NEWS
NEWS

Inside the guts of the great Chinese startup laboratory where DeepSeek emerged: "We are recent graduates"

Updated

In Hangzhou, one of China's technological powerhouses, the 'chatbot' that has made a splash in the global technological scene was born, and like its domestic counterparts, censorship is its weak point

The chatbot was born in Hangzhou.
The chatbot was born in Hangzhou.Josetxu L. Piñeiro

In the office of DeepSeek, nestled among financial companies in a skyscraper in the Chinese city of Hangzhou, there is no one these days answering the thousands of calls and emails coming from journalists worldwide seeking some statement after the earthquake caused by the new artificial intelligence phenomenon in the markets this week.

No one picks up the phone because their employees, less than a hundred, are on vacation for the Chinese New Year. This is the first reason. But it is also true that even if it were not a holiday, there would be no one to answer questions. The company tends to keep such a low profile that it does not have a press department.

Not even an employee dedicated to public relations. Someone who is currently working in the Hangzhou skyscraper is the security guard, who has not quite understood the hustle and bustle of journalists and curious onlookers roaming around the building these days. The guard has never heard of DeepSeek, the startup that developed the AI assistant, which shares the same name, and has been the most popular in the world this week; the one that has disrupted the credentials that seemed unshakable for American giants as leaders in AI.

The guard, however, mentions that among the veteran finance profiles that dominate this office tower, he is struck by the fact that for a few years now, most of the employees on one of the floors are carefree twenty-somethings. "They are extremely young and full of vitality," he asserts.

The office address of DeepSeek is the same as that of High-Flyer Quant, the hedge fund that made Liang Wenfeng, a brilliant 40-year-old engineer, son of a rural primary school teacher from the southern Chinese province of Guangzhou, who was bullied as a child.

Other kids made fun of him for his haircut and because they said he was a geek. Liang, after amassing a fortune by successfully integrating AI learning models to predict market trends, developed the chatbot that has shattered industry norms. Chinese media reported that to develop DeepSeek, Liang sought out the best young talents who excelled in Chinese programming and electronic engineering schools, and also recruited national experts working abroad, especially in American technological institutions.

"Our team is mainly made up of recent graduates and workers with just one or two years of work experience," Liang stated in an interview last December after launching his AI models, DeepSeek-V3 and DeepSeek-R1, which were available in app stores on January 20.

Liang had claimed that the performance of his chatbot was on par, or even superior, to OpenAI's ChatGPT in complex tasks such as mathematics, coding, and natural language reasoning. No one believed him until Silicon Valley began testing the R1 with math problems that it solved much faster than ChatGPT.

The surprise was even greater in the technology hub when DeepSeek stated that the application had cost only 5.58 million dollars, hundreds of millions less than its American counterparts, and had been developed without relying on high-performance chips, which were believed to be essential for launching such advanced models.

Additionally, it implemented an open-source policy for its products, exposing the code and model architecture in a way that allows the public to view, use, and modify it. This means that many small and medium-sized companies can directly use their model and greatly reduce R&D costs.

How had a completely unknown company achieved this? How had the Chinese startup managed to overcome American technological restrictions and develop such a sophisticated low-cost model whose emergence had sent shockwaves through the actions of major Wall Street tech companies? We need to go back to the beginning to find the answers.

Hangzhou was one of China's seven ancient capitals and is one of the country's main technological hubs, with the giant Alibaba, headquartered in this city, as its main cash cow. Prosperity during the golden years of economic growth turned this megacity of 12 million inhabitants into very fertile ground for all kinds of startups. Those that succeeded the most at the beginning were those linked to e-commerce. Entire neighborhoods were flooded with 5,000 companies manufacturing streamers capable of selling all kinds of products to an audience of over 600 million people, that is, Chinese hooked on video streams on platforms.

In this favorable environment for entrepreneurs, leading companies in different sectors have emerged, such as Yushu Technology, which has created an army of robot dogs revolutionizing the robotics world; Qiannao Technology, a pioneer in brain-computer interface development; or Game Science, the video game developer behind the phenomenon Black Myth: Wukong, the title that dominated the market last year.

All these powerful companies are based in different technology parks scattered throughout Hangzhou. Therefore, one of the things that caught the attention of those looking for DeepSeek's address was that its office was located in a skyscraper dominated by financial companies.

Several Chinese TV programs have sent their cameras and reporters to do live broadcasts at the building's entrance after DeepSeek rocked the global stock markets last Monday, causing the value of Nvidia, the American chip giant, to plummet and triggering a tech market cataclysm due to the panic among major investors who realized that the Asian giant threatens the dominance of the current AI sector leaders. In China, they are delighted that their low-cost chatbot challenges ChatGPT. And, more importantly for the ruling Communist Party, the unexpected impact of DeepSeek has been a blow on the geopolitical chessboard, especially for US technological supremacy after years of Washington trying to suffocate Beijing with all kinds of restrictions to limit its access to the most advanced chips.

"A few years ago, when he was still a nobody and was setting up the hedge fund, Liang Wenfeng told his partners that his purpose was to change the rules of the game in the AI industry by building a great, very cheap model accessible to everyone. Of course, no one took him seriously then, very close sources to Liang told the Chinese newspaper Guagming Daily.Using AI and algorithms to identify patterns that could affect stock prices, Liang amassed a great fortune after founding High-Flyer Quant in 2015, which four years later already managed assets of over 10 billion yuan (about 1.3 billion euros). It was in 2023 when the company, which had been able to buy less advanced chips from Nvidia before the US began a harsh wave of restrictions, presented its first tests of a series of AI models, including DeepSeek."For Liang, DeepSeek is more of a secondary project or a hobby, driven by deep curiosity and a commitment to fundamental research.

His approach is to understand the essence of human intelligence and the processes that support it, and he believes that such exploration is crucial despite the lack of immediate commercial incentives," analyzed a portrait drawn by the Chinese state newspaper The Paper.

"We see that Chinese artificial intelligence cannot continue to have a lower profile than the rest. We often say that there is a gap of one or two years between China and the United States, but the real gap is between originality and imitation. If this does not change, China will always be an imitator," Liang affirmed in the only interview he granted - to a Chinese media outlet - after presenting his chatbot.

"The cost was so low compared to other models because we believe that AI services should be affordable and accessible to everyone."Even though the country is on vacation for the New Year, in China's great startup laboratory, everyone is talking about the latest phenomenon. Here, at home, just like in Silicon Valley, they also did not anticipate the success of Liang's AI model.

In the bubble of engineers and programmers in Hangzhou, it is highlighted that, in addition to the profile of novice and very young DeepSeek employees, the company pays its deep learning researchers and developers very well: up to 110,000 yuan per month, which is almost 15,000 euros at the exchange rate. A very high salary considering that the average cost of living in Hangzhou is much lower than that of any major Western city where the workers of AI titans live.

DeepSeek made its latest update on Monday, January 27, just hours before the West woke up to the shake-up in the markets and the collapse of the big tech companies, starting with the $600 billion drop in Nvidia, the ultra-profitable giant that recorded over $63 billion in profits in its last four quarters, and which Wall Street considered impregnable due to the extreme dependence that major AI companies had on its chips. That premise crumbled after the emergence of a chatbot that did not require Nvidia's high-end chips.
"The R1 from DeepSeek is an impressive model, particularly in terms of what it can offer for that price," wrote OpenAI CEO Sam Altman in X, after DeepSeek became the most downloaded free app in the AppleiPhone store in the US. Altman also noted that it was "exciting to have a new competitor" and that his company will be advancing some of its new product releases.
Throughout the week, major international media outlets have provided ample space for their tech analysts to spew their opinions on whether Monday's panic sales were an overreaction by the markets. Or if it was indeed a serious wake-up call because Silicon Valley had underestimated China's technological prowess and its ability to bypass numerous export restrictions. Undoubtedly, and what is most celebrated in Beijing, where they aspire to achieve technological self-sufficiency, was the show of strength in the technological war between the two global superpowers.

Just a couple of weeks ago, before leaving the White House, the Biden Administration once again pointed towards China with the announcement of a series of new controls on the export of chips and other crucial technologies for advanced AI projects. More quotas were imposed on the sales of advanced microcomponents, and US companies were also limited in sharing technical details of the most comprehensive AI models.


It remains to be seen what path Donald Trump will take regarding these restrictions, but among the executive orders he signed right after assuming the US presidency was one that overturned Joe Biden's requirement for companies to report their most significant advancements to the government.
Last year, a regulation came into effect prohibiting US private equity and venture capital companies, as well as joint ventures with Chinese conglomerates, from investing in Chinese artificial intelligence, quantum computing, and semiconductors.

Regarding the coveted chips, the US government has been implementing export controls under the official pretext of hindering China's ability to use US technology in military applications.


"DeepSeek's achievements could change China's national destiny amid its prolonged technological war with the United States," wrote Feng Ji on Weibo, a local social network similar to X, founder of the video game developer Game Science. "DeepSeek has revolutionized the world by launching two new powerful AI models, built at a lower cost and with fewer computing resources than those used by the largest tech companies," noted Zhou Hongyi, president of the Chinese cybersecurity company Qihoo 360.


DeepSeek's success is undeniable. However, it limps on the same leg as previous chatbots that have come out of China: censorship remains very noticeable when asked about politically sensitive topics for a regime that, despite selling openness outwardly, continues to be under the yoke of virtual repression at home. It's possible that many of the enthusiastic Chinese users currently using DeepSeek may discover that there are many aspects of their country's dark history that they are unaware of and that their government has hidden from them.