GuShiio.com Wen| Pai Finance, author| xiaojiu
Since the beginning of this year, DeepSeek has shocked the investment circle and capital market.
On February 21, DeepSeek posted on social platform X that it had built a small team exploring AGI (General Artificial Intelligence) and would open source five code bases starting next week to share research progress in a completely transparent manner.
The day before yesterday, DeepSeek just denied that external financing was false news. Previously, foreign media reported that DeepSeek was considering external financing for the first time to cope with the growing demand for computing power. As early as early as early February, it was reported that Alibaba planned to invest in DeepSeek at a valuation of US$10 billion and subscribe for a 10% stake. However, Ali Vice President Yan Qiao later refuted the rumor on social media platforms that the online news was false.
In fact, since the beginning of this year, many leading companies and investment institutions have approached DeepSeek. However, DeepSeek has strong financial support behind it and is not short of money.
How strong is DeepSeek? Who benefited from DeepSeek?
How strong is DeepSeek?
During the Spring Festival this year, Deepseek released the open source model DeepSeek-R1, which quickly swept the world with its strong reasoning capabilities and extremely high cost performance. Once released, it attracted widespread attention inside and outside the industry.
How popular is Deepseek? On the Apple App Store, Deepseek ranks first in the free app store list in China, and has also surpassed the generative AI products of American technology companies such as ChatGPT, Meta’s social media platform Threads, Google Gemini, and Microsoft Copilot in the United States, soaring to first place. According to statistics, Deepseek has also won the championship in the app store download list in 140 markets around the world, becoming a popular target of users around the world.& zwnj; Behind this ranking is the high recognition of Deepseek in the user market, and it is also ironclad evidence that it has rapidly emerged and gained a place in the global artificial intelligence market.
Before DeepSeek, what shocked the global AI industry was ChatGPT developed by Open AI. Since OpenAI released o1-preview in September 2024, it has launched a large-scale catch-up in the market, and inference models that can match or even exceed its performance have emerged one after another.
The main reason why DeepSeek was able to break through is that it not only took the lead in achieving the effect of the Open AI-o1 model, but also reduced the cost of the reasoning model to a very low level.
Specifically, the technological breakthrough of DeepSeek-R1 lies in using pure deep learning methods to allow AI to spontaneously emerge reasoning capabilities. According to DeepSeek’s official website, DeepSeek-R1 uses reinforcement learning technology on a large scale in the post-training stage, greatly improving the model’s reasoning capabilities with only very little annotated data. In terms of tasks such as mathematics, code, and natural language reasoning, the evaluation performance is close to the official version of the GPT-o1 model developed by the United States Open Artificial Intelligence Research Center (OpenAI). For the first time, the o1 model has achieved true universal reasoning capabilities and can surpass human experts in doctoral scientific question and answer sessions.
How strong is DeepSeek R1? Some testers used commonly used collision detection to verify, requiring large models to write a program to make a small ball bounce in a slowly rotating geometric shape and keep the small ball stay in the shape. Collision detection has a huge test on the model’s reasoning ability, and a slight carelessness can lead to visible logistics errors. On the surface of the test results, R1 ‘s performance is significantly better than the OpenAI o1 pro model.
In addition to its strong performance, another major advantage of the R1 is its high cost performance. It is understood that the training cost of the DeepSeek-R1 model is only US$5.6 million, which is far lower than the hundreds of millions or even billions of dollars invested by many American technology giants in artificial intelligence technology, such as ChatGPT-4 training. The cost is as high as over US$100 million. According to technical reports, R1’s service for developers is priced at 14 yuan per million tokens (input strings), which is only about 1/30 of that of O1.
Liang Wenfeng, founder of DeepSeek, once said in an interview with the media that both APIs and AI should be universally beneficial and affordable to everyone.
It is worth mentioning that DeepSeek adopts a completely open source strategy, which on the one hand lowers the threshold for users to use. On the other hand, DeepSeek after open source has attracted the attention of many developers and researchers, further promoting the collaborative ecosystem of the AI developer community. Prosperity and promote the development of AI technology.
Previously, OpenAI has also expressed its hope to promote the development of digital intelligence in a way that is most likely to benefit all mankind, without being limited by the need to generate financial returns. However, OpenAI’s open source strategy only persisted until the release of GPT-3. After the release of GPT-4, it hid its training data and model weights and gradually moved towards closure.
Jim Fan, senior research scientist at Nvidia, said on his personal social platform that we are at a historical moment: a non-American company is continuing OpenAI’s original mission to empower all mankind through truly open cutting-edge research.
DeepSeek, a mysterious Eastern force, has caused the secondary market to question the competitiveness of the U.S. technology industry. Affected by this, the share price of U.S. chip giant Nvidia plunged 17%.
DeepSeek’s parent company, DeepSeek, was established on July 17, 2023. It is supported by the famous quantitative private equity magic square fund and has strong financial strength, which also provides solid support for its continued exploration of technology without making profits.
As early as May 7, 2024, DeepSeek-V2, which was released in depth, also adopted the open source model and became a comprehensive Chinese capability in the open source model once released (AlignBench) is the strongest representative, and was included in the same echelon as GPT-4-Turbo, Wenxin 4.0 and other closed-source models in the evaluation; the English comprehensive ability (MT-Bench) is in the same echelon as the strongest open source model LLaMA3 – 70B, surpassing the strongest MoE open source model Mixtral 8x22B; the results of knowledge, mathematics, reasoning, programming, etc. are among the top.
In addition, its API price is only 2.7% of GPT-4o, further intensifying the price war of domestic large models. In the following week, Byte, Alibaba, Baidu, and Tencent successively announced price cuts.
At the end of 2024, DeepSeek-V3 was released. In the fields of knowledge tasks, long text understanding, programming and mathematical operations, its performance is close to or even exceeds the world’s top closed-source models such as GPT-4o and Claude-3.5-Sonnet.
The entry of DeepSeek has directly triggered a chain reaction in the global AI field. Looking at the global market, cloud computing platforms such as Microsoft, Nvidia, and Amazon have connected to DeepSeek, which has also stimulated OpenAI to launch the OpenAI o3-mini series of models in disguise, opening it to users for the first time. Use rights for the inference model.
Who connected DeepSeek?
Under the DeepSeek wave, according to incomplete statistics, more than 200 companies have announced access to DeepSeek, covering basic telecommunications companies, cloud computing, chips, finance, automobiles, mobile phones and other fields. Among them, many cloud platforms including Huawei, Alibaba, Baidu, Tencent, Jingdong, etc. have announced access to the DeepSeek model, and about 20 car companies have announced that they have completed deep integration with DeepSeek in smart cabin scenarios or AI operations. In addition, a number of securities firms, banks, and public funds have also said they have connected to the DeepSeek model.
In fact, domestic manufacturers have previously made arrangements for their own AI large-scale model strategies. The access of their main APP to third-party large-scale model services can be seen as a strategic change for the domestic manufacturers and a bet on the future of the company.
Tencent launched the DeepSeek-R1-based AI search function in a recent grayscale test, which is seen as a strategy to actively embrace AI. Immediately afterwards, Tencent Maps, QQ Music, Tencent Documents and other products were quickly connected to DeepSeek.
Immediately afterwards, Baidu Search also followed up with DeepSeek. Before announcing access to DeepSeek, Baidu had changed its development path of the big model. First, it announced that all were free to use, then changed its closed-source strategy and began to embrace the open source route, saying that it would launch Wenxin Big Model 4.5 series, towards open source.
Among the several major Internet companies currently, Byte is more cautious about embracing access to DeepSeek. Currently, on the business side, only Feishu official announced that the DeepSeek series of models has been launched, but the entrance is also relatively hidden.
What is worthy of recognition is that DeepSeek’s open source ecosystem has greatly accelerated the implementation of AI scenarios and promoted the further transformation of the AI industry from hardware facilities to software application layers. Dachang’s cloud service was the first to connect to DeepSeek, which is due to the unprecedented popularity and extremely low cost advantages brought by DeepSeek. It is expected to become a breakthrough in breaking the market price war and further consolidate its existing market position.
How strong is DeepSeek?With the explosion of DeepSeek, a large amount of traffic poured in, and accompanied by some hacking attacks, DeepSeek’s official website became stuck from time to time.
Compared with other large model language products on the market, the difference between DeepSeek on the user side is that it displays a detailed thinking process.
How to make good use of the AI model and make it a tool to effectively improve work efficiency is also a matter of knowledge. Careful netizens have already summed up a set of rules. If you become a leader of AI, you will face some employment difficulties like a real leader.
A smart but non-working deepseek, a stupid but hard-working doubao, a moderate kimi who doesn’t give you emotional value, and a chatGPT who I can’t afford to study overseas.
Some netizens also carefully concluded that some management talents need to be utilized. First use smart deepseek to formulate strategies (you can only answer once, you need to cherish the opportunity), and then use hard-working bean buns to fill the content (you can output countless times and use unlimited). Kimi can be used for horse racing. Compare the kimi and bean buns with the best. Information involving overseas can be checked again using chatGPT.
But it is worth noting that these AIs will deceive you if you don’t carefully distinguish between being deceived every minute.
At the application level, DeepSeek’s huge profits and worries
With the explosion of DeepSeek, a group of people have already made money.
Share on social media platforms how ordinary people can collect tuition fees through DeepSeek to start a sideline business and start a self-media business. Common words are that low-threshold entrepreneurship is easy to get big money and the biggest monetization outlet in 2025.
Some bloggers said that the relevant community attracted 4000 people to pay to join in four days, earning nearly 200,000 yuan.
The topic #deepseek on Xiaohongshu has 820 million views and 5.263 million discussions; on Douyin, the topic #deepseek has 11.46 billion views.
There is a Yiwu landlady who uses DeepSeek+Al to do multinational business. She only needs to say 12345 in front of the camera and can generate videos in dozens of languages.
Some netizens followed DeepSeek to buy lottery tickets and win, while others tried to use it to calculate the fortune of the next year.
Xiaohongshu user @Yapie programmer brother used deepseek to create the song “Seven Days of Love” that became popular on the NetEase cloud. From February 11 to 0:00 on February 17, the song’s effective playback volume was 186,000, the collection volume reached 9555, and the number of comments was 1769. This song also brought him an average income of 150 yuan per day.
It is worth mentioning that following the explosion of DeepSeek, some risks and challenges have emerged.
On the one hand, it comes from the risk of data leakage. Cyberattacks may lead to the theft of DeepSeek’s user data, technical data, etc. Once data is leaked, it will not only harm the interests of users, but also expose DeepSeek to legal liability and reputation damage, and may even be used by competitors to obtain commercial benefits.
In addition, Deepseek is a small start-up AI company, and its internal talents face the risk of poaching talents from major domestic companies such as Alibaba, Baidu, Tencent and major American Silicon Valley companies;
In addition, there is some pressure from the U.S. government. The United States launched an investigation into DeepSeek on national security grounds, behind which it is trying to maintain its hegemony in the AI field. More policy restrictions on DeepSeek may be introduced in the future, and other countries may even jointly suppress it and limit its development in the global market.
The most important thing is that the AI model has not yet seen the prospect of large-scale commercialization despite its vigorous development. Under the price war, how to recover hundreds of millions of R & D costs needs to continue to wait and see.
