Sources say DeepSeek's new funding round aims to raise about 100 billion yuan, with Tencent and CATL leading

Article body

On October 22, market reports said that the domestic large-model company DeepSeek plans to raise about 100 billion yuan (about 15 billion US dollars) in its latest funding round, with Tencent and CATL as the largest investors in this round, and the deal is expected to be officially completed within the next few days.

As of now, DeepSeek, Tencent and other relevant investors have not publicly responded. The reporter cross-verified with multiple sources close to the company and the investors and confirmed that the above information is accurate. Sources said that the fundraising target for this round was initially set at 50 billion yuan, and was subsequently raised multiple times to about 100 billion yuan. It has not yet been officially announced, and the final amount is still not fully determined. The funds will continue to be used for model training, computing power infrastructure and talent recruitment to support DeepSeek in global competition. Some sources also believe that this round of financing will pave the way for the company's planned IPO on the STAR Market.

In June this year, DeepSeek completed its first round of financing of about 51 billion yuan. Among this, Liang Wenfeng personally contributed about 20 billion yuan, Tencent about 10 billion yuan, the CATL system about 5 billion yuan, and NetEase, JD.com, Lisi Capital, IDG Capital and the National Artificial Intelligence Industry Investment Fund (about 980 million yuan) also participated. Except for the National AI Fund, the funds from all parties in this round were injected into a limited partnership managed by Liang Wenfeng, and the investors do not have voting rights.

According to a previous report by Caijing, the pre-money valuation of this round remains at about 500 billion yuan. A number of industry insiders believe that this reflects that the company cares more about obtaining sufficient long-term funds than about raising its valuation. According to The Information, Liang Wenfeng once revealed at an investor meeting that the company's annualized revenue run rate has exceeded 1 billion US dollars, compared with less than 500 million US dollars a few months ago; customers did not leave after the API price increase, and revenue mainly comes from model API calling services.

Computing power is the primary destination of this 100 billion yuan financing. According to a previous report by this newspaper, DeepSeek is building a large data center in Ulanqab, Inner Mongolia. Hu Yanping, a distinguished professor at Shanghai University of Finance and Economics, believes that the parameters of DeepSeek's models in training are said to have exceeded 2 trillion, and next year most domestic flagship models will rise to the range of 5 trillion to 8 trillion, or even approach about 10 trillion, so computing power investment will need to grow several times. This round of financing is not the end; continuing to raise funds through listing is an inevitable path to take next.

In Hu Yanping's view, Tencent's heavy investment in DeepSeek for two consecutive rounds is an important part of its AI layout. Developing in-house while taking stakes in leading model companies is becoming a new approach for big tech companies to participate in large-model competition. He analyzed that Tencent does this for both financial and strategic considerations. Tencent has always insisted on in-house development. Although Hunyuan has narrowed the gap through catching up, its competitiveness is still not strong enough; if Tencent wants to make up for its AI shortcomings, the success of WorkBuddy alone is not enough. DeepSeek's foundation models and open-source full-stack R&D capabilities lead in China. Investing in a range of domestic model companies including DeepSeek can both make up for the shortcomings of in-house development and cope with future competition, and is also equivalent to locking in the dividends of this wave of the AI industry in advance.

Regarding the future competitive landscape of the industry, Hu Yanping believes that the gap between Chinese and US models should be viewed from two aspects: in terms of conventional intelligence, DeepSeek's Flash model can already catch up with the performance of overseas flagships in low-computing-power mode; while in terms of deep intelligence, the gap has been widening since the second and third quarters of this year. Whether a "flywheel effect" can be formed through financing and listing will determine whether domestic models can "emerge as soon as possible" in the next round of competition. In his view, the key to widening the gap among domestic models lies in four points: funding, computing power, deep intelligence, and market progress achieved in the "model-agent integration" layout.

Industry insiders believe that with the landing of DeepSeek's new round of financing, the "computing power arms race" among domestic large models will further accelerate. Whether these funds can be efficiently converted into training capability and model iteration speed will determine whether DeepSeek can narrow the gap with overseas flagship models on the deep intelligence track. For the entire industry, leading companies obtaining sufficient long-term capital also means that the competitive threshold is rising, and the concentration of capital, computing power and talent may accelerate the differentiation and reshuffling of domestic large models.

(Source: Shanghai Securities News) Middle of the body Embedded advertisement