小厂，被困在 AI 时代, HSK5 reading

In the first quarter of 2025, many Chinese internet companies released their earnings reports one after another. Unlike large companies, these smaller platforms rarely mention AI revenue separately in their financial reports. Zhihu's quarterly revenue was about 650 million yuan, with Non-GAAP net profit of about 17.16 million yuan—a sign of recovery—but its income still came mainly from traditional businesses such as paid content and marketing services. iQiyi's revenue was 6.23 billion yuan, with nearly 70% from memberships and nearly 20% from advertising. Of Weibo's 2.9 billion yuan in revenue, advertising accounted for more than 2.5 billion. Zhiwen Group's revenue of nearly 2.4 billion yuan came almost entirely from memberships and live-stream tips. Platforms like Huya and Douyu have made money in much the same way for years. In other words, these smaller companies' AI revenue has not yet clearly shown up in their earnings reports.

But that does not mean they have stayed out of AI. After ChatGPT went viral at the end of 2022, these smaller companies actually moved quickly. Zhihu worked with ModelBest to launch "Zhihaitu AI" and later rolled out "Zhihu Direct Answer." iQiyi set up an AIGC center, developed its own large model, and launched "Taodou" for users. Weibo introduced AI comment bots and AI search products. Huya and Douyu applied AI to live streaming, esports, and other scenarios. Objectively speaking, these efforts were not worthless. AI has genuinely helped them improve efficiency in areas such as content moderation, internal productivity, and creative assistance.

The problem is that most of these results stay limited to "working inside their own apps," and struggle to become new growth engines. Compared with big tech firms, smaller companies' self-developed models are not strong enough and rarely appear on industry rankings. They also lack the funds for large-scale promotion. Users always chase the strongest and newest AI products, while smaller firms can neither offer standout offerings nor afford paid user-acquisition battles. As a result, many AI projects follow the same arc: a loud launch at the start, then a gradual fade once the hype dies down. AI features look more like add-ons to old businesses than independent new tracks.

Deeper difficulties come from a huge gap in capital and resources. China's top-tier AI companies now invest hundreds of billions of yuan each year. Alibaba plans to invest 380 billion yuan in AI infrastructure, and ByteDance is rumored to be putting in 500 billion yuan. These giants have steady cash flow from their core businesses and support from capital markets, so they can invest for the long term even when returns are not obvious in the short term. By contrast, smaller companies usually earn only a few billion yuan a year. If they tried to "all in AI" at the same intensity as big firms, that money might last only a few days. Long-term pressure on their stock prices has also weakened their ability to raise funds. The foundation of the AI industry—large models, AI apps, coding tools, agent frameworks, and the like—increasingly looks like a game only well-funded giants can stay in long term.

Mid-size companies seem to have found two workable paths. One is to concentrate resources on vertical AI and focus on a niche scenario—for example, Kuaishou's "Kling," which entered the top tier in AI video generation and was reportedly valued at $18 billion after being spun off. The other is to sell traffic to big companies and help AI products acquire new users. Bilibili's ad revenue has grown quickly in recent quarters, with AI-related ads as a major source. Smaller companies have also tried to copy these paths: platforms like Zhihu and Weibo promote AI content in advertising, while iQiyi and Huya explore vertical AI products. But smaller firms have limited traffic and struggle to build a highly active, high-conversion AI content ecosystem. Their vertical AI efforts also hit bottlenecks early. They can imitate business forms, but they cannot replicate the computing power, talent, engineering capability, and capital allocation behind them.

Perhaps an even harder problem than lack of money is that the AI industry evolves too fast for smaller companies to find a stable place. In 2023, everyone was competing for foundation models. Then came AI search, AI chat apps, and other applications. Today, competition has extended to agents, AI coding tools, and AI hardware. At every stage, big firms quickly seize key positions, while smaller ones take part but rarely catch the window of opportunity. After 2024, AI at big firms and AI at small firms are almost no longer the same kind of competition: the former can do full-stack R&D, iterate quickly, and even invest in AI chips; the latter rely more on open-source models and build customized features within their core businesses, with noticeably weaker exploration.

An even bigger change is that AI is reshaping the "entry point." In the mobile internet era, phones and app stores distributed apps but did not make every product themselves, leaving room for smaller companies in niche scenarios. Many of their highlight moments happened in that period. But in the AI era, giants are building "super entry points": they are traffic distributors, service providers, and even demand creators. More and more users are used to asking questions, looking things up, listening to music, and watching videos in Doubao. These high-frequency needs, once spread across small apps, are being absorbed more efficiently by AI. ByteDance launched a Doubao phone, WeChat is exploring agent-style integration, and OpenAI and Microsoft are building agent platforms. If users can complete most tasks just by talking to AI, why open a separate niche app? That is exactly what smaller companies fear most.

The article's conclusion is not optimistic: smaller companies are not blind to AI's value, nor have they done nothing, but under pressure from funding, users, technology, scenarios, and the pace of change, they keep falling further behind. Big and mid-size companies have offered partial "reference answers," but smaller ones struggle to copy them in any real way. To break out, they cannot treat AI as a decorative line in an earnings report—they need deeper, bolder change. Otherwise, as the AI era arrives faster, these smaller firms may not just be running slowly; they may have no clear place left at all.

In one sentence: well-funded, resource-rich big companies are pulling further ahead in the AI era, while many smaller internet firms started early but cannot afford to burn cash or keep up with the industry's pace, making it hard to turn AI into real business and revenue as their room to survive keeps shrinking.