Since the explosion of DeepSeek R1 on January 20, 2025, the AI venture capital circle has undergone earth-shaking changes in just 100 days. AI entrepreneurs who were once on the verge of bankruptcy have turned around thanks to DeepSeek, and the financing environment has also undergone a subtle shift. DeepSeek not only demonstrates the power of China’s AI, but also influences the direction of AI development in the world to a certain extent. However, while DeepSeek has brought new hope to the AI industry, the challenges faced by entrepreneurs have not diminished.
Zhang Yang (pseudonym), who is still on the verge of bankruptcy at the end of 2024, received the wealth brought by DeepSeek at the beginning of the year.
“It didn’t go bankrupt, it all depended on DeepSeek.” As the founder of an AI hardware company, Zhang Yang visited the docks of domestic investment institutions last year, and although he was born in a large factory, he did not hold the title of executive and failed to get the ideal financing amount; Although the companion AI hardware products developed have been mass-produced, they are on the verge of bankruptcy because “the market demand is too small, and the old capital is burned for research and development”.
Just after DeepSeek started a fire in 2025, Zhang Yang suddenly got an order from a well-known brand – Party A asked to develop a smart home for children based on the DeepSeek R1 model. At the beginning of the year, Zhang Yang also shifted from almost selling himself to “business running”, and the company’s fate drew a big upward curve because of DeepSeek.
“I really want to say goodbye to Liang Wenfeng.” Zhang Yang joked.
This shift is not an isolated case. Li Po, who left Microsoft in July 2023 to invest in AI entrepreneurship, downloaded the DeepSeek APP on his mobile phone for his parents in his hometown during the Chinese New Year, “There has never been such an AI application, so that AI has truly broken the circle.”
Subsequently, Li Po, who was still in the seed round of financing last year, faced the review of the conditions required for Pre A and A rounds of financing, and this year, his team has “chatted smoothly” with several overseas leading dollar funds.For Li Po, who went overseas early, DeepSeek, which went abroad, has become the overseas financing business card of Chinese entrepreneurs, “DeepSeek not only proves that the level of domestic AI technology has caught up with foreign countries, but also proves that Chinese AI teams often have more potential in cost reduction and commercialization.” ”
Looking back at the release of DeepSeek R1 in January 2025, within 100 days, the AI venture capital circle has turned around.
In August 2024, Zhu Xiaohu, managing partner of GSR Venture Capital, still asserted that “no independent large model company will survive in five years”, believing that the only way out for AI Xiaohu is to sell himself to a large factory. But within 60 days after DeepSeek became popular, in March 2025, he posted in the circle of friends: “If DeepSeek opens financing, I am willing to enter the market at any valuation.” ”
However, one detail worth noting is that according to the observation of Alphabet List (ID: wujicaijing), AI investors seem to have regained confidence, but they have chosen to bypass the large model entrepreneurship field where DeepSeek is located.
B-end product manager’s ability model and learning improvement
The first challenge faced by B-end product managers is how to correctly analyze and diagnose business problems. This is also the most difficult part, product design knowledge is basically not helpful for this part of the work, if you want to do a good job in business analysis and diagnosis, you must have a solid …
View details >
Since 2025, a total of nearly 30 financings have occurred, with high financings appearing in the fields of AI applications and embodied intelligence. For example, Huachuang Capital invests in silicon-based liquidity, and its inference acceleration service for DeepSeek is a key factor in driving it out of the circle. Zhen Fund and others have bet on embodied intelligence, Unitree Technology raised 1 billion yuan in B round of financing to set a track record, and Shi Zhihang also set a record with an angel round of financing of 120 million US dollars. Private equity institutions have begun to explore vertical fields such as “AI + healthcare” and “AI + law”, such as the valuation of Harvey’s legal contract generation tool has doubled in half a year.
At the same time, DeepSeek, which focuses on low-cost and open source, not only has the threshold for AI application development plummeted, but also made AI entrepreneurship more volatic.
Taking AI Six Tigers as an example, the dark side of the moon was cut off 70% of the marketing budget of the Kimi chatbot, and instead focused on model training, trying to replicate DeepSeek’s technical route; Zhipu AI, which once laid off 20% of its workforce in 2024 due to commercialization pressure, not only announced the open source GLM series model in March to join the open source army, but also received more than 1.5 billion yuan of state-owned assets in Hangzhou and Zhuhai, trying to compete for the right to speak as a “national team”.
The AI Agent “WorkfxAI” developed by Li Po focuses on vertical scenarios such as e-commerce, banking, and medical care, and has achieved nearly one million revenue with the fine-tuning model of vertical field data.
“Now the competition is no longer about technical parameters, but about who can deeply bind the model to the scene faster and better.” Li Po said.
DeepSeek has not stopped iterating. On March 25, DeepSeek officially announced that the V3 model had completed a minor version upgrade. DeepSeek doesn’t seem to stop iterating. Subsequently, on April 30, the DeepSeek-Prover-V2-671B large model (focusing on the field of mathematical proof) was released. Whether it is good or bad, this earthquake caused by DeepSeek’s “unintentional” will obviously bring deeper and farther shocks.
01
“Domestic financing looks overseas, and the emergence of DeepSeek has made overseas investors pay more attention to China’s AI team.”
Li Po is one of the first members of Microsoft’s Bing search team, from Alibaba to Microsoft, and then to Microsoft in July 2023 to start an AI business, he has set up teams at home and abroad, and last year, he failed to get a satisfactory financing amount.
As of January 2025, before DeepSeek exploded out of the circle, AI entrepreneurs such as Li Po faced a sluggish investment market.
In the field of Agent where Li Po is located, if you want to integrate angel wheels in China, you need to meet the requirements of annual revenue from 0 to 1 million US dollars within half a year.At that time, whether it was the B-end or the C-end in China, there was no clear understanding of AI, which made start-ups such as Li Po feel a little harsh, “Rongangel round, seed round, investors are basically according to the standards of PreA or A round, not a large number of users can get money, and the amount of revenue and payment rate are required according to the A round/Pre A.” ”
Abroad, when it comes to the Chinese team, “backward technology, copy” has become a key word, and Li Po’s team has not been able to get attention.
After DeepSeek exploded out of the circle during the Spring Festival, the wind direction changed.
Li Po doesn’t seem to have to bother to introduce what AI is to relatives and friends, and DeepSeek’s out-of-the-circle has made the parents in his hometown understand.At the same time, DeepSeek’s technological breakthroughs have also made overseas investors realize that “China’s AI technology has caught up, although it may not be as good as OpenAI in terms of performance innovation, but DeepSeek has proven that we can bring down the price.” ”Overseas funds that ignored Li Po last year have also become inclined to look at Chinese teams this year.
Now, Li Po’s seed round of financing has talked about the final round, “The goal is to raise $5 million.” ”
Fang Cheng (pseudonym), a partner of a leading domestic venture capital fund, also told the alphabet list that after DeepSeek knocks down the price, the AI application market will usher in an explosion, and it has become a consensus to invest in AI upstream such as chips, and downstream such as applications and hardware to actively shoot, “The emergence of DeepSeek has indeed made everyone’s cognition converge.” ”
Fang Cheng found thatThe emergence of DeepSeek has not only made large companies rush to launch deep thinking models, but even AI entrepreneurs are learning DeepSeek.
Li Po’s “WorkfxAI” has also begun to add the process of thinking and presentation throughout the product design and various joints. The popularity of DeepSeek made him realize that whether it is ToB or ToC, users want to be more transparent and explainable, and users want to know what AI is doing.
Like Li Po, after DeepSeek came out of the circle, Cheng Sen not only replaced part of the product model with DeepSeek, but also modified the internal process of the product under inspiration.
In September last year, after running away from a leading Internet company to start a business, Cheng Sen rushed to launch his own Agent tool in January this year, and when analyzing the reasons for DeepSeek’s exit from the circle, Cheng Sen realized that the high generation quality of DeepSeek was due to internal thinking to show the outside world, so that C-end users could directly issue instructions in the CoOT link (Chain of Thought, that is, Chain of Thought, which refers to AI understanding and responding to user problems through logical reasoning and information processing) to constrain the large model.
thereforeCheng Sen also added nodes within the product to add the machine thinking process to the next round of output, “After adding the thinking process, the output quality is indeed higher.” “Although the monthly token consumption cost is 2-3 times higher than before, the number of registered users has also increased from 4,000 in February to more than 30,000 today.
However, Fang Cheng admitted that although the team has doubled the number of projects they have seen recently compared with November last year, the industry has tacitly bypassed the large model base field where DeepSeek is located, and is more inclined to invest in AI applications and embodied intelligence tracks.
“DeepSeek proves that large models do not have to pile computing power to improve performance, but it has also raised the barriers of large models again. Fang Chengcheng said that even if investors have loosened their mouths about entrepreneurs, in fact, there are not many who really start.
02
For entrepreneurs, today’s investment market seems to be like the spring sun, looking bright but difficult to transmit.
“Compared with last year, there are more investors looking at the agent direction, but it is more difficult to raise funds.”
Cheng Sen told the alphabet list that last year, like Li Po, he was blocked by almost harsh financing conditions, and the title of the leading Internet giant could not be successfully “realized”, and finally he chose to take the money of individual investors as a start-up fund.
With the explosion of DeepSeek, investors look more active than last year, but DeepSeek has further lowered the technical threshold for ordinary developers, and with it, there are more competitors, and it is more difficult to get money.
In factIn the 100 days since DeepSeek exploded, the large model track has become more and more crowded.
AppGrowing data shows that since February, Kimi’s delivery has declined sharply, from more than 100 million yuan in December and January 2024 to 44.25 million yuan in February.
At the same time, when DeepSeek announced that it had a theoretical profit margin of 545%, even at a very low token price, it could also make a profit, with a valuation of more than 20 billion yuan, and Zhipu, which had just received more than 1.5 billion yuan in financing in various places, reported that the sales in 2024 would be 300 million yuan, and the loss would still reach 2 billion yuan.In front of the open source and more well-known DeepSeek, AI entrepreneurs, even the leading six tigers, are facing a fierce knockout round for the commercialization of B-end business.
MiniMax also recently reported the resignation of Wei Wei, the head of the open platform. In response, MiniMax responded that the domestic B-end business has ushered in a new stage of development, and the team will be led by other responsible persons. The commercialization pressure of large model companies brought about by the DeepSeek craze seems to be not immune to MiniMax.
And even if you do a vertical scene, the volume will only increase.Facing AI entrepreneurs is a commercialization problem that needs to be answered urgently.
Zero One Thing, founded by Kai-Fu Lee, has stopped “pre-training” large language models and instead focused on using DeepSeek’s models to sell customized artificial intelligence business solutions. Baichuan Intelligent turned to the healthcare market.
Cheng Sen chose to develop customized services for smaller B-end enterprises, delineating two age groups: one old and one young.
Under the popularity of Deepseek, Cheng Sen announced on Xiaohongshu that his agent had been connected to Deepseek, and the number of likes could reach more than 4,000, which was the most popular post since his founding. But as the number of users has increased from 4,000 in February to more than 30,000, and the number of tokens consumed per month has increased by 2-3 times, Cheng Sen still needs to find a way for the team.
However, whether it is in the form of team settlement support or exploring the charging model of commercial SaaS, Cheng Sen is frank,These are just possible sources of cash, “After all, domestic commercial SaaS has been developing for many years, and large manufacturers have not yet formed a particularly good model.””
Different from the confusion of investors when he started his business at the beginning of last year, Cheng Sen’s team decided to narrow down the user group to two age groups, although the spending power of these two groups was questioned, Cheng Sen decided to do “valuable things, it doesn’t matter whether you can make money this year.” ”
03
It seems that the waves stirred up by the stone of DeepSeek are still limited.
For entrepreneurs, in the 100 days of DeepSeek’s explosion, what has not changed is the pressure caused by the never-ending technology competition.
Li Po’s WorkfxAI maintains an update frequency that will be iterated in 1-2 months, and the model function updates of large manufacturers such as foreign Google, OpenAI, and domestic bytes, Alibaba, DeepSeek, etc. will become WorkfxAI’s update progress reference, “As soon as Gemini and OpenAI are updated, we will also update immediately.” ”
In order to maintain the technical barriers of providing an agent platform for B-end users, Li Po’s team had to walk on the road of continuous iteration.
Just recently, DeepSeek Vv3 has implemented feature updates, and the R2 model focusing on reasoning capabilities is also expected to be released before May. In March, OpenAI made major updates to GPT-4o and Sora, launching a new Wensheng graph model. Not only does it support continuous questioning, style transformation, image PPT, but it even surpasses Midjourney, the leading platform in this field, in some performance.
A flapping of wings from a major technology company may be a storm for entrepreneurs, “The algorithms and models that have spent so much time and manpower to tune may be replaced by the update of large models at one time.” Cheng Sen added.
In order to keep up with technological iteration, entrepreneurs need to compete with large factories for people and resources. Although after the explosion of DeepSeek, the strategic adjustment of large companies such as Byte, Alibaba, Tencent and other large companies has suddenly made many excellent technical talents appear in the market, but “there are still not enough people, good algorithm talents, and the annual salary package is not something that ordinary startups can catch.” Fang Chengcheng added.
It should be noted that in this unknown competition, whether it is a large company or an entrepreneur, the determination to continue to invest in AI will not change, and even within more than 60 days of DeepSeek’s explosion, it has been further strengthened.
In addition to following DeepSeek and grabbing spillover dividends, Tencent not only actively promotes ingots, but also launches its own T1 deep reasoning model in a timely manner. At the same time, Alibaba has also recently increased the importance of quark again, positioning it as Alibaba’s flagship AI application.
According to AppGrowing data, since the end of February, Tongyi Qianwen has gradually increased its investment in advertising, and entered the TOP10 after the release of the open source reasoning model QwQ-32B on March 6. Byte’s Doubao large model department (Seed) also proposed at the plenary meeting that the model application should be done in accordance with the model capabilities for a long time, and the beanbag inference model should be fully updated.
“Big models are like nuclear weapons, big companies have to have them anyway. Because if there is a major breakthrough in model capabilities and you don’t keep up, your users will be robbed. Fu Sheng’s judgment on the competition of contemporary AI applications also echoes the actions of large manufacturers.
generally“DeepSeek is more like an inspiring, its emergence has made the market have a consensus, so that entrepreneurs are more excited, but the real actual effect is not obvious, DeepSeek may need to equal OpenAI in all aspects of general capabilities, if it can still have such a low price at that time, it will completely lead an AI revolution.”Li Po added.
Whether it is a big factory or an AI Six Tigers, or even DeepSeek, this battle around the base model and AI application has not yet come to an end, and the 100 days of DeepSeek’s explosion may be more like a technical footnote, Chinese entrepreneurs have begun to take the wave, and next, there are more unknowns waiting for them.
