摘要：AsianFin -- Chinese tech start-up DeepSeek is causing a stir in the U.S. AI sector with its open-source approach and affordable mo

(Image Source: Shutterstock US)

AsianFin -- Chinese tech start-up DeepSeek is causing a stir in the U.S. AI sector with its open-source approach and affordable models, gaining attention from both U.S. media and AI forums. Despite facing stringent semiconductor restrictions imposed by the U.S. government, DeepSeek has managed to develop AI systems that compete with those of leading U.S. tech giants, sparking curiosity about the company's innovative capabilities.

As of press time, DeepSeek has climbed to the sixth spot in the top charts of iOS download apps. ChatGPT ranks the top.

"China's cheap, open AI model DeepSeek thrills scientists" is the title of an article published on the website of Nature on January 23 after a large language model called DeepSeek-R1 was released by the Chinese company on January 20. Its performance on some tasks in chemistry, mathematics and coding is on par with that of OpenAI's o1, the Nature report said.

"How China's new AI model DeepSeek is threatening U.S. dominance," a 40-minute video made by CNBC on Friday expressed concerns over DeepSeek's impact on U.S. dominance in AI. It said that "a little-known AI lab out of China has ignited panic throughout Silicon Valley after releasing AI models that can outperform America's best despite being built more cheaply and with less-powerful chips."

Unlike the years of work and billions of dollars invested by American tech giants, DeepSeek has reportedly achieved its breakthrough in just two months, spending only $5.6 million, CNBC reported. This represents a stark contrast to OpenAI, which spends over $5 billion annually on research, or Google, which had estimated $50 billion in capital expenditures for 2024.

A New York Times report said "The company built a cheaper, competitive chatbot with fewer high-end computer chips than U.S. behemoths like Google and OpenAI, showing the limits of chip export control," in a report titled "How Chinese AI Start-Up DeepSeek Is Competing With Silicon Valley Giants."

Attention on the Hangzhou-based startup began in December 2024 following the launch of its large-scale AI model, "DeepSeek-V3." The new model swiftly gained popularity among AI enthusiasts, igniting conversations that crossed borders and spread across international social media platforms and tech forums.

Founded by Liang Wenfeng, DeepSeek is reportedly linked to Chinese hedge fund High Flyer Quant, which manages around $8 billion in assets.

DeepSeek's journey to becoming a significant player in the AI space has been rapid. Starting in 2021 with the purchase of thousands of GPUs to build AI computing power, DeepSeek has made remarkable progress in just four years. Liang, often described as a "small-town prodigy," was born in Zhanjiang, a third-tier city in Guangdong Province. He excelled academically, earning a master's degree in Information and Communication Engineering from Zhejiang University in 2010. After graduation, Liang led a team to explore fully automated quantitative trading using machine learning. In 2015, he co-founded High-Flyer Quant, aiming to create a world-class quantitative hedge fund using mathematics and AI.

By 2021, High-Flyer Quant managed over 100 billion yuan in assets. That same year, Liang began investing in AI technology, purchasing thousands of Nvidia GPUs (likely including models like the RTX 4090, A100, and L40). However, by 2023, the fund's total assets under management had decreased to around 40 billion yuan. Despite this, DeepSeek's AI capabilities continued to grow.

DeepSeek's rise has been accompanied by significant media attention, particularly in China, where it is often portrayed as evidence that Chinese AI companies are surpassing their U.S. counterparts. However, this narrative is somewhat exaggerated. DeepSeek's early models were relatively crude, relying heavily on open-source data and even utilizing GPT-3.5 APIs. While DeepSeek has made significant strides, it is not yet on par with leading U.S. companies like OpenAI. The media's focus on DeepSeek as a symbol of China's AI superiority is, in many ways, a case of "cherry-picking" and does not reflect the broader landscape of AI development.

DeepSeek has chosen not to seek external funding or pursue an IPO in the short term. Instead, it relies on strong cash flow to recruit top AI research talent, creating a research-focused environment. The company's team is highly skilled in infrastructure and chip design, and Liang has brought some of the best talent from the hedge fund industry to DeepSeek. This strategy has allowed DeepSeek to focus on cutting-edge research without immediate commercial pressures.

According to a notice released by DeepSeek in an open communication group, the company said that it "does not engage in external project cooperation, nor does it provide privatization deployment and related support services." Additionally, DeepSeek emphasized that it will concentrate on research and development to develop more advanced models, encouraging everyone to "please look forward to it."

Nvidia Senior Research Manager Jim Fan said on X that "We are living in a timeline where a non-U.S. company is keeping the original mission of OpenAI alive - truly open, frontier research that empowers all."

Meta's AI department is feeling the pressure, according to an anonymous post on Teamblind, a forum for verified Big Tech employees. The post wrote that Meta is "in panic mode. Engineers are moving frantically to dissect DeepSeek and copy anything and everything we can from it."

Meta's Mark Zuckerberg took to Facebook, saying that in 2025, Meta aims to develop an AI assistant that can serve more than a billion people. "This will be a defining year for AI," Zuckerberg wrote.

The New York Time left the question of China-US competition in AI open-ended. It wrote that "If the best open source technologies come from China, they argue, U.S. developers will build their systems atop those technologies. In the long-run, China could emerge as a central player in AI innovation globally."

DeepSeek’s success is not an isolated case in China’s rapidly advancing AI landscape. Other Chinese companies, such as Kai Fu Lee’s 01.AI, are also making significant strides, which has earned the unicorn start-up status just eight months after its establishment. Similarly, Alibaba’s Qwen AI model has dramatically reduced the cost of training large language models, making it more attractive for developers.

The speed with which Deepseek has caught up to the global leaders has prompted a reevaluation of the AI arms race. Former Google CEO Eric Schmidt, who previously claimed that China was 2-3 years behind the U.S. in AI, has now acknowledged that China has caught up in a matter of months, noting the impressive progress made by models like Deepseek.

The availability of powerful open-source models is also reshaping the landscape. Deepseek’s model is offered free of charge and allows developers to customize and fine-tune it for their needs. This openness, combined with the low cost—just $0.10 per million tokens compared to $4.40 per million tokens for OpenAI—has made it an attractive alternative to more expensive proprietary models. As a result, the widespread adoption of open-source models could shift the dynamics of the global AI market.

U.S. President Donald Trump recently announced a $500 billion AI infrastructure investment in the U.S. Three top tech firms will create a new company, called Stargate, to grow artificial intelligence infrastructure in the U.S., according to CNN.

During the World Economic Forum in Davos recently, Alexandr Wang, the founder of Scale AI, made comments about DeepSeek, saying said that DeepSeek's large AI models are roughly on par with the best models developed in the United States. Wang acknowledged that the U.S. has likely led the AI race over the past decade, but he suggested that DeepSeek's recent advancements could "change everything."

Wang also said he believes the AI sector will reach a trillion dollars, on par with estimates that the generative AI market is poised to top $1 trillion in revenue within a decade.

“The United States is going to need a huge amount of computational capacity, a huge amount of infrastructure,” Wang said, later adding, “We need to unleash U.S. energy to enable this AI boom.”

Wang also hinted at potential challenges for DeepSeek due to U.S. export controls. He noted, "DeepSeek has about 50,000 H100 computing cards, which they can't talk about obviously because it is against the export controls that the United States has put in place, and I think it is true that you know I think they have more chips than other people expect but also on a go forward basis they are going to be limited by the chip controls and the export controls that we have in place."

Those who see DeepSeek's performance and think 'China is surpassing the U.S. in AI' are mistaken, said Yann LeCun, Chief AI Scientist at Meta and Turing Award winner.

The correct interpretation is that 'open-source models are surpassing proprietary models.' DeepSeek's success demonstrates that open-source AI models are competitive, but it also raises questions about whether the company will face targeted restrictions from the U.S. government, which could hinder its access to critical computing resources, he added.

来源：钛媒体APP