Is It Difficult to Develop Large Models Domestically? ByteDance, Baidu, and Unicorns Compete Overseas, Who Will Profit First from Over 70 AI Products?
Produced by | Sohu Technology
Author | Liang Changjun
“Every day from 9 AM to 3 PM, I have meetings with foreign teams for remote development, internal testing, or bug fixing.” Entrepreneur Li Bojie, who is about to launch an AI product overseas, has been exceptionally busy recently.
This is a C-end AI evaluation product that helps users recommend different AI models or products. He hopes to make this product the “TikTok of the large model era.”
More than a year ago, when Li Bojie decided to leave Huawei to start his own business, he aimed to enter the overseas market. At that time, domestic large models were still in the stage of fierce technical competition, but now more and more companies are choosing the same direction as him.
Whether it’s ByteDance, Baidu, Alibaba, or large model unicorns like MiniMax, Dark Side of the Moon, and Zero One Everything, they are all accelerating their overseas expansion to tap into the global market.
Many companies are quietly making a fortune. Sohu Technology has learned that several overseas products have achieved rapid growth in users and revenue, and some have even started to become profitable. Some products have seen a surge in traffic with AI support, and are expected to achieve profits of 70 to 80 million yuan this year.
In the mobile internet era, Chinese companies went overseas and created TikTok. Now everyone is trying to create the TikTok of the AI era. This is a huge opportunity, but also full of challenges.
Big Companies and Startups Compete in AI Overseas, ByteDance is the Most Aggressive
While domestic large models are busy competing in technology, finding applications, and cutting prices, many companies are eyeing overseas markets.
Not long ago, the domestic large model rising star Dark Side of the Moon was reported to be entering the overseas market, developing AI role-playing chat applications Ohai and music video generator Noisee for the US. The company denied having plans to develop and release overseas products at that time.
However, Sohu Technology noticed that Ohai and Noisee have already launched web versions overseas, with the main entity registered in Sunnyvale, California, under the company Tranquillitatis. Its sole director is Yuxin Wu, which sounds the same as one of the co-founders of Dark Side of the Moon, Wu Yuxin.
Before Dark Side of the Moon, another large model unicorn MiniMax launched the AI role-playing chat application Talkie in the US as early as June last year.
Zero One Everything and Zhipu AI also regard globalization and internationalization as important strategies. The former focuses on productivity and social tracks, while the latter serves international companies coming to China based on B-end services and collaborates with domestic companies to go overseas.
In the context of domestic big companies leading the competition in applications and prices, AI startups can avoid the range of big companies by going overseas. As Kai-Fu Lee bluntly stated, “If it’s so competitive domestically, I’d rather lose everything than let you win, then go to foreign markets.”
However, big companies are also frequently taking action overseas. In just over half a year, ByteDance has launched or upgraded 7 C-end products overseas, involving AI chat, AI image generation, AI video editing, AI education, etc.
Baidu is also testing the waters in overseas markets, launching three native AI applications overseas since last year, including the AI chat application SynClub, AI social application WiseAI, and AI camera Meira.
In addition to launching AI image and video processing products such as Duiyou, Pic Copilot, and Luban AI for overseas e-commerce business, Alibaba has also built five new data centers overseas and launched the international version of the large model service platform Bailian.
It can be said that ByteDance is currently the most aggressive in AI overseas, replicating the “APP factory” approach to the AI era.
“ByteDance’s matrix/factory approach is very correct, just like how TikTok grew back then,” said Li Bangzhu, founder of the AI product list that has long focused on overseas markets, to Sohu Technology. However, this approach is not feasible for startups.
Zhang Yiming once emphasized that ByteDance cannot miss AGI. After Liang Rubo criticized the team for not researching large models in a timely manner at the beginning of the year, ByteDance quickly took action in this wave of AI overseas expansion.
In the view of Hong Yong, an expert at the China Digital and Real Integration 50 Forum think tank, ByteDance is trying to build a global ecosystem based on AI technology, reflecting its ambition for diversification and globalization in the AI era.
For big companies, globalization has always been an unavoidable development proposition, and it is the same in the AI era.
“Going overseas is an important step for these companies to expand their market share and achieve a global layout, and it is also an opportunity to showcase China’s AI technology strength and innovation capabilities to the world,” said Wang Peng, a researcher at the Beijing Academy of Social Sciences.
AI Tool Products Dominate, Interesting Products Attract More Users
The number of overseas AI products is growing rapidly. According to Feifan Research data, as of early June this year, there are more than 70 AI products with a scale of over 10,000 monthly visits overseas, doubling in less than half a year.
These products mainly focus on two directions: one is tool products that can be used to improve efficiency, which are commonly referred to as useful products in the industry; the other is social products that focus on emotional companionship and role-playing, which are interesting products.
Currently, overseas AI is mainly dominated by useful products, mainly targeting image and video generation or editing fields, especially the image field, which has attracted many companies to focus on layout.
For example, the traditional image editing tool Fotor has been rejuvenated after launching AI functions, with monthly visits reaching tens of millions. Image enhancement tools like Cutout Pro and Yodayo also have monthly active users exceeding ten million.
In the video field, ByteDance’s overseas version of Jianying Capcut launched AI editing and AI video generation functions this year. The product’s web version had over 40 million visits in June (according to Feifan Research), making it ByteDance’s most popular overseas application after TikTok.
Li Bangzhu believes that video generation products are still maturing technically, and Sora is a futures product that is still unusable after so long. But video generation will soon have some very interesting applications, such as the restoration of old photos, which corresponds to a strong psychological need.
In addition, social overseas products targeting the American large model unicorn Character AI have also received a lot of attention and have gained higher user stickiness and retention.
The AI companion application CrushOn launched by a domestic startup team overseas had 18.77 million visits in June this year (according to AI Product List data), making it the only AI social product with over ten million monthly active users overseas.
“If you only look at one data point for a product, look at the usage time, which represents user stickiness. Another data indicator reflected is retention,” Li Bangzhu explained. Role-playing and other social products perform better in terms of user time and retention compared to tool products.
According to Similarweb data, CrushOn’s single visit duration can reach 20 minutes, once ranking first among global AI products. In comparison, Character AI reaches 14 minutes, ChatGPT maintains seven to eight minutes, and Midjourney is around 10 minutes.
Li Bangzhu believes that general-purpose tools face retention issues, and ChatGPT has not seen much growth after reaching three hundred million monthly active users. “This is not an AI capability issue, not a technical issue, nor a team issue. The essence is that ChatGPT does not have a scenario.”
“AI social products, the value of satisfying users’ emotions is underestimated, and specific groups are very willing to use such products and are prone to impulsive consumption,” he mentioned. AI social products are sometimes serious about role-playing, such as with games and anime, but borderline cases are also common.
Therefore, the domestic AI social field faces greater regulatory pressure. MimiMax’s Glow, launched domestically last year, was reported and removed, prompting a shift overseas. Subsequently, AI chat apps like Zuo Ge AI, He, and AI Lisi were also removed one after another. Earlier, Tencent’s Weiban also announced the cessation of services in mainland China, focusing its operations overseas.
“Domestic regulatory pressure is greater, and once something goes wrong, it’s over. Going directly overseas is the first choice,” Li Bangzhu said, which is also why domestic companies tend to go overseas for AI social.
Li Bojie believes that in the early stages when large model technology is not very mature and accuracy is not very high, making interesting products is easier to land, with relatively low barriers.
But this track is also becoming increasingly crowded. When Li Bojie first started his business, he chose a social entertainment project similar to Stanford AI Town, aiming to create interesting Agents, but later he changed his mind.
“After finishing, I felt that the homogenization of Agents was too serious, making it difficult to stand out.” Li Bojie hopes to do something with a moat, “Whether it’s interesting or useful products, there is a recommendation problem, which will be an independent track.”
Overseas Can Make Money Faster, Some Products Will Achieve 70 to 80 Million Annual Profits
Currently, domestic companies’ AI overseas expansion mainly focuses on the C-end. Li Bojie said that doing C-end domestically is not as good as overseas because domestic big companies have greater advantages, and super apps like WeChat and Alipay can easily become AI entry points.
But he believes that domestic companies have stronger product development capabilities than overseas, and overseas business models are easier to run through. Traditional internet products, such as website development, have one-time development costs, and operational costs are relatively low. However, large models not only have high training costs but also very high inference costs.
“This means that doing it domestically requires a high enough customer unit price to recover costs and make a profit. Overseas, users have stronger payment capabilities and willingness, making it easier to recover costs.” This is also why Li Bojie focuses on overseas entrepreneurship.
Sheng Jingyuan, general manager of MiniMax’s overseas business, who has worked in Silicon Valley for many years, also mentioned that Chinese ToC entrepreneurship has a great advantage, and Americans cannot come up with a heavy user operation business model. “Chinese companies have a certain methodology for making global ToC products, and there is a possibility of overtaking on the application side.”
At the same time, she believes that interesting products are easier to run through business models. “The main direction of the domestic hundred-model battle is usefulness, but the business model is not very clear.”
Li Bojie said that interesting products can make money faster in the short term, but useful products may have more value in the long term. “The market for interesting products is not small, but in the long run, the imagination space is not as large as useful products.”
Currently, many overseas AI products are quietly making a fortune.
Sohu Technology has learned that MimiMax’s global C-end users have reached 40 million, with more than half overseas, contributing more than half of the revenue. Public information states that Talkie has accumulated 14 million downloads in one year, with total revenue exceeding $800,000, and achieving positive ROI.
Industry insiders also revealed to Sohu Technology that Fotor is expected to have a profit of 70 to 80 million yuan this year. This means that the company behind the product, Hengtong Technology, is expected to exceed 100 million yuan in revenue this year.
Kai-Fu Lee previously publicly stated that Zero One Everything’s overseas applications have nearly ten million total users, and a single ToC product relying on subscriptions will achieve 100 million yuan in revenue this year, with a product ROI close to 1, initially getting rid of the money-burning customer acquisition model.
Some AI overseas companies have achieved explosive revenue through the B-end, such as HeyGen, which became popular by using AI to make American singer Taylor Swift speak Chinese. The company’s founder, Xu Zhuo, revealed that by the end of last year, the company’s annual revenue had reached $18 million, with a 50% month-on-month growth for nine consecutive months.
Regarding the prospects of AI overseas, Li Bangzhu believes that the B-end is easier to make money at this stage, while the C-end has more potential. Li Bojie said that ToB emphasizes relationships and resources more, while ToC is easier to scale up.
Hong Yong pointed out that although the higher willingness to pay in the United States provides a favorable business model environment for AI going abroad, it does not guarantee success. Factors such as product differentiation, user experience, and brand building must also be considered.
Currently, domestic AI going abroad mainly relies on a user subscription model. “Overseas companies need to carefully design their business models to balance the freemium model with direct charging,” Hong Yong stated.
Overseas competition is equally fierce, with regulation being a greater potential challenge
Going abroad is undoubtedly a huge opportunity, but the competition in the overseas AI market is equally intense. Startups in the United States are more active, financing is easier, and related AI applications are emerging one after another.
In the June statistics of this year’s global AI WEB TOP100 products by Feifan Research, nearly 80 are overseas products. Domestic products going abroad are difficult to compete in terms of both quantity and traffic.
Opportunities for entrepreneurs themselves are also limited. Li Bojie believes that large models are currently not capable of creating purely new phenomenal applications, and most opportunities lie with large companies. “The opportunity for entrepreneurship is in areas that large companies have not seen, or have seen but cannot or are unwilling to do.”
At the same time, market preferences are also changing. Due to financing difficulties, Character AI recently reported considering selling to Google and Meta. This company, valued at $5 billion, had revenue exceeding $15 million last year and nearly 280 million monthly active users, but has not secured financing in the past year.
Earlier, Inflection, also in the AI social track, sold to Microsoft. These star companies in the AI social field giving up independent development to some extent show the survival difficulties of large model startups, and the logic dominated by user traffic still faces scrutiny in the AI era.
This may also be an invisible pressure for AI companies going abroad. Originally, going abroad might be a new story that startups could tell investors, but if they cannot maintain user growth and convert traffic, the story of going abroad will be difficult to sustain.
From a product perspective, Li Bojie believes that current AI social is very limited to a one-on-one form, only suitable for people or scenarios with inner emptiness wanting to kill time, and may eventually lead to borderline cases.
“I think the future should be a society composed of multiple people and multiple agents interacting, which is very crucial.” Li Bojie mentioned that some products are moving in this direction, but overall progress is relatively slow.
The greater potential challenge of going abroad lies in regulation. Hong Yong stated that in the increasingly stringent regulatory environment, domestic companies going abroad will face challenges such as compliance, data security, cross-cultural adaptability, and brand trust building.
Wang Peng also mentioned that as AI applications deepen in the social field, user privacy and data security issues will become increasingly prominent. At the same time, Chinese AI companies need to strengthen their awareness of intellectual property protection and take corresponding measures to prevent potential risks.
“Currently, smaller startups may not face TikTok-like regulatory pressure. Domestic companies going abroad generally establish overseas companies to ensure there are no legal compliance issues, but it may not be easy once they grow larger.” Therefore, Li Bojie chose to register a company in the United States at the start of his venture.
For companies going abroad to do B-end business, the challenges may be greater. “Customers may have concerns if they see the Chinese entity behind it, so now some companies are moving their entities overseas,” Li Bangzhu mentioned.
Sohu Technology noticed that HeyGen adopted a similar approach. Its previous domestic entity was Shiyun Technology, which later canceled its domestic entity and fully shifted to the overseas market.
For companies going abroad, localization is key. It may not be as simple as establishing a company overseas but also requires design in data, operations, human resources, and other aspects to meet overseas policies, regulations, and regulatory requirements.
“In the AI era, China is fully capable of producing global products similar to TikTok.” Many industry researchers are optimistic about the future prospects of Chinese AI going abroad.
However, this path is destined not to be smooth, and companies going abroad need to be prepared to deal with risks.