Exclusive: A Complete Record of the Visit to DeepSeek's Office After Its Global Sensation

A Chinese - developed open - source AI model, DeepSeek - V3, has made a huge splash in the global AI community with its shorter training time, lower cost, and equally excellent performance.

DeepSeek, the Chinese large - model company that developed this model, was founded in July 2023 with a registered capital of $ 1.37 million.

This extremely low - key startup was once out of the mainstream spotlight. However, it suddenly became a sensation after the launch and simultaneous open - sourcing of the DeepSeek - V3 model in December 2024. Public information shows that DeepSeek spent only about $5.58 million to train this model. Amid this sudden fame, DeepSeek chose to dive deep and had no desire to interact with the outside world in any form.

Nevertheless, DeepSeek's Hangzhou office has still become a popular spot.

Before the Chinese Spring Festival, after DeepSeek became a global hit, the popularity of its office on Huancheng North Road in Hangzhou skyrocketed. The property staff of the building said that even during the Spring Festival, there were forty or fifty groups of people coming every day.

A Field Visit to DeepSeek

The registered address of Hangzhou DeepSeek is Room 1201, West Building 1, Huijin International Building, No. 169 Huancheng North Road, Gongshu District, Hangzhou.

On February 5th, Yinshi Finance arrived at the address shown on High - Flyer's(DeepSeek's parent company) official website. The building complex consists of 4 buildings. High - Flyer is located in Building A, and DeepSeek is in Building B. Yinshi Finance noticed that the Huijin International Building has a strict security system. Each building has access control on the first floor. You need to swipe a card to enter, and there are security guards on duty at the access control. Even if you pass the access control, you also need to swipe a card to use the elevator to reach the corresponding floor.

The property management of Huijin International Building said that to go upstairs, you need to make an appointment with the corresponding company in advance, and the company staff will come to lead you upstairs. The property has no right to help you swipe the card to enter, nor can they make calls on - site.

The inability to enter the building to see DeepSeek up close did not dampen the enthusiasm of the visiting reporters. A property staff member said that after DeepSeek triggered a frenzy on January 27th, which led to a sharp drop in NVIDIA's stock price, the number of visitors here increased significantly, and the popularity even remained high during the Spring Festival holiday. There are forty or fifty groups of people coming here every day, and some are tourists visiting Hangzhou.

Posts about visiting Huijin International Building began to appear on social media one after another. From the content, it can be seen that the posters all came to visit after hearing that DeepSeek has its office here.

While waiting downstairs at Huijin International Building, Yinshi Finance met many people who came to visit DeepSeek, including tourists and media from all over the world.

A reporter from Japan's Yomiuri Shimbun told Yinshi Finance that he had been waiting here for 5 days to interview DeepSeek but had not yet received permission. The reporter also said that during his waiting period, he had encountered companies such as iFlytek that were trying to seek cooperation with DeepSeek, but these companies also failed to see the true face of DeepSeek.

Many citizens also came to visit DeepSeek. A resident living nearby said that he was very curious after hearing that a great company was born in Hangzhou, so he came to its office to have a look. A delivery guy said excitedly that DeepSeek was incredibly amazing and he believed it would transform the world and even humanity in the future.

Focusing on Research

Facing the enthusiastic visitors, DeepSeek's attitude was extremely calm, basically turning down all visitors.

A visitor information registration form provided by the property shows that on the morning of February 5th alone, there were 13 registration records from various parties such as the media and securities firms, all companies and individuals who came to visit DeepSeek. However, DeepSeek did not send anyone to receive these people.

The property staff said that DeepSeek had informed the property that it wanted to focus on research and politely declined all interviews. To cooperate, the property had increased some security forces. 

Yinshi Finance contacted DeepSeek as a media outlet. The other side said that they wanted to focus on research and did not accept interviews. Even when Yinshi Finance said that it only wanted to visit the company without conducting an interview, it was still rejected.

Yinshi Finance then contacted High - Flyer as an investor. The other side said that all its products had been closed and there was no plan to reopen them for now. They said that if you want to learn about the products, you can follow the company's official account.

Subsequently, Yinshi Finance managed to reach DeepSeek's office floor and found that this company, which single - handedly caused NVIDIA's stock price to plummet, was very low - key. The company's door was tightly closed, equipped with a face - recognition access control system.

Looking inside through the glass door, only the company name deepseek and red - colored New Year decorations could be seen. The internal layout of the company and its staff could not be seen, and the voices of employees communicating could not be heard either.

Someone who had entered DeepSeek's office revealed: DeepSeek's office is empty and spacious, extremely simple overall, and the original R & D team is all in Beijing.

What Subversions Has DeepSeek Brought to the AI Industry?

DeepSeek, with its lower cost, higher - performing models, lower requirements for chip performance, and open - source strategy, has brought a catfish effect to the AI industry. At the same time, it has also raised questions about the necessity of huge investments by technology companies in AI infrastructure. Recently, other large - model manufacturers have taken measures such as price cuts and free services, and urgently launched new models to deal with the impact of DeepSeek. A little earlier, several major US technology giants announced that they would increase their investments in AI infrastructure in 2025.

In contrast, DeepSeek has abandoned the blind pursuit of the computing power arms race and instead focused on improving the efficiency and practicality of its models. In terms of inference costs, OpenAI o1 charges $15 for every million input tokens and $60 for every million output tokens, while DeepSeek R1 only charges $0.55 and $2.19 for the same input and output respectively. This means that DeepSeek R1 achieves GPT - o1 - level performance at less than one - tenth of the cost. 

In terms of the market pattern, the rise of DeepSeek has broken the monopoly of US giants in the AI field, set an example for Chinese AI enterprises, and promoted the diversified development of the global AI industry. In terms of application scenarios, DeepSeek has applied AI technology to a wider range of scenarios, empowered the transformation and upgrading of traditional industries, and promoted the inclusive development of AI technology.

The future development of the artificial intelligence field is still full of possibilities. DeepSeek's technological breakthroughs have not only reduced costs and improved resource utilization efficiency but also further broadened AI application scenarios and lowered the thresholds for enterprise training and inference, allowing the commercialization of artificial intelligence to progress faster. From this perspective, the complexity of models and the total demand for computing power may further increase. 

Just as the Jevons Paradox expounds, after the efficiency of steam engines was greatly improved, the total consumption of coal did not decrease but increased instead. That is, while technological progress improves resource utilization efficiency, the growth in resource demand stimulated by cost reduction may exceed the resource savings brought about by efficiency improvements.