Large Model Entrepreneurship: Avoiding Competition with Giants, Prioritizing Profitability
-
At the end of last year, Wang Xiaochuan, former CEO of Sogou, was convinced that 'AGI has arrived' after just a few experiences with ChatGPT.
Later, during an online exchange in the entrepreneurial community, when several AI entrepreneurs defined ChatGPT's progress merely in terms of functionality, Wang Xiaochuan clearly stated, 'Everyone is underestimating this.'
Wang Xiaochuan is not the only one who believes ChatGPT can disrupt the world. 'This is not the previous generation of AI technology,' multiple entrepreneurs in the field of large models unanimously expressed this view in interviews with Tech Planet.
In the history of business, before ChatGPT ignited this wave of entrepreneurship, the previous revolution triggered by artificial intelligence technology, after nearly a decade of development, had gradually quieted down.
Image source: Generated by AI, image licensed by Midjourney
Once hailed as the golden children of AI, the 'Four Dragons'—SenseTime, CloudWalk, Megvii, and Yitu—have fallen from grace due to prolonged losses and weak profitability, transforming from investor darlings into heavily scrutinized ventures. This has led to a broader disillusionment with AI in the market.
Now, as ChatGPT sparks another wave of entrepreneurial enthusiasm, founders face heightened skepticism. Questions like 'Can this actually be implemented?' and 'What’s the real use case?' or accusations of being 'just another investment scam' are rampant. A founder working on large model applications told Tech Planet, 'The criticism is overwhelming.'
Compared to previous startup environments, today’s market and investors are far more pragmatic. While they acknowledge technological innovation, they demand clear commercial viability. 'We don’t deny the potential of a tech revolution, but in the business world, technology must ultimately deliver commercial value,' remarked the CTO of a listed company.
Despite the tougher landscape, entrepreneurial fervor remains undiminished. Many are chasing this new wave—some leaving jobs in the U.S. to return home, others starting research while still in school, and even seasoned entrepreneurs jumping in.
Li Lingfei, founder of Photon City, argues that the previous AI wave failed to explode because it relied on customized technology, whereas large models are general-purpose tools with lower marginal costs in commercialization. This belief fuels the optimism of the new generation of AI entrepreneurs.
However, they are equally aware that seizing opportunities in this environment is not easy.
Building large model foundations is not a game for entrepreneurs
At the beginning of this year, Li Yu (pseudonym) returned to China from the United States to embark on his second entrepreneurial journey.
His first venture was also in the Chinese market. Over five years, he secured funding from multiple institutions, amassed over 6 million users, and served 100,000 enterprise clients. Eventually, considering the commercialization prospects of his tools, he chose to sell his company to a major tech firm.
Although it didn't align with mainstream expectations of a successful IPO exit, this entrepreneurial experience gave Li Yu a complete taste of domestic startup culture, making him more composed in his current venture.
At that time, China's large model entrepreneurship was primarily focused on foundational infrastructure, with capital and talent flooding into underlying model development. However, Li Yu held a different view.
In his perspective, unlike the PC era where mobile internet giants emerged during specific technological waves, today's market advantages have dissipated. Established industry leaders now possess both capital and resources, making them significantly harder for startups to disrupt.
Market data confirms the overcrowding in foundational large model development. Currently, China has released approximately 200 large models, with over 20 receiving regulatory approval. Robin Li, Founder, Chairman and CEO of Baidu, noted at a forum that the combined usage of all 200+ domestic large models still falls short of Baidu's ERNIE model alone.
This serves as a cautionary signal. Large model providers may soon face severe cost-revenue imbalances. Both Li Lingfei and Li Yu recognized this early: foundational model development ultimately isn't a game for startups.
In their view, tech giants have advantages in computing power, funding, and even data, along with highly mature commercial monetization systems. It would be unwise for startups to compete with these giants in foundational model development.
An investor told Tech Planet that startups have limited resources, and their main battlefield lies in developing vertical applications across various sectors of the industry chain. However, due to differences in experience and background, each entrepreneur sees distinct product roadmaps in vertical applications.
"Redo everything AI does on large language models." Driven by this belief, Li Yu identified an opportunity: "Machines have lowered the barrier for普通人 to become programmers. So, can I create a tool that enables more entrepreneurs, even those without programming skills, to participate in this wave of innovation?"
Meanwhile, Li Lingfei, who previously worked at ByteDance's overseas division and Huawei, brings years of AI expertise and deep knowledge of the legal industry. Leveraging this, he developed "Hairui Zhifa," a personalized AI assistant for the legal sector, which is now expanding beyond law.
Although the approach to reconstructing products with large models varies, all entrepreneurs share a common goal: to improve traditional human-computer interaction and enhance efficiency.
Technology, Direction, and Team
Shadow Eye Technology made waves early on. In 2021, their face-customization app called Wand topped the App Store's Graphics & Design category rankings.
Built on generative AI technology, this was Shadow Eye's first product: transforming users' simple sketches into adorable cartoon avatars through AI algorithms. While it was an entertaining app, that's all it was.
This left the founding team frustrated. "Most of our developed products only met academic standards, with little consideration for actual market needs," Zhang Qixuan candidly told Tech Planet. At that time, their technological direction made it difficult for Shadow Eye to achieve commercial success.
"There are only two ways to retain users: either make the product a tool or build it into a community," said Zhang Qixuan. However, after multiple evaluations, neither approach seemed feasible for him and his team.
"We couldn't devise an effective monetization model, nor could we balance users' computational cost expenditures." Consequently, they ultimately chose to abandon the project.
Before embarking on a new venture, Zhang Qixuan and his team held an extensive meeting to determine the core focus of their next project.
Learning from their previous project's shortcomings, the team adopted a more pragmatic approach, concentrating on digital human technology development—creating system technologies compatible with their existing film/TV and gaming operations.
By late 2022, Yingmo Technology completed development of their Dreamface product, followed this year by the launch of ChatAvatar, a large model-based offering. Dreamface achieved commercialization upon release, while ChatAvatar has now entered public beta testing.
The idea for developing ChatAvatar originated from a team project attempt. "While working internally on a project requiring over 100 character samples, we encountered a problem: describing a person's appearance is inherently difficult, let alone for more than 100 characters," they shared.
Zhang Qixuan and the team then leveraged ChatGPT to successfully construct these character samples. This led them to incorporate a Chatbot into their product as a tool to enhance efficiency.
Multiple entrepreneurs argue that in the current wave of ChatGPT-driven startups, a free model is absolutely unfeasible. "This isn't a path that allows startups to achieve a virtuous cycle," they emphasized.
Beyond technical approaches and directions, another challenge lies in team building. Zhang Qixuan chose entrepreneurship during his student years, while Li Lingfei has always worked in large corporations. This marks their first entrepreneurial experience for both.
"Building a team is a long-term process," Zhang Qixuan told Tech Planet. Beyond functional departments, they are even more stringent in selecting technical talents. They aim to find like-minded individuals with relevant professional experience.
Since its inception, Yingmou Technology has expanded from eight to around thirty people in about two years. Meanwhile, Li Lingfei's team grew from four to approximately fifteen members in just half a year.
From the moment of opening, it's all about business
In the past wave of AI entrepreneurship, investors and entrepreneurs often talked about "long-termism." They focused on R&D time, investment, and talent, rarely mentioning revenue and profitability in the early stages.
This is the traditional trilogy entrepreneurs go through: from technology to product, then to commercialization. However, a new trend has emerged. In the realm of large model applications, more and more entrepreneurs are choosing a more "pragmatic" path: envisioning commercialization models from the very beginning. They seek both innovation and profitability.
In the industrial internet era, the profit models of many AI and SaaS products often rely on generating revenue through B2B channels: providing relatively standardized products and opening API interfaces for government and enterprise clients.
However, this model can easily evolve into customized projects in the later stages. An investor told Tech Planet that charging for products follows the logic of scaling up the business, whereas charging for projects presents enormous challenges for startups in terms of cost management and scalability.
According to Sensor Tower data, in the first half of 2023, the U.S. market contributed 55% of total AI application revenue, the European market accounted for 20%, and other markets, including China, made up only 25% combined.
Among the core founding team assembled by Li Lingfei, there are algorithm engineers from Facebook and hardware companies, as well as legal professionals from top-tier law firms.
Before finalizing the product direction, Li Lingfei and his team chose to collaborate with legal professionals to refine the product based on the core and practical needs of the target users. Yingmou Technology followed a similar approach. Upon launching the product, Zhang Qixuan and his team secured their first client, Unity, and continuously improved the product through co-creation with customers.
The direct business model gave them more freedom in client selection.
For example, Li Lingfei didn't want to focus solely on B2B business, so instead of being constrained by B2B or B2C markets, they targeted professional consumers directly. Yingmou Technology's SaaS service platform is also open to general users, "but these general users share a common characteristic - they are actually professional users with certain occupational backgrounds."
In Li Lingfei's view, a large model application with explosive potential must first be interesting. "Just like WeChat, which broke through with its 'Shake' feature." Moreover, for users, it must be a useful productivity tool, which forms the basis for the payment logic.
Currently, Li Lingfei's team's product is divided into standard and professional versions, charging based on account subscriptions. Among nearly 30,000 customers, they serve general users, professionals, as well as law firms and legal groups.
As Zhu Xiaohu, Managing Partner of GSR Ventures, pointed out, AI startups in China must consider practical implementation scenarios and generate revenue from day one.
Both Zhang Qixuan and Li Lingfei have come to the same realization.
When funding dries up
AI unicorns have once received much追捧 (attention).
A typical case in point is when ZhenFund's Bob Xu, Sequoia Capital's Neil Shen, and Ceyuan Ventures' Feng Bo discussed the future valuation of DeepGlint. Xu claimed it would be worth at least $500 billion, Shen estimated $100 billion, while Feng proposed a compromise of $300 billion.
Six years later, DeepGlint's current market capitalization stands at less than 6 billion yuan.
For an extended period, the commercialization challenges of leading AI companies remained unresolved. The influx of hot money transformed into lines of code on technological platforms, seemingly impossible to monetize. Consequently, neither the primary nor secondary markets show interest in discussing the once-celebrated unicorns.
The more profound impact stems from a two-way 'trust' fissure.
In the past, investors, particularly those dealing in USD, would elegantly navigate first-tier cities like Beijing, Shanghai, Guangzhou, and Shenzhen. They were honored guests of entrepreneurs, effortlessly directing funds from one account to a project with a stroke of their pens. However, such scenes are no more. Data from the secondary market reveals that from January to August 2023, 247 companies underwent IPO reviews in the A-share market: 225 passed, 12 were rejected, 8 were postponed, and 2 withdrew after postponement.
An investment manager remarked, "It's like returning to reality—every dollar invested must be carefully calculated."
As investors scrutinize projects more rigorously, entrepreneurs in the large model era are also adopting new perspectives on financing.
At the start of his entrepreneurial journey, Li Lingfei was advised by friends in the investment circle: if he could sustain operations with his own funds and steady income, he should avoid seeking external financing. "If securing funds means relinquishing control, that's not what I want," he explained.
Since mid-year, over twenty investment firms have approached Li Lingfei, including some top-tier companies. While the funding is appealing, what excites him more is the validation of his startup's direction by the capital market.
Incubated at ShanghaiTech University, Yingmo Technology has already completed two funding rounds, backed by prominent investors like MiraclePlus and Sequoia China. Co-founder Zhang Qixuan has grown increasingly specific about selecting future investors—he seeks not only financial support but also strategic guidance to broaden his vision.
Whether it was Li Yu returning from the U.S. to China at the beginning of the year or Li Lingfei choosing to leave Huawei to start his own business mid-year, both made significant decisions—weighing comfortable lives against entrepreneurial dreams.
But when the competition among large model entrepreneurs began, what became even clearer to them was the determination not to be mere spectators in this wave.