"Finance" learned that recently, the first round of financing negotiations for DeepSeek, a large AI (artificial intelligence) model company, is nearing completion.Investors involved in the transaction negotiations said that this round of financing was led by the National Integrated Circuit Industry Investment Fund (referred to as the "National Large Fund"), and several market-oriented investment institutions are also on the negotiation list.


DeepSeek's model iteration requires continuous investment in large-scale training computing power, and sufficient funds are an important source to support this long-term investment.

Since April this year, the sentiment in the primary market has been high, and DeepSeek has become a hot investment target. A person close to the transaction said, "Everyone wants to get in touch with (DeepSeek), but no one guarantees 100% investment." Until the last moment, there are variables in both the participating investors and the amount of investment. However, the person recently revealed to Caijing that the list of the first round of financing has been basically finalized by the end of this month.

This is the first time that DeepSeek, which was founded three years ago, has opened financing to the outside world. A month ago, DeepSeek's valuation was still US$20 billion. According to media reports on May 22, DeepSeek's latest round of financing has increased to 70 billion yuan, approximately US$10 billion, and its pre-financing valuation has reached US$45 billion. If this round of financing is successfully completed, DeepSeek will become the company with the largest first round of financing in the history of China's large AI models.

Who wants to get involved?

An investor familiar with DeepSeek's senior management initially learned that the financing was dominated by state-owned assets and did not consider taking action. "The market is already very large. If market-oriented investment institutions enter, the return on investment may not be very high." In the financial circle, "white horse stocks" usually refer to those "high-achieving students" with large market capitalization and relatively stable trends. He judged that DeepSeek may be a "big white horse."

The turning point occurred after May Day this year. He sensed that DeepSeek was opening up opportunities to market capital. Many well-known investment institutions, including Monolith Capital and Hillhouse Capital, are intensively exposed to DeepSeek. He quickly changed his tune, and his agency later joined the talks.

In the venture capital circle, DeepSeek's share is hard to come by. It was once rumored that Liang Wenfeng opened an email and asked various institutions to send emails to introduce themselves and explain their understanding of AI. This rumor has been disproven. The above-mentioned investors who are familiar with DeepSeek’s senior management said that funds that do not know Liang Wenfeng and lack reputation will basically not be able to invest.

"Finance" learned that in this round of financing, some investors tried to package assets into tradable securities through SPV (Special Purpose Vehicle, special purpose entity) and bring in new LPs (limited partners). A person close to the trade told Caijing that some less well-known LPs would not be able to invest in DeepSeek under normal circumstances, but they can enter the game through such channels.

Penetrating investment through SPV requires additional management fees and channel fees. Therefore, leading funds rarely enter the financing plate through this method. In recent years, the global AI investment boom has been increasing, and SPV penetration structures have been widely used in primary market transactions.

A trader told Caijing that such operations would only occur in particularly popular transactions. "Even if you pay a layer of channel fees, based on today's valuation of (DeepSeek), you can still make money after it is listed."

However, there are risks associated with trading through an SPV due to the potential risk of investment fraud. Last week, American AI unicorn Anthropic upgraded its ban on unauthorized share transactions, explicitly prohibiting investment in Anthropic’s past or future financing rounds through SPVs, saying that transactions conducted through unauthorized channels are considered invalid. This move caused shocks in the secondary private equity market, and the prices of some listed funds holding shares through SPVs fell sharply.

Different from the overall optimism of the market, some investors who have never been willing to contact believe that DeepSeek's current valuation is too high and "cannot understand". It takes a lot of time and money to participate in negotiations, and the possibility of final investment is too low. Considering the return on investment, this transaction is not cost-effective.

Chen Shi, an investment partner at Fengrui Capital, told Caijing that it is good to invest in DeepSeek, but it does not make much sense to get a small share through a "particularly indirect" and unguaranteed relationship. "You should not lose money if you invest now, but I personally think it is better to invest in opportunities that I have discovered and projects that can obtain a larger share, which are more valuable."

There are also several major manufacturers on the negotiation list previously disclosed by the media. According to media reports, new energy giant CATL is planning to enter the game, strategically pointing to the energy variables behind the computing infrastructure. The media also previously reported that Tencent and Alibaba are also negotiating with DeepSeek. Alibaba has publicly denied the news and expressed little interest. As for whether Tencent is still in contact, Caijing asked Tencent for confirmation but received no positive response.

Several AI investors have analyzed that among the two major companies, Tencent and Alibaba, Tencent is the one with more motivation to enter the game. Its hybrid model has not yet caught up with the first echelon and has reached a strategic cooperation with DeepSeek. Alibaba has established its own ecosystem from models to chips, and its strategic appeal for investing in DeepSeek is not that strong.

DeepSeek's Crossroads

From 2025 to now, several large AI model companies that have survived the "Battle of 100 Models" are accelerating their financing and commercialization processes. In January this year, Zhipu Technology and Minimax, both of the AI ​​"Six Little Dragons", were listed on the Hong Kong stock market. Zhipu's latest market value was HK$6,353.26 (approximately US$810.93), and MiniMax's latest market value was HK$266.59 billion (approximately US$340.28).

Dark Side of the Moon recently completed US$2 billion in financing, with a valuation of over US$20 billion. It is dismantling its VIE and red-chip structures in preparation for listing. Among the leading large-scale AI model startups, DeepSeek was the latest to start financing, but its current first-round financing scale of US$10 billion far exceeds its competitors.

Regarding DeepSeek's rising valuation, many investors believe that it is still within a reasonable range and there is still room for upside. From the perspective of model inference efficiency, DeepSeek is recognized in the industry as being at the forefront of the industry. After the release of new models, it still maintains low cost and high performance. Taking V4-Pro released in April as an example, officials revealed that in a long context scenario of 1 million tokens, the FLOPs of single-token reasoning are only 27% of the previous generation V3.2. This efficiency advantage is directly reflected in the pricing of the API: V4-Pro is 0.025 yuan, which is the lowest level among mainstream models in the world.

How to commercialize it is an important question hanging on DeepSeek's high valuation. According to Caijing's previous report, in March this year, DeepSeek executives revealed at a meeting that the outside world is paying attention to the implementation of DeepSeek's business model and technical progress. DeepSeek has been making a lot of efforts and attempts in this regard, and has also found some paths through preliminary verification.

Several investors in the AI ​​field told Caijing that DeepSeek chose to open up financing for two main reasons: first, to allow employee options to be market-verified to retain talents; second, the funds required for model training will only get higher and higher, and its own assets originally created by magic square quantification may not be able to meet research and development needs in the future.

In addition, a trade source close to DeepSeek said that DeepSeek employees also want the company to go public.

Regarding whether DeepSeek should build a product, an investor close to Liang Wenfeng told Caijing that Liang Wenfeng probably realized the importance of productization at the end of last year. "He wants to build a product." The person said that DeepSeek's product team is still small and has not recruited the right person in charge. The company has struggled on the product route. "Liang Wenfeng has to think about the product himself."

According to media reports on May 22, Liang Wenfeng stated at a meeting for investors that the company will continue to promote open source AI models and aim to achieve general artificial intelligence (AGI). In his statement, technological breakthroughs are the core mission of DeepSeek, which is to break through the boundaries of technology rather than pursue profits.

The above-mentioned investors close to Liang Wenfeng believe that there is no conflict between realizing AGI’s technical ideals and commercialization. A more realistic consideration is that DeepSeek's model iteration requires continuous investment in large-scale training computing power, and sufficient funds are an important source to support this long-term investment.

Chen Shi believes that the business model of leading large model companies such as Anthropic and OpenAI is already at the top of the industry and is also the "most stressful business model" because the Scaling Law (law of scale) is still in effect. The training cost of each generation of models is higher than that of the previous generation, requiring continuous investment in computing power, data and talents. Once they fall out of the poker table, they will be forgotten by the market. As a start-up company with large AI models, today's primary market is no longer able to meet the funds required for model iterations. National funds can support one step, and the next step is to enter the secondary market.

DeepSeek has not made efforts on the product side before, and its focus has always been on model training; but today's market environment has changed, and both Coding and Agent applications require a large amount of high-quality token consumption. "If DeepSeek is determined to invest in commercialization, its revenue should increase rapidly and will not be less than other current competitors." Chen Shi said.

The national team led the investment, which is the most eye-catching point of this round of financing. Song Xiangqing, vice president of the China Society of Business Economics, told Caijing that a major national fund led the first round of financing in DeepSeek. This is the first cross-border investment in a pure large model company by a major national fund in the 12 years since its establishment. The AI ​​industry is entering a stage of "leading the national team and following up with market capital." He believes that DeepSeek is rising to become a national AI strategic platform.

AI and computing power have been upgraded to national strategies, which are also reshaping the capital structure of leading model companies. DeepSeek’s deep integration with the domestic chip ecosystem is one of the backgrounds for the entry of the national team in this round. DeepSeek-V4, released in late April this year, is the first to support Huawei's new generation AI chip Ascend 950PR, which means that model training will not completely rely on the NVIDIA CUDA chip ecosystem.

Changes in capital structure do not necessarily mean changes in route. "Liang Wenfeng is a relatively pure person." A person close to Liang Wenfeng told Caijing that even if the national team leads the investment in the secondary market, DeepSeek's daily operations and technical routes will not necessarily be bound by the capital market. Listing is just a capital operation, not an inflection point to change direction.

However, when DeepSeek stepped out of the "self-sufficient" technological utopia, facing the capital market, it could not avoid the problem that all technology startups will face: how to coexist between technical ideals and commercial returns in the long term.