Source: Deepnet Tencent News
Humans do not share the same joys and sorrows. Since 2016, the first year of artificial intelligence, the AI industry has gone through several rounds of reshuffles. With the help of ChatGPT, DeepSeek has stirred up the entire large model market like a catfish. Compared with it, other large model startups and the "six little dragons" regarded as upstarts in the industry are in a situation where the sun rises in the east and the rain falls in the west.
After DeepSeek shocked the industry with its low-cost DeepSeek-V3, which is comparable to GPT-4o in performance, it then released the R1 model on January 20. Six days after its launch, it topped the Apple App Store's global download list, and the cumulative downloads exceeded 110 million times within one month of its launch. During this period, major cloud vendors quickly launched the open source versions of V3 and R1, and products such as Baidu Search and WeChat are actively embracing DeepSeek.
Kimi's global reinforcement learning model k1.5 and step reasoning model Step R-mini, which were released at the same time as DeepSeek, are close to o1 in many aspects of model capabilities, but they are still drowned in the hot public opinion of DeepSeek.
Compared with the noise of DeepSeek, the "Six Little Dragons" also broke news one after another: Zero One Everything was further split, the budget and arbitration case of the Dark Side of the Moon were not settled, and another senior executive of MIniMax resigned...
And behind this are the frustrated VCs: none of the projects supported by real money has reached the popularity of DeepSeek. At present, four of the "Six Little Dragons" have not released any financing news for more than half a year. In 2024, the industry said that two of the "Six Little Dragons" have fallen behind. In 2025, who will be the next to fall behind?
Only three companies continue to take root in large models
DeepSeek's explosion is not without signs. Since the launch of its first model DeepSeek Coder on November 2, 2023, more than 10 different versions of the model have been launched in more than a year. Among them, the V2 model released in May last year is comparable to GPT-4 Turbo in performance, but the price is only 1% of GPT-4. Therefore, DeepSeek is called the "price butcher" and "AI Pinduoduo", and at the same time set off the first round of price wars in the large model industry.
On January 27, 2025, DeepSeek surpassed ChatGPT and topped the Apple APP Store free list in China and the United States, attracting global attention. What makes DeepSeek so successful is its large inference model DeepSeek-R1. According to information released by DeepSeek, R1 scored close to the official version of o1 in many authoritative tests, and even scored higher than the official version of o1 in some tests.
In addition to the rankings, open source + cost-effectiveness are the key combination that made DeepSeek so popular. Impacted by DeepSeek, Baidu founder Robin Li, who was once a believer in closed source, also announced that he would join the open source team. OpenAI founder Sam Altman also reflected that the company has always been on the "wrong side" in terms of open source strategy.
MiniMax, one of the "Six Little Dragons" of large models, released its first open source model on January 15. Its founder Yan Junjie also said in an interview with "LatePost" that "I don't have a lot of experience in starting a business for the first time. If I could choose again, I should open source it on the first day." Among the other five little dragons, only Zhipu was the first to walk on two legs of open source and closed source. After nearly two years of hard work, the development direction of the "Six Little Dragons" has gone in opposite directions.
Zero One Everything is the first basic big model company to publicly make major adjustments. It first laid off the pre-training algorithm team and the Infra team. Some personnel joined Alibaba in the form of job-hopping. Later, it announced the establishment of an industrial big model joint laboratory and an industrial big model base with Alibaba Cloud and Suzhou High-tech Zone respectively.
In terms of personnel, Huang Wenhao, head of model training, Lan Yuchuan, who is responsible for the big model API open platform, and Cao Dapeng, head of productivity products, have all resigned one after another. Zero One Everything, which tried to stay at the table, could not cover up its decline in this round of big model competition.
Baichuan Intelligence has made it clear that it will enter the medical track in 2024, and recently launched its first "AI pediatrician". Baichuan does not seem to be doing well in the commercialization of To B. Its co-founder and head of commercialization, Hong Tao, resigned years ago. According to an employee of Baichuan, it was indeed not as expected. "Now that we have DeepSeek, the pressure this year has only increased."
The head of To B commercialization also resigned from MiniMax, Wei Wei. Previously, Wei Wei said in an interview that many B-side customers would not easily pay this money to support the revenue of large model companies. They can only help customers align output effects in actual scenarios based on R&D capabilities and algorithm capabilities, which also proves that the commercialization of large models is not easy.
In this way, the only ones still focusing on large model technology innovation and pursuing AGI are the Dark Side of the Moon, Zhipu, and Step Star. Influenced by DeepSeek, Step Star has also joined the open source camp, but unlike DeepSeek, which focuses on text models, Step Star's latest open source is two multimodal models - Step-Video-T2V and Step-Audio.
In the early morning of February 23, the Dark Side of the Moon released its latest paper "Muon is Scalable for LLM Training" and open sourced the MoE model Moonlight, which only requires 3B model activation parameters. Many industry insiders believe that this is "intercepting the open source week" because DeepSeek announced earlier that it would release open source projects for 5 consecutive days.
For the dark side of the moon, the most pressing issue may be its Kimi product, which has invested heavily in traffic.
It is difficult to become the top brother of the list by spending money and investing traffic
Like the "Six Little Dragons" of large models, DeepSeek also has a C-end product with the same name, which did not attract much attention in the market in the first week after its launch. According to data disclosed by QuestMobile to the media, from January 13 to January 19, 2025, the weekly download volume of DeepSeek App was only 285,000, far less than Doubao (4.52 million) and Kimi (1.557 million).
After the release of R1 on January 20, 2025, the number of downloads of DeepSeek began to grow steeply. Sensor Tower research shows that DeepSeek was downloaded more than 16 million times within 18 days of the launch, almost twice the 9 million times when OpenAI's ChatGPT was first released.
The surge in visits once caused DeepSeek to crash, but even so, the growth momentum is still very strong, with monthly downloads exceeding 110 million. No one can ignore the brilliance of DeepSeek. At the internal staff meeting of ByteDance on February 13, CEO Liang Ruobo talked about DeepSeek and reflected on the insufficient follow-up speed. This year, we will pursue intelligent online.
Tencent's WeChat grayscale test connected to DeepSeek's AI search, and after the usage exceeded expectations, it called on the AI application Yuanbao to support WeChat search. On February 22, Tencent Yuanbao surpassed ByteDance's Doubao and rose to the second place in the Apple free APP download rankings in China, and DeepSeek continued to top the list.
The "big brother" of the first and second place changed hands in just one month, forcing Doubao and Kimi, who burned money for growth, to lose their advantages. The difference between the two is that the former is a noble born with a "golden key", while the latter is a "new entrepreneur". Previously, some media calculated that Kimi's daily investment in the iPhone channel alone was close to 200,000, while Doubao's was 2.48 million.
Under the influence of DeepSeek, Dark Side of the Moon was recently exposed to have drastically cut its product launch budget, including suspending the launch of multiple Android channels and cooperation with third-party advertising platforms. According to an insider who revealed to "AI Light Years", the promotion has indeed been adjusted accordingly, "there are natural additions, but they cannot be compared with the growth of DeepSeek."
Kimi's current troubles are more than these: "Undercurrent Waves" exclusively learned that the long-pending Kimi arbitration case did not complete the settlement as expected, but entered the next process of the arbitration case. According to insiders, the two parties in Kimi's arbitration case, the old shareholders of Circular Intelligence and Yang Zhilin, have paid the fees at HKIAC (Hong Kong International Arbitration Center) at the end of January and late February, respectively, and the court has been formed. Zhang Yutong, the more critical protagonist behind the whole incident, may be sued separately.
MiniMax also has high hopes for To C products because its star product Talkie became the fourth most downloaded AI application in the United States in the first half of 2024, which made it taste the sweetness. But the good times did not last long. In mid-December, Talkie quietly disappeared from the Apple App Store in the US market, while the Android platform was not affected.
Step Star, Zero One Everything, Zhipu AI and Baichuan Intelligence also have their own AI application products, but according to the AI product list, in January 2025, none of the top 20 AI applications with monthly activity were related to these four manufacturers. Previously, an employee of Baichuan Intelligence told AI Light Years, "It's not surprising that Baixiaoying's user retention and growth are very poor. We basically don't do advertising, and let other companies burn money to complete user education first."
Currently, DeepSeek, Tencent Yuanbao, and Byte Doubao occupy the top three of Apple's free APP download rankings. If the "Six Little Dragons" of large models want to be on the list, the competition will only be more intense. Nano Search, which is currently ranked seventh, Zhou Hongyi is personally "bringing goods".
Another opponent that cannot be ignored is Alibaba. After AI Application Tongyi was incorporated into Alibaba Intelligent Information Business Group, Alibaba's AI To C business recently launched a large-scale recruitment, with hundreds of positions, focusing on product and technology research and development positions related to AI large models. There are wolves in front and tigers behind, which is a true portrayal of the current situation of the "Six Little Dragons" of large models.
When the technology story is no longer romantic, commercialization is not as expected, and the monthly active user growth of the product is not proportional to the investment, the "Six Little Dragons" of large models are full of ideals, but the reality is skinny.
The threshold for the next round of financing will be raised
It is a recognized fact that large model pre-training costs money. Kai-Fu Lee once revealed that the cost of pre-training is about three to four million US dollars. Even the lower-cost Yi-Lightning used 2,000 GPUs for training, which took one and a half months and cost more than three million US dollars.
Even DeepSeek, which claims to be low-cost, has an inestimable investment in the early stage. The third-party organization SemiAnalysis estimates that DeepSeek actually has a huge computing power reserve: a total of 60,000 NVIDIA GPU cards, including 10,000 A100s, 10,000 H100s, 10,000 "special edition" H800s, and 30,000 "special edition" H20s.
"We estimate the training cost of a general large model to be around 1 billion US dollars, which is only the computing power part, and does not include the other two very expensive parts, one is data, and the other is labor cost. Talent in the field of large models is very scarce in the world now." Dr. Du Feng, founding partner of Jiangmen Venture Capital and former head of Microsoft Ventures Greater China, once told the author.
Due to the need for such a high investment, a saying has been popular in the industry for a long time: the entry ticket to invest in large model companies is 100 million US dollars. Another signal behind this sentence is that a large model startup company will find it difficult to survive if it cannot get financing.
After the 100 Model Wars in 2023, there will be financing news released almost every month, but as the AI bubble theory becomes more and more popular, starting from September 2024, there will be no hundreds of millions of hot money flowing to the "Six Little Dragons" of large models for a long time. Until before the Spring Festival in 2025, Zhipu and Jieyuexingchen announced that they had received "winter money", the former announced the completion of a new round of 3 billion yuan in financing, and the latter completed a round of B financing of hundreds of millions of dollars.
The other four of the "Six Little Dragons" have been away from the last release of financing dynamics for more than half a year: MiniMax officially announced the completion of a $600 million round of B financing in March last year, Baichuan Intelligence received a 5 billion yuan round of A financing in July last year, Zero One Everything completed a new round of hundreds of millions of dollars in financing in August last year, and the Dark Side of the Moon completed a $300 million financing in August last year.
During the Spring Festival, DeepSeek was popular all over the world, and public opinion did not hesitate to praise DeepSeek and its founder Liang Wenfeng. In the venture capital circle, there have been a lot of news circulating recently about whether DeepSeek will start financing and how much it is valued.
Previously, there was news that Alibaba would invest $1 billion at a valuation of $10 billion and hold a 10% stake. In response, Yan Qiao, vice president of Alibaba, quickly refuted the rumor through WeChat Moments, saying, "The information circulating outside that Alibaba invested in DeepSeek is false news." Later, foreign media reported that "DeepSeek considered raising external funds for the first time", and people related to DeepSeek denied the rumor, saying that the financing news was all rumors.
"Many investors are directly or through relationships with Liang Wenfeng. I predict that the valuation should be far higher than the current 'Big Model Six Little Dragons'." An investor from CICC Capital said, "DeepSeek has become a benchmark. The threshold for the six little dragons to get new financing in the primary market is obviously higher."
In fact, since the big model entrepreneurial boom has been set off, the industry generally does not believe that the "six little dragons" can finally survive as independent "big model companies". Several founders of the "six little dragons" have also expressed similar views in public. For example, Yan Junjie, the founder of MiniMax, believes that there will only be five big model companies left in the world in the future.
"China will definitely have its own ChatGPT. This is the same as search engines. We have our own compliance requirements. But the Chinese version of ChatGPT will only be produced in five companies: BAT+Byte+Huawei." Xunlei founder and Yuanwang Capital Cheng Hao once told the author.
Under the continued popularity, the already polarized “Six Little Dragons” will accelerate the reshuffle.
Weatherly
Anais