and Zhejiang University graduate, the DeepSeek origin story is a modest one. Liang reportedly gathered as many as 50,000 Nvidia A100 chips just prior to the export ban — then, after the ban, supplemented these with the best of what he could still get his hands on.
The resulting R1 model, powered by open-source DeepSeek-V3, was reportedly much cheaper to develop, requiring just $5 million to $6 million in training costs versus the billions spent by U.S. competitors.
Even so, by collaborating with other Chinese AI companies and refining the way its model learns, DeepSeek achieved AI performance comparable to OpenAI’s ChatGPT and Google’s Gemini, while using just 3-5% of the resources its rivals require — making it a cheaper and more sustainable option.
Conner and others are skeptical of these claims, however. “I don’t believe the 5 or 6 million they’re talking about. That might have been their investment,” Conner said.
He also expressed skepticism about DeepSeek’s creation.
“They say they used early vintage Nvidia chips that were cheaper in those pieces, but we also know those can’t perform more complex tasks at speed or cost,” he continued. “We just don’t know what that infrastructure is yet. We may never know.”
The Dangers of DeepSeek
Amid the widespread enthusiasm for DeepSeek — including from Trump, who called it a “wake-up call” and speculated it could pressure American companies to lower costs and increase efficiency — security experts have pointed out a number of potential dangers.
Some have speculated that the sudden availability of a low-cost, open-source AI model could jeopardize the profits of rivals who have already invested a considerable amount into AI infrastructure, thus stifling further innovation.
With DeepSeek already having been impacted by a cyberattack, security is another concern. Earlier this week, the company announced it was “temporarily limiting registrations to ensure continued service” due to a suspected DDoS attack. Since then, security firm Kela has identified several vulnerabilities in the company’s chatbot that make it susceptible to “jailbreaking” and allowed it to generate harmful content, such as instructions for creating explosives.
But what DeepSeek isn’t talking about is almost as concerning. “If you’re housing your data in China and your models are built in China, by nature, you’ve got some biases,” Conner said. “Some of the answers may not be what you think.”
For example, due to China’s strict laws governing information, DeepSeek’s AI models have been documented avoiding sensitive topics related to the Chinese government, such as the 1989 Tiananmen Square massacre, calling such things outside its “current scope.”
But by far the most talked-about concern is privacy.
“This is a Chinese algorithm. If you’re augmenting their LLM, your data for your customers or for yourself is stored explicitly in China,” Conner explained—referring to the portion of the privacy policy that states “We store the information we collect in secure servers located in the People’s Republic of China.”
“Chinese tech, from Huawei to TikTok, has repeatedly been subject of allegations that the firms are linked to the Chinese state, and fears that this could lead to people’s data being harvested for intelligence purposes.”
The privacy policy goes on to outline the data collected—which includes basic things such as email, phone number and birthdate, but also includes recorded text and audio, chat histories, IP addresses and even keystroke patterns. DeepSeek claims it uses this information to improve “safety, security and stability.” But it also states that it shares this information with third parties and that the information will be kept “for as long as necessary” and used “to help match you and your actions outside the service.”
While some have argued that U.S. privacy agreements at times aren’t much better, Conner pointed out that since the data is being sent to and stored in China, it could raise compliance concerns. “If you’re subject to GDPR or other regulations, you’ve got to be careful with it,” he said.
The Arrival of the AI Arms Race
DeepSeek’s emergence signals the beginning of what some are calling an “AI arms race” between China and the U.S., with different nations and companies vying for dominance in a rapidly evolving field.
“Certainly the U.S. would be considered the Wild West, and the innovator. But there are fast followers, and there are innovators around the world with deep technical know-how, and in this case, deep technical pockets,” Conner said. “In some cases, there are governments that are going to spend a lot more money than ours has in developing that innovation and capability.”
Ironically, as this arms race heats up and governments investigate both good and bad uses for AI, collaboration may be the key in forging a positive future.
“We’re now starting a new game” in AI, Conner said, and “all business people and technology people will have to come together to forge a public/private partnership in order to understand and make use of it.”