DeepSeek's continuous rise has accelerated the reshuffling of the already diverging "Six Little Dragons."
Author: Wu Qianyu
The joys and sorrows of humanity are not interconnected. Since the dawn of the artificial intelligence era in 2016, the AI industry has undergone several rounds of reshuffling. Riding the wave of ChatGPT, DeepSeek has stirred the entire large model market like a catfish, while the other large model startups, regarded as the new elite "Six Little Dragons," find their situations starkly different, akin to the sun rising in the east while it rains in the west.
After DeepSeek shocked the industry with the launch of its low-cost DeepSeek-V3, which rivals GPT-4o in performance, it subsequently released the R1 model on January 20. Just six days after its launch, it topped the global download charts on the Apple App Store, accumulating over 110 million downloads within a month. During this period, major cloud vendors quickly launched open-source versions of V3 and R1, with products like Baidu Search and WeChat actively embracing DeepSeek.
In contrast, Kimi's global reinforcement learning model k1.5 and the step reasoning model Step R-mini, released around the same time as DeepSeek, are close to o1 in various aspects of model capability but remain overshadowed by the overwhelming buzz surrounding DeepSeek.
Compared to DeepSeek's noise, the "Six Little Dragons" have also been hit with a series of news: Zero One Everything further splits, the budget and arbitration case of the Dark Side of the Moon remain unresolved, and another executive from MiniMax has left…
Behind this, there are disappointed VCs: none of the projects they financially supported have reached the popularity of DeepSeek. Currently, four of the "Six Little Dragons" have not released any financing news for over six months. The industry predicts that two of the "Six Little Dragons" will fall behind by 2024; who will be next in 2025?
Only Three Continue to Root for Large Models
DeepSeek's explosive growth was not without warning. Since launching its first model, DeepSeek Coder, on November 2, 2023, it has released over ten different versions of models in just over a year. Among them, the V2 model released in May last year rivals GPT-4 Turbo in performance but is priced at only 1% of GPT-4, earning DeepSeek the titles of "price butcher" and "the Pinduoduo of AI," while igniting the first round of price wars in the large model industry.
On January 27, 2025, DeepSeek surpassed ChatGPT to top the free app charts in both China and the United States on the Apple App Store, drawing global attention. The driving force behind DeepSeek's success is its reasoning large model, DeepSeek-R1. According to information released by DeepSeek, R1 scored close to the official version of o1 in several authoritative tests, with some tests even exceeding the official version of o1.
Beyond the rankings, the combination of open-source and cost-effectiveness has been a significant factor in DeepSeek's immense popularity. In response to DeepSeek's impact, Baidu's founder, Robin Li, a former proponent of closed-source, announced his entry into the open-source camp, while OpenAI's founder, Sam Altman, reflected that the company has been on the "wrong side" of the open-source strategy.
Among the "Six Little Dragons," MiniMax released its first open-source model on January 15. Its founder, Yan Junjie, stated in an interview with "Late Point" that "the first time entrepreneurs lack many experiences, if I could choose again, I should have gone open-source from day one." Among the other five dragons, only Zhipu has been the earliest to adopt both open-source and closed-source approaches. After nearly two years of trial and error, the development directions of the "Six Little Dragons" have diverged significantly.
Zero One Everything was the first foundational large model company to publicly make significant adjustments, first laying off its pre-training algorithm team and Infra team, with some personnel moving to Alibaba, and later announcing joint ventures with Alibaba Cloud and Suzhou High-tech Zone to establish an industrial large model joint laboratory and industrial large model base.
In terms of personnel, model training head Huang Wenhao, head of the large model API open platform Lan Yuchuan, and productivity product head Cao Dapeng have all left. Zero One Everything, trying to stay in the game, cannot hide its decline in this round of large model competition.
Bai Chuan Intelligent has clearly defined its path in the medical sector, recently launching its first "AI Pediatrician." However, its commercialization in the To B sector seems to be struggling, as its co-founder and commercialization head, Hong Tao, left before the new year. According to an employee at Bai Chuan, the performance has indeed fallen short of expectations, stating, "With DeepSeek now, the pressure this year has only increased."
The departure of the To B commercialization head also affected MiniMax, with Wei Wei previously stating in an interview that many B-end clients are reluctant to spend money to support the revenue of large model companies, and can only help clients align output effects in practical scenarios based on R&D and algorithm capabilities, confirming that large model commercialization is not an easy task.
Thus, it appears that only the Dark Side of the Moon, Zhipu, and Step Star are still focused on large model technological innovation and the pursuit of AGI. Influenced by DeepSeek, Step Star has also joined the open-source camp, but unlike DeepSeek's focus on text models, Step Star's latest open-source offerings are two multimodal models—Step-Video-T2V and Step-Audio.
In the early hours 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 requires only 3B model activation parameters. Many industry insiders believe this is a "preemptive strike during the open-source week," as DeepSeek had previously announced it would release open-source projects for five consecutive days.
For the Dark Side of the Moon, the pressing concern may be its substantial investment in the Kimi product.
Burning Money for Traffic Fails to Become the Top Brother
Like the "Six Little Dragons" of large models, DeepSeek also has a similarly named C-end product, which did not attract much attention in the market during its first week after launch. According to data disclosed by QuestMobile, from January 13 to January 19, 2025, the weekly download volume of the 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, DeepSeek's download volume began to surge sharply. Research from Sensor Tower showed that DeepSeek exceeded 16 million downloads within 18 days of its launch, nearly double the 9 million downloads of OpenAI's ChatGPT during its initial release.
The surge in traffic once caused DeepSeek to crash, yet even so, the growth momentum remains strong, with monthly downloads exceeding 110 million. DeepSeek's brilliance can no longer be ignored by any competitor. On February 13, during an internal meeting at ByteDance, CEO Liang Rubo reflected on DeepSeek, acknowledging that the company needed to catch up in terms of speed this year.
Tencent's WeChat has integrated DeepSeek's AI search in a gray testing phase, and after usage exceeded expectations, it called upon the AI application Yuanbao to support WeChat search. On February 22, Tencent's Yuanbao surpassed Byte's Doubao, rising to second place in the free app download rankings in China, while DeepSeek continued to hold the top spot.
The "top two brothers" changed hands in just one month, forcing Doubao and Kimi, who relied on burning 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 entrepreneurial elite." Previous media estimates indicated that Kimi's daily advertising spend on iPhone channels was nearly 200,000, while Doubao's was 2.48 million.
Under the influence of DeepSeek, the Dark Side of the Moon has recently been reported to significantly cut its product advertising budget, including suspending multiple Android channel promotions and collaborations with third-party advertising platforms. According to insiders, promotional adjustments have indeed been made, stating, "There is some natural growth, but it cannot compare to DeepSeek's surge."
Kimi's current troubles extend beyond this: "Dark Underflow Waves" has exclusively learned that the long-stalled Kimi arbitration case has not reached the anticipated resolution but has entered the next phase of arbitration. According to insiders, the parties involved in the Kimi arbitration case, including the old shareholders of Loop Intelligence and Yang Zhilin, completed their payments at HKIAC (Hong Kong International Arbitration Centre) at the end of January and late February, respectively, and the formation of the tribunal has also been completed. The key figure behind the entire event, Zhang Yutong, may face separate legal action.
MiniMax also holds high hopes for its To C product, as its flagship product Talkie became the fourth most downloaded AI application in the U.S. in the first half of 2024, reaping the benefits. However, the good times were short-lived, as Talkie quietly disappeared from the Apple App Store in the U.S. market in mid-December, while the Android platform remained unaffected.
Step Star, Zero One Everything, Zhipu AI, and Bai Chuan Intelligent also have their own AI application products, but according to the AI product rankings, none of the top 20 AI applications in January 2025 are related to these four companies. An employee from Bai Chuan previously told "AI Light Year," "It's not surprising that the user retention and growth of Bai Xiao Ying are poor; we basically do not do advertising, letting others burn money to complete user education first."
Currently, DeepSeek, Tencent Yuanbao, and Byte Doubao dominate the top three spots in the Apple free app download rankings. The competition for the "Six Little Dragons" of large models to make the list will only become more intense, with the seventh-ranked Nano Search, led by Zhou Hongyi, personally stepping in to "promote."
Another formidable competitor is Alibaba, which has recently integrated its AI application Tongyi into Alibaba's Intelligent Information Business Group and has begun large-scale recruitment for its AI To C business, with hundreds of positions focused on products and technology development related to large models. With wolves in front and tigers behind, this accurately reflects the current predicament of the "Six Little Dragons" of large models.
When the technological narrative is no longer romantic, commercialization falls short of expectations, and the growth of monthly active users does not correlate with investment, the "Six Little Dragons" of large models face a stark reality.
Next Round of Financing Thresholds Raised
It is a well-acknowledged fact that pre-training large models is costly. Li Kaifu once revealed that the cost of a single pre-training session is about three to four million dollars. Even the lower-cost Yi-Lightning used 2,000 GPUs during training, taking a month and a half and costing over three million dollars.
Even DeepSeek, which prides itself on low costs, has incurred immeasurable initial investments. Third-party agency SemiAnalysis estimates that DeepSeek actually possesses a massive computing power reserve: it has stacked 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 for a general large model to be around 1 billion dollars, and this is just for the computing power part, not including the other two costly components: data and labor costs. Talent in the global large model field is currently very scarce," Dr. Du Feng, founding partner of Changmen Venture Capital and former head of Microsoft Ventures in Greater China, told the author.
Due to the need for such high investments, a saying has circulated in the industry for a long time: the entry ticket for investing in large model companies is 100 million dollars. Another signal behind this statement is that a large model startup that cannot secure financing is unlikely to survive.
After the outbreak of the hundred-model war in 2023, financing news was released almost every month. However, as the AI bubble theory gained traction, from September 2024 onwards, there was a long period without hundreds of millions of hot money flowing into the "Six Little Dragons" of large models. It wasn't until just before the Spring Festival in 2025 that Zhipu and Step Star announced they had secured "winter money," with the former completing a new round of financing of 3 billion yuan and the latter completing a Series B financing of several hundred million dollars.
Among the "Six Little Dragons," the other four have not released any financing updates for over six months: MiniMax announced the completion of a $600 million Series B financing in March last year, Bai Chuan Intelligent secured 5 billion yuan in Series A financing in July last year, Zero One Everything completed a new round of several hundred million dollars in financing in August last year, and the Dark Side of the Moon completed $300 million in financing in August last year.
During the Spring Festival, DeepSeek gained global popularity, and public opinion did not hesitate to praise DeepSeek and its founder, Liang Wenfeng. In the venture capital circle, there have been many rumors circulating recently about whether DeepSeek will initiate financing and what its valuation might be.
Earlier reports suggested that Alibaba would invest $1 billion for a 10% stake at a $10 billion valuation. In response, Alibaba Vice President Yan Qiao quickly refuted the rumor through social media, stating, "The information circulating about Alibaba investing in DeepSeek is false." Subsequently, foreign media reported that "DeepSeek is considering raising external funds for the first time," but DeepSeek representatives denied that any financing news was true.
"Many investors are directly or indirectly trying to meet with Liang Wenfeng. I predict the valuation should far exceed that of the current 'Six Little Dragons' in the large model sector," said an investor from CICC Capital. "DeepSeek has become a benchmark, and for the Six Little Dragons to secure new financing in the primary market, the threshold is clearly higher."
In fact, since the wave of large model startups began, there has been a general skepticism in the industry about whether all "Six Little Dragons" can survive as independent "large model companies." Several founders among the "Six Little Dragons" have expressed similar views in public, such as MiniMax founder Yan Junjie, who believes that in the future, only five large model companies will remain globally.
"There will definitely be a Chinese version of ChatGPT. Just like search engines, we have our own compliance requirements. However, the Chinese version of ChatGPT will only emerge from five companies: BAT + Byte + Huawei," said Cheng Hao, founder of Xunlei and partner at Yuanwang Capital, in an interview with the author.
With the ongoing explosive growth, the already diverging "Six Little Dragons" will accelerate the reshuffling.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
