In 2026, where will the money for AI applications go?

This article may contain an automated translation. The original source is linked for verification.

In 2026, where will the money for AI applications go?

Fixed focusOne

2026.11.11 09:42

· From Hebei

Full text 5488 words

Text | Dingjiao One (dingjiaoone), author｜Li Mengran, editor｜Wei Jia

In 2026, the "biggest" money in the AI financing market will still flow to basic capabilities such as large models and computing power. But more and more frequent financing stories begin to occur at the application layer.

According to statistics from Zhixixi, in the first half of 2026, among the hundreds of millions of yuan in financing in the domestic AI industry chain, the AI application direction had the most financing events, with a total of 33 cases. Among them, video, image and 3D generation accounted for nearly 40%, and AI Agent accounted for about 20%. Enterprise applications, AI medical/drugs, AI finance, AI marketing and other directions have also seen billion-yuan financing.

Megan is the founder of the GTM public welfare community of Origin Community and a domestic C-end application overseas trader. She told "Focus One" that from the experience of communicating with investment institutions, capital's attitude towards AI applications has experienced a round of changes this year. The first half of the year was generally positive, but by the end of summer and early autumn, some institutions turned cautious. There are two reasons: first, tracks such as embodied intelligence diverted attention; second, some AI applications failed to achieve daily activity, popularity, and commercialization after investing a lot of money, and the exit path began to receive more discussion. After entering October, market attention has picked up, but investors pay more attention to self-developed model capabilities, technical barriers or software and hardware combinations that can create differences.

Investors have become cautious, and the flow of money has become uneven. AI creation company Yanyu Technology completed nearly US$300 million in Series B+ financing in June, with a post-investment valuation of more than US$2 billion. In industries such as medical care, education, marketing, and security, more companies have received financing ranging from tens to hundreds of millions of yuan. At the same time, major manufacturers such as Byte, Alibaba, Tencent, and Kuaishou appear on almost every popular application track.

Why is money concentrated in a few tracks? Do industries with limited financing have smaller opportunities? Under the attack of big manufacturers, how can start-up companies survive?

01. Video, image and 3D, be the first to get “big money”

Putting this year's large financings of AI application companies together, the flow of funds will be more intuitive.

IT Orange statistics show that in the first half of this year, a total of 84 financings were raised in the AIGC application track represented by image and video generation, totaling 59.605 billion yuan, with an average of 710 million yuan per transaction, the highest among all AI subdivisions. Among them, Shengshu Technology, Sand.ai, LiblibAI/Yanyu Technology, and Aishi Technology, four AIGC application companies have raised a total of 9.378 billion yuan.

Since the beginning of this year, large-scale financing has continued to focus on companies related to visual creation.

After the AI video company Aishi Technology completed a US$300 million Series C round of financing in March, the overall Series C financing scale reached 2.98 billion yuan in July; in June, LiblibAI parent company Yanyu Technology completed a B+ round of financing of nearly US$300 million, with a post-investment valuation of more than US$2 billion; in July, Shengshu Technology completed a US$500 million B+ round of financing; AI 3D company VAST announced the completion of Series B and B+ rounds of financing in September, with a total amount of approximately 3 billion yuan, and a cumulative financing of approximately 5 billion yuan in less than half a year; Meshy completed nearly US$400 million in Series B financing in July.

Why are video and 3D the first to make big money?

Ding Boran, an investor in Dewu Capital, told "Focus One" that the first thing capital looks at is the market ceiling. In ToC applications, video is still one of the media forms with the richest content and the most widely demanded content, and 3D may become a new content carrier in the next stage. For capital that needs to bet on "big opportunities", this type of track is more likely to accommodate large companies and is more worthy of continued investment.

The market is big enough, which is just a prerequisite. Bryan Liu, investment director of Alumni Ventures, told "Fixed Focus One" that visual generation also has another feature that is easily favored by capital: its product feedback cycle is shorter.

"After generating a video, a picture or a 3D material, whether the content is available and whether the user is willing to use and pay can often be verified in a relatively short period of time." This feature makes visual generation easier than some industry applications that require long-term verification of effects to show how this generation of products is better than the previous generation.

A group of visual generation companies have already achieved quantifiable commercialization indicators in advance.

PixVerse, a subsidiary of Aishi Technology, will launch commercialization in November 2024. By October 2025, the disclosed ARR has exceeded US$40 million, with more than 100 million users. Revenue mainly comes from member subscriptions, mainly in overseas markets; Yanyu Technology disclosed that as of May 2026, the company's overall ARR has reached US$300 million. Capital is beginning to be able to use indicators such as user scale, subscription revenue, and API calls that are closer to traditional Internet companies to measure an AI application company.

However, rapid commercialization is not the whole reason. Visual generation is also a category with heavy capital investment among AI applications. Aishi Technology's financing purposes include video generation basic models, real-time world models and global product growth; VAST not only operates Tripo, but also continues to invest in the research and development of 3D basic models, data and world models, which require continuous consumption of funds.

Aishi Technology previously revealed that if the investment in self-developed large models is not included, its product business can basically be self-financing. In other words, there is a layer of continuous model investment between the revenue generated by the product and the capital needs of the entire company.

Another type of company has proven that it can still achieve high valuations without developing its own underlying model. For example, Yanyu Technology does not rely on self-developed general basic models to build products. Instead, it aggregates different models and puts more capabilities into creative tools, workflows and user ecology. The company has further expanded from the LiblibAI community to design Agent Star Stream and video creation platform LibTV. As of May this year, ARR has reached US$300 million.

Therefore, behind "video and 3D attract the most money", there are actually two different logics superimposed: one type of company needs to use heavy capital to continue to improve model capabilities; the other type can also rely on the application itself to form users, revenue and ecosystem.

In addition to visual generation, Agent is another direction where large amounts of financing are beginning to appear. In October, Today AI, a personal agent company that had just launched its product, announced that its cumulative financing exceeded US$200 million. Another star agent company, Manus, completed a new round of financing of more than US$500 million in the same month, with a target valuation of approximately US$4 billion, becoming one of the most watched large-scale financings in the Agent track this year.

Different from visual generation, the money received by the individual agent is mainly invested in the team and entrance. Today AI's products were only launched in September. Capital is interested in the founding team's product experience from major manufacturers and the possibility of personal agents becoming the entrance to the next generation of AI.

02. Where the scale of financing is not that large, are the opportunities smaller?

In terms of financing scale, video, imaging and 3D are almost the most prominent tracks for AI applications this year. But if you look at the frequency, more vertical industry applications are more active.

iiMedia Consulting data shows that in the first half of 2026, although the financing amount of AI industry applications (excluding AIGC and AI general applications) was only 25.792 billion yuan, 133 financings occurred, which is the category with the most financing events among the major directions in its statistics, with an average of about 194 million yuan per transaction.

It can be seen that the frequency of getting money from industry applications is not low, but the funds are more dispersed.

In Bryan Liu's view, the two types of tracks themselves cannot be simply compared horizontally. Video, images and 3D are technologies and product forms, while medical care, education, marketing and security are industries. What really widens the gap is the different ways in which different businesses verify their value.

The first type is a business that takes longer to prove “results”. Medical care and education are typical "long feedback" businesses.

Ding Boran believes that both medical and education are subject to data security, privacy, compliance or capitalization paths. For medical startups, the cycle of obtaining industry data, completing product verification and entering real business processes is longer, while education companies are difficult to achieve capitalization.

Bryan further explained that medical treatment needs to prove clinical effects, education needs to verify learning effects, and some ToB scenarios need to further connect procurement, payment and responsible entities. “It’s not that these industries have smaller opportunities, but that the verification paths are different and some industries require more time.”

SenseTime is an example. This company, which was spun off from SenseTime, completed a Series A financing of more than 500 million yuan in April this year, with a valuation exceeding US$1 billion. In July, it completed a Series B financing of more than US$100 million, with a post-investment valuation of more than 10 billion yuan. Its SenseCare platform currently discloses that it has cooperated with more than 500 hospitals and obtained more than 30 medical device registration certificates. Compared with a C-side application that can be verified by user growth after being launched, medical AI obviously needs to accumulate not only model capabilities, but also hospitals, clinical data and compliance systems.

Education is similar. Zhibei Technology, founded in 2023, completed the Pre-A round of financing in January this year, with an amount of several million US dollars; Dance with Love, founded in the same year, and its AI teacher product "Ai Xue" has exceeded tens of millions of yuan in monthly revenue. As of November 2025, it has completed 4 rounds of financing, with a total financing amount of approximately US$150 million. Its products spent several months testing indicators such as course completion rate and refund rate before going online. Whether an AI education product can truly improve learning effects often requires longer observation.

For this type of company, capital needs to wait for the product to prove its effectiveness.

The second type is a business whose value can be directly proven by its operating performance. Their financing scale may not be large, but they may have clear revenue and customer data earlier. GEO, which is heating up rapidly this year, is an example.

The smart push era, which was only established in 2025, completed tens of millions of yuan in angel rounds this year. As of the first half of this year, the company disclosed that ARR has exceeded 100 million yuan; it currently serves more than 400 customers, and it is expected that the ARR for the whole year of 2026 will be close to 200 million yuan. Deepin Yunhai, which was also established in 2025, completed a multi-million angel round of financing in October 2026, with annual revenue reaching tens of millions of yuan.

Compared with training video-based models, the financial needs of such companies are obviously not at the same level. However, the smaller financing amount does not prevent them from verifying their needs with customers and income first.

The third type is to wait for businesses with mature demand, and the pace of financing depends on the maturity of the market. AI safety is such a track.

An AI security practitioner told "Focus One" that it is not that there is no demand for security, but that many companies' emphasis on AI risks has not fully kept up with the speed of application. Before there is a strong enough risk event, enterprises are not willing to allocate budget specifically for AI security.

But as Agents begin to access internal enterprise data, call tools, and even perform tasks autonomously, this demand has begun to become apparent. After EgeCloud completed hundreds of millions of yuan in Series B financing in May this year, it further extended its products from office security to unified security management of "people + AI". In September, it completed nearly 100 million yuan in Series B+ financing, and funds continue to be invested in the corporate governance platform of people and AI Agents.

Opportunities in some industries will wait until AI truly enters the core business process, and then the original potential demand will become a clear purchase.

Therefore, the amount of financing does not equal the size of the opportunity. Some tracks have to wait for the results to be proven, some have to wait for income to speak first, and some have to wait for demand to mature. To judge whether a track has a chance, instead of looking at how much money it takes, it is better to look at how it uses and how long it takes to prove its value.

03. When the giants are doomed, start-up companies still have to pass three hurdles

Competition for AI applications is much more complex than it was a few years ago.

On the one hand, large model capabilities are still being rapidly upgraded. A product that seems established today may be covered by an update to the basic model tomorrow. On the other hand, Doubao and Yuanbao are competing for the general AI entrance. Alibaba, Tencent, and Byte are all working on offices and agents. Video generation, AI programming, education, marketing and even medical treatment are also crowded with large companies.

In Megan's view, the advantage of large manufacturers is not just more money. Compared with entrepreneurial teams, they produce products faster, have more mature investment channels and user growth mechanisms, and have a large number of existing users and data. If new applications do not have these experiences and just learn user growth, they will be much slower than them.

For startups, they need to prove what they can leave behind after the big manufacturers enter.

The first is whether startups can establish their own independent value when big companies are both allies and rivals.

Now, the relationship between big companies and startups is becoming more and more complicated. In October, the personal agent startup Today AI announced that it had completed a total of more than 200 million US dollars in financing. Among the investors, Tencent and Ant Group also appeared, and the angel round and the large model company Step Star participated. Its founder Qi Junyuan was previously the vice president of Byte Feishu products and the person in charge of the PC version of Doubao.

But on the other hand, Tencent is promoting its own Agent products, and Ant-owned Lingguang is also increasing its personal task execution capabilities. For Today AI, industrial capital has brought funds and resources, but it has not eliminated competition. It still has to compete with shareholders' products for the next generation of personal AI entrance.

Such relationships are becoming increasingly common. While large companies invest in external companies, they also engage in similar businesses themselves. A startup needs to prove that even without the resources of shareholders, it still has value that cannot be easily replaced.

Secondly, with the rapid iteration of model capabilities, the technical barriers of application companies are also constantly being impacted.

Bryan Liu believes that if a company's advantage is just to call a better general model, this ability is not exclusive. Model upgrades may even have two opposite effects simultaneously. On the one hand, it can help application companies reduce development and delivery costs; on the other hand, it can also turn a company's originally charged functions directly into basic capabilities that everyone can obtain.

Today AI acknowledges this pressure. In an interview with the media this year, his team stated that Personal AI is still in a highly homogeneous competition stage, and the differences in one or two functional points may soon be filled by competitors.

In Megan's view, one type of opportunities for startups comes from scenarios that are segmented and experimental enough that big companies temporarily lack the motivation to enter; another path is to first build communities and relationships around a small group of core users, and then let the product grow. The advantages of such companies do not necessarily come from stronger technology, but may come from a deeper understanding of specific users and the team's accumulation in community, growth or a certain industry.

When Ding Boran judges AI application companies, he values unique data, industry know-how, customer relationships, and embedding in real workflows. For example, ToB applications can quickly acquire customers and embed them into customer workflows, while ToC applications can quickly acquire customers and accumulate user data, forming a data flywheel. These are things that are difficult for large manufacturers to copy in the short term. If enough financing can be obtained and subsequent customer acquisition can be accelerated, certain barriers can be formed.

But Megan believes that “data” itself does not necessarily constitute a natural barrier. Compared with static data obtained once, what is more valuable is the behavior, interaction and memory left by users using the product for a long time, as well as the continuous user understanding formed by the enterprise in the process.

In addition, startups also need to answer a more realistic question: whether the advantages can be replicated on a large scale.

Bryan reminded that some vertical AI companies can easily become a project-based service company in the process of "deepening the industry".

Some AI companies can earn good revenue by deeply serving a small number of customers in the early stage. However, if each additional customer requires redesigning the process and investing in more engineers and service personnel, then as revenue increases, costs will also increase. In this case, even if you have deep industry know-how, you may just be expanding a project-based business rather than forming a truly scalable AI product. For example, the AI medical company needs to prove that it can copy the model of the leading hospital to more hospitals, and the GEO company needs to prove that growth depends on products, not manpower.

Therefore, when major manufacturers fully enter the application layer and the underlying models are constantly upgraded, for the top players in visual generation, model capabilities are still the ticket, but for more AI application companies, these three test questions may be more decisive than "what model is used" to determine how far they can ultimately go.

Respect for the original work, if you are qualified, you should be appreciated.

According to the real-name requirement of the Cybersecurity Law, please bind your mobile phone number before posting a comment.

Express