文 | 吴怼怼
本文基于QuestMobile 2019—2026年历年报告。
2026年7月，一个指标悄悄翻了篇：抖音占中国Top 50应用总时长的19.4%，第一次超过了微信的18.8%。
这不是一次偶然的月度波动。把QuestMobile过去七年的报告串起来看，它是一条早就写好的曲线走到了终点——也是下一段故事的起点。
这篇文章想回答三个问题：中国人的手机时间是怎么涨上来的？它流向了谁？它的终点在哪里？
问题一：时间是怎么涨上来的？
2019年二季度，QuestMobile第一次记录到移动互联网用户单季净减少200万人。那一年，被很多人视为“人口红利”终结的元年。
但接下来的七年，时长并没有停下。月活从约11.4亿增长到12.82亿，只多了约13%；月人均使用时长却从128.2小时涨到189.3小时，接近五成。
换句话说，总时长的增量里，约四分之三来自“每个人用得更久”，而不是“更多人来用”。
曲线上唯一的一次明显回落，发生在2023年。那年12月，人均时长同比下降近12小时，约-7%。原因不难猜：线下生活恢复了。
这个小插曲很重要。它说明手机时长的天花板不在于生理极限，而在于“一天怎么在线上和线下之间分配”。
2024年起，时长重回增长；2025年5月以来，同比增速稳定在7%—9%。还有一个细节值得留意：2026年6月，人均使用次数同比只涨了3.1%，远慢于时长。每一次打开手机，停留得更久了——这正是视频化消费留下的指纹。
问题二：时间流向了谁？从“聊天”到“刷”
2019年，即时通讯还是吃掉最多时间的行业，占全网总时长的26.5%；短视频只有15.2%。三年后，位置对调：2022年短视频升到28.5%，即时通讯降到20.7%。交叉点，大约在2021年。
但短视频的份额并没有一路涨上去。2023年5月，它停在28%左右；到2024年12月，短视频和即时通讯的行业人均时长同比都是+3.2%，分别为62.9小时和40.2小时。这场大迁移，已经结束了。
新一轮迁移，正在更细的赛道里发生：
短剧在抢长视频的时间。2026年6月，短剧视频App月活突破4亿，同比+71.5%。5月份，短剧App人均每月看26.9小时，接近在线视频（14.7小时）的两倍。
小游戏在抢碎片时间。2026年6月，微信小游戏月活6.52亿，同比+21.2%。
工具型应用在失血。论坛贴吧、搜索引擎、图片美化，用户规模和人均时长同比双降——它们都是“找信息、做处理”的场景，恰好是AI最先替代的部分。
腾讯与字节的“时间换位”
2019年一季度，腾讯系占全网总时长的45.9%，接近一半；字节系只有10.3%。到2022年末，这组数字变成了33.6%对24.5%。
再到2026年7月，按Top 50应用口径，字节系五款App合计40.9%，腾讯系29.1%。
有意思的是，腾讯输掉的不是用户。2026年6月，腾讯、阿里、抖音集团的去重用户分别是12.81亿、12.65亿、12.59亿，几乎都覆盖了全网。微信DAU 9.31亿，仍明显高于抖音的7.14亿。
差别全在“深度”。2026年7月，抖音总时长同比+29%，背后是DAU增长19%、每人每天时长增长8%；微信只涨了3%，而且全部来自DAU增长。
视频号是最好的注脚。2022年6月，它的月活已经达到8.13亿，超过抖音；但日均使用时长只有35分钟，抖音是118分钟。
“渗透率，换不来时长。”
为什么每一轮都是字节赢？
回看这七年，每一次内容形态切换——短视频、免费小说、短剧——都制造了一条“没有终点的信息流”，而每一次的主要受益者都是字节系。
它在反复复制同一套模式：免费内容＋算法分发＋广告或激励。抖音、抖音极速版、番茄免费小说、红果短剧，本质上是同一台机器的不同型号。2026年7月，红果短剧DAU达到1.68亿，同比+107%，总时长同比+157%。
人群结构在背后托底。2026年6月，45岁以上用户占比39.5%，三、四线城市用户占41.5%。银发和下沉人群闲暇多、偏爱视频，“看视频领金币”的极速版又把这部分时间放大了。
不过也有例外：游戏。2026年5月，手游App人均每天玩111.2分钟，同比+6.9%；在百万月活以上的手游里，腾讯游戏的总时长占比超过七成。短视频崛起这么多年，并没有系统性地挤掉游戏时间——这也是腾讯时长份额在掉、利润却在涨的原因之一。
新变量：AI，是时间的吞噬者还是节省者？
AI正在同时做两件方向相反的事。
在工具侧，它替你办事，节省时间。AI原生App月活从2024年12月的1.2亿涨到2026年6月的4.99亿，同比+85.4%。但它的用法是“高频、短时”：2024年12月，人均每月打开49.6次，每次只有约2.7分钟，一个月加起来2.2小时，不到全网人均时长的1.3%。
在内容侧，它让短剧、漫剧的生产成本骤降，制造出更多“可刷”的东西。这一面，正在吞噬时间。
与此同时，聊天机器人的热度已经见顶：2026年7月，豆包DAU 1.68亿，环比持平，结束了自2025年8月以来的连续增长。大厂的竞争重心，正转向办公Agent。
我的判断
AI不会让总时长下降，而会改变时长的构成：被AI省下来的注意力，大概率流向更便宜、更无限的AI生成内容。真正被AI改写的，是“时长”这个指标本身——当Agent在后台替你比价、下单、写文档，平台的价值会越来越少地体现在你盯着屏幕的分钟数上。
问题三：终点在哪里？
先列一个最朴素的公式：
总时长 ＝ 月活规模 × 人均时长
月活这一项，已经没有故事了。12.82亿，覆盖全国约14亿人口的九成以上，每年只增加约1500万人。
人均这一项，还剩最后一段。189.3小时/月，折合每天6.3小时。按全年龄平均（包括老人和孩子），扣掉睡眠、工作学习中没法用手机的时间，我把现实上限定在每月210—225小时，也就是每天7—7.5小时。
按近一年7%—9%的增速推算：2027年中约203—206小时，2028年中约217—225小时——触顶。
换成总量：2026年6月，全网每月的总使用时长约2427亿小时；触顶时大约2700亿—2900亿小时。剩下的空间，只有10%—20%。
“触顶之后，任何一家多出来的一分钟，都得从别人那里抢。”
2023年那次回落，其实就是零和世界的一次预演。
以上数字为作者基于QuestMobile数据的推算，并非QuestMobile的结论。
三种可能的结局
下面三种情景并不互斥，后面的概率是我的主观权重，用来排序，而不是预测。
三个终局判断：
1、时长的终局，是天花板加零和。
2019年用户见顶后，行业靠人均时长又涨了近五成。这条路，大约在2028年前后走完。
2、谁拥有下一种内容形态的分发权，谁就拿走下一轮时长。
过去三轮——短视频、免费小说、短剧——都被字节吸收；视频号的8亿月活，没有换来同等的时长。腾讯的护城河在关系链、支付和游戏，而不在时长份额。当然，微信擅长慢工出细活，悬念交给时间。
3、时长与收入，会继续脱钩。
2026年二季度，腾讯营销服务收入同比增长22%，达到436亿元，同期它的时长份额仍在下滑；字节则在主动压低电商、生活服务的信息流加载率。终局的变量是“每分钟的变现效率”和“Agent入口”，而不是总分钟数。
写在最后
过去十年，互联网公司的估值逻辑里，有一条近乎公理的假设：用户数和时间，就是生意本身。
这条公理正在松动。时间的总量也快到顶了，时间的价格开始分化，而AI正在把一部分时间从屏幕上拿走。下一个十年，比的不是谁让你停留得更久，而是谁让你的每一分钟更值钱。
- 数据说明：本文数据来自QuestMobile历年公开报告。各报告取月不同，存在季节性；“月人均单日使用时长”与“月人均使用时长”口径不同，本文以后者为主线。终局推演与情景概率为作者推算，不构成投资建议。
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Man, Wu!
This is based on the QuestMobile 2019-2026 calendar year report.
In July 2026, one indicator quietly turned over: Douyin accounted for 19.4 per cent of the total time spent on Top 50 applications in China, and for the first time exceeded 18.8 per cent of WeChat.
This is not an accidental monthly fluctuations. Bringing together QuestMobile's reports for the past seven years, it is a long-written curve that has reached its end — and it is the beginning of the next story.
Three questions were asked in this article: How did Chinese cell phone time rise? Who's it headed to? Where's its end?
Question one: How did time rise?
In the second quarter of 2019, QuestMobile recorded for the first time a net decrease of 2 million mobile Internet users on a single-season basis. That year was seen by many as the end of the millennium of the demographic dividend.
But for the next seven years, it didn't stop. Monthly activity increased from approximately 1,140 million to 1,282 million, an increase of only about 13 per cent; the average monthly time spent increased from 128.2 hours to 189.3 hours, close to 50 per cent.
In other words, about three-quarters of the total increase comes from “everyone uses it longer” rather than “more people use it”.
The only clear fall on the curve occurred in 2023. In December of that year, the per capita length dropped by almost 12 hours per person per year, or -7 per cent. The reason is not hard to guess: life is back online.
This little episode is important. It shows that the time-long ceiling of a mobile phone is not a physical limit, but rather “how one day is distributed online and between lines”.
Since 2024, there has been a return to long-term growth; since May 2025, the increase has remained stable at 7-9 per cent. It is also worth noting that, in June 2026, the per capita use increased by only 3.1 per cent over the same period. Every time the phone was turned on, it stayed longer — that's the fingerprints left by video consumption.
Question two: Who does the time go to? From chat to brush?
In 2019, instant messaging was still the sector that ate the most time, accounting for 26.5 per cent of the total Internet time; short video was only 15.2 per cent. Three years later, the short video rises to 28.5 per cent in 2022 and instant communications drops to 20.7 per cent. Crosspoint, around 2021.
But short video's share did not go up all the way. In May 2023, it stopped at around 28 per cent; by December 2024, the average per capita length of short video and instant communications was +3.2 per cent, at 62.9 and 40.2 hours, respectively. This major migration is over.
A new round of migration is taking place in a more detailed track:
Short operas take long video time. In June 2026, the short play video App made over $400 million a month, compared to +71.5 per cent. In May, Amp per capita saw 26.9 hours per month, close to double the online video (14.7 hours).
The little game is robbing debris time. In June 2026, WeChat games lived 652 million per month, +21.2 per cent.
Tool-type application in blood loss. The forum posts, the search engine, the image beautification, the size of the user and the per capita time drop — they're all scenes of “getting information, doing processing”, which is exactly the first part of AI to replace.
Tencent and byte's "time change"
In the first quarter of 2019, Tencent accounted for 45.9 per cent of the total length of time of the network; byte accounted for only 10.3 per cent. By the end of 2022, the figure had changed to 33.6 per cent to 24.5 per cent.
By July 2026, according to the Top 50 caliber, byte was a total of five App 40.9% and Tencent 29.1%.
Interestingly, Tencent didn't lose the user. In June 2026, Tencent, Ali and Douyin groups had 1,281 million, 1,265 million and 1,259 million users, respectively, almost all of which covered the entire network. WeChatDAU 931 million, still significantly above 714 million Douyin.
The difference is in depth. In July 2026, the total length of Douyin was +29 per cent, followed by a 19 per cent DAU increase and an 8 per cent increase per person per day; WeChat increased only 3 per cent, all from DAU growth.
Video numbers are the best footnotes. In June 2022, its monthly activities had reached 813 million, more than Douyin; however, the daily average was only 35 minutes, and Douyin was 118 minutes.
"Infiltration rates, not for long."
Why is every byte win?
Looking back at these seven years, each content morphological transition — short video, free novels, short dramas — has created a “flow of information without endpoints”, each of whose main beneficiaries is byte.
It is replicating the same model: free content + algorithm distribution + advertising or motivation. Douyin, Douyin hypervelocity, tomato-free novels, red fruit dramas are essentially different models of the same machine. In July 2026, the red fruit short play DAU reached 168 million, +107 per cent for the same period and +157 per cent for the same period.
The crowd structure is in the back. In June 2026, 39.5 per cent of users were over 45 years of age and 41.5 per cent were urban users on three or four lines. This period was amplified by the high-speed version of the "See the video" gold medals, which was used in the free and popular video by the silver-haired and sinking population.
But there are exceptions: games. In May 2026, App per person played 111.2 minutes per day, compared to +6.9 per cent, while Tencent games accounted for more than 70 per cent of the total length of time on hand for over a million months. short video has risen for so many years without systematically squeezing out game time — and this is one of the reasons why Tencent's long share is falling and profits are rising.
New variable: AI, Time Eater or Saver?
AI is doing two different things at the same time.
On the tool side, it works for you, saves time. The monthly life of the original AI App increased from 120 million in December 2024 to 499 million in June 2026, an increase of +85.4 per cent. However, it is used as “high frequency, short hours”: in December 2024, 49.6 times per person per month was opened for approximately 2.7 minutes per month, or 2.2 hours per month, less than 1.3 per cent of the total network.
On the content side, it brought down the production costs of short dramas and comics, and created more “brushable” things. This side, it's eating up time.
At the same time, the heat of the chat robot has peaked: In July 2026, DoubaoDAU was $168 million, a flat ring, which ended its continuous growth since August 2025. The focus of competition in large factories is shifting to office Agent.
My judgment.
AI does not let the total length of time decline, but changes the long-term composition: the attention saved by AI probably goes to cheaper, more unlimited AI-generated content. The real change from AI is the "long" indicator itself -- when Agent is on the backstage to compete for you, to print and to write, the value of the platform becomes less and less in the minutes you stare at the screen.
Question three: Where is the end?
First, a simplest formula:
Total duration = monthly activity size x per capita duration
There's no more story. 12.82 billion, covering more than 90 per cent of the country ' s population of about 1.4 billion, is increasing by only about 15 million annually.
This is the last one left for everyone. 189.3 hours/month, equivalent to 6.3 hours per day. On an average basis at all ages (including the elderly and children), the time taken off from sleep, work or study without the use of a cell phone is limited to 210-225 hours per month, or 7.5 hours per day.
Based on an increase of 7 to 9 per cent in nearly a year: about 203 to 206 hours in 2027 and 217 to 225 hours in 2028 — the peak.
In June 2026, the total use of the network was approximately 24.27 billion hours per month, and the peak was about 270 billion to 290 billion hours. Only 10-20% of the remaining space.
"As soon as it hits the top, any family gets another minute from someone else."
That fall in 2023 was actually a rehearsal of the zero-sum world.
These figures are based on the author ' s extrapolation of QuestMobile data and are not the conclusions of QuestMobile.
Three possible endings.
The following three scenarios are not mutually exclusive, and the latter probability is my subjective weight and is used to rank rather than predict.
Three final judgements:
The end of the period is the ceiling plus zero.
After the peak of users in 2019, the industry increased by nearly 50 per cent on a per capita basis. This road is about 2028.
2. Whoever has the power to distribute the next form of content takes away the next round.
The last three rounds — short video, free novels, short plays — were absorbed by byte; 800 million months of live by video numbers were not exchanged for the same length of time. The moat of Tencent is in the chain of relationships, in payment and in play, rather than in long-term shares. Of course, WeChat is good at slow work and leaves the suspense to time.
3. Time and income continue to be delinked.
In the second quarter of 2026, Tencent's revenue from marketing services increased by 22 per cent over the same period, to 43.6 billion yuan, while its long-term share continued to decline over the same period; byte was actively pressing down the load rate of the electricity suppliers and the flow of information for services. The final variables are “realization efficiency per minute” and “Agent portal” rather than the total number of minutes.
At the end.
In the valuation logic of Internet companies over the past decade, there has been a nearly reasonable assumption that the number of users and time is the business itself.
This justice is relaxing. The total amount of time is at its peak, the price of time is starting to divide, and AI is taking some of the time off the screen. The next decade is worth more than who made you stay longer, and who made you worth more every minute.
- Data description: This data is from the QuestMobile public report for the calendar year. The reports vary from month to month, and there is seasonality; the term “one-day per month per person” is different from the caliber “one-month per person per person per month”, which is the main line in this paper. Finalization and scenario probabilities are extrapolated by authors and do not constitute investment recommendations.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.