9:1 36Kr | Doubao's Monthly Active Users Drop by 6.1 Million After Launching Subscription; Anthropic Calls for Global Slowdown in AI Development, Warns of 'Self-Improvement' Risks; Luo Yonghao Steps Down as Executive Director of Smartisan Software Company
Doubao’s loss of 6.1 million monthly active users served as a costly, open lesson for the entire Chinese AI industry. This wake-up call was loud enough to jolt every team eager to rush AI products into a paid model: In China, users’ faith in "free" remains unshakable. The timeline for commercialization is not dictated by the pace of technological iteration but by the speed of shifting user mindsets. Li Bangzhu’s remark that "China’s free AI era is far from over" was too polite—it should have bee
Analysis
Doubao’s loss of 6.1 million monthly active users served as a costly, open lesson for the entire Chinese AI industry. This wake-up call was loud enough to jolt every team eager to rush AI products into a paid model: In China, users’ faith in "free" remains unshakable. The timeline for commercialization is not dictated by the pace of technological iteration but by the speed of shifting user mindsets. Li Bangzhu’s remark that "China’s free AI era is far from over" was too polite—it should have been more direct: Free is not a phase; it is an ecosystem. Relying on its traffic empire, ByteDance still revealed in its AI commercialization push the inadequacy of its familiar "blitzscaling" mindset when confronting entirely new user relationships. Users today may cheer for the impressive performance of your free model, but tomorrow they will turn away without hesitation the moment a paywall goes up. Those 6.1 million lost users are not mere data—they are angry, clear-eyed consumers voting with their feet.
In stark and ironic contrast to Doubao’s "premature push," there is Anthropic’s blog post calling for a global slowdown in AI development. On the surface, it brims with lofty concern for humanity’s fate. But peel back the warm, public-relations-friendly rhetoric, and the core reveals a shrewd strategic positioning: run the fastest yourself, then turn around and shout to the chasers, "Danger—slow down!" This move carries a whiff of "a thief crying 'stop thief.'" While Anthropic flaunts its muscle by disclosing internal data on its models’ soaring capabilities, it warns of the risks of "self-improvement." What it may really be saying is: "Friends, the road ahead is too dangerous—why not let me use my lead to occupy it first?" In the business world, noble calls are often the most elegant competitive tactics. The so-called global coordination resembles more of a soft barrier designed to buy time for the leader and set hurdles for the followers. We must, of course, be mindful of risks, but we should also see clearly the deep-seated anxiety behind these calls—the fear of being easily overtaken by latecomers.
Turning back to the domestic scene, a Tencent executive’s claim that "most of our code this year is generated by AI" reads less as a technical fact than as a compelling efficiency narrative. In a precise technical context, the term "most" requires very specific definition: Does it mean AI assists in completing most lines of code? Or does it dominate the development of most functional modules? The image of engineers "spending more time on architecture design while delegating coding to AI" feels more like an idealized preview of future workflows than today’s widespread reality. However, the true value of this statement lies in what it signals: Major tech companies have officially upgraded their positioning of AI from a "novel tool" to "core productivity infrastructure." At the same time, the explosive growth in Tencent Cloud’s daily token consumption—surpassing 5 trillion—offers a more solid, unembellished signal. It indicates that in the B2B sector, large model applications have moved beyond the proof-of-concept stage and entered a practical phase of large-scale data processing and task execution. The growth logic here diverges subtly but crucially from Doubao’s setbacks in the B2C market: Enterprise clients pay for deterministic efficiency gains, their willingness and ability far exceeding that of individual users paying for uncertain experiential upgrades.
Thus, the current AI landscape is fragmented. On one side, the consumer market is locked in a fierce struggle within the mire of free models, where paywalls act as revealing mirrors exposing the illusion of user loyalty. On the other side, the enterprise services market surges ahead in real-world scenarios, with token consumption emerging as the new yardstick for growth. Meanwhile, across the ocean, the frontrunner constructs its moat through a narrative of "risk." This game is no longer a mere technological sprint. Doubao’s tuition tells us that respecting the stages of market education matters more than chasing the speed of technological monetization. Anthropic’s call reminds us that in competition, the ability to define issues and set rules is sometimes just as critical as technological leadership itself. As for those grand, beautiful tales of "AI-generated code," we might do better to listen to the steady, tangible pulse of tokens on cloud platforms—where the truest and most powerful undercurrents of AI commercialization in this stage truly churn.
Disclaimer: The above content is generated by AI and is for reference only.