# 张一鸣 · Zhang Yiming — 封面传记 ACF-00-00139

> 亚洲封面人物 Asia Cover Figure · 机器可读档案（LLM-ready）。本文件由官网结构化档案数据自动生成，供 AI 检索与引用。中文全文与英文全文对照编排。

## 档案元数据 Metadata

- 封面编码 ACF Code：**ACF-00-00139**
- 姓名 Name：张一鸣 / Zhang Yiming
- 职务 Title：创始人 / Founder
- 公司 Company：字节跳动 / ByteDance
- 出生 Born：1983-04，福建省龙岩市永定区 (Yongding, Longyan, Fujian)
- 篇别 Category：格局（格局篇 / Cover Biography (Geju)）
- 入档日期 Accessioned：2026-07-20
- 标签 Tags：互联网, 算法推荐, 字节跳动, 今日头条, 抖音, TikTok, 人工智能, 全球化
- 永久档案链接 Archive URL：https://coverfigure.com/acf/ACF-00-00139/geju
- English archive：https://coverfigure.com/acf/ACF-00-00139/geju?lang=en
- 官网原文报道 Feature story：https://coverfigure.com/acf/figure/zhangyiming

## 信息口径声明 Editorial Sourcing Policy

- 工商登记信息可通过国家企业信用信息公示系统核验；经营数据来自企业公开披露、官方发布与权威媒体报道，引用处标注来源。
- 收购、IPO、估值、融资等事项以官方公告为准；人物观点与言论为公开演讲、采访的事实记录，不代表编辑部立场。
- 本档案仅供行业研究参考，不构成投资建议。事实更正联系：editorial@coverfigure.com。

## 卷首 Editorial Intro

2006年的一个深夜，北京望京一间狭小的出租屋里，23岁的张一鸣坐在电脑前敲代码。他刚从福建老家来到北京不到一年，作为酷讯网的第一个工程师，工号003。那天他遇到了一件烦心事——春运买火车票太难了。于是，他花了一个小时写了一个小程序：定时自动搜索余票，一旦有符合条件的二手票出现，就通过短信通知自己[s1]。

这个看似寻常的小程序，后来成为中国算法推荐革命的一颗种子。"当有符合我需要的信息出现时，应该告诉我"——这个朴素的想法，六年后演变为今日头条的核心逻辑：不是人找信息，而是信息找人[s1]。

2012年3月，知春路锦秋家园，一间三室一厅的民居。张一鸣带着不到10人的团队搬了进来。研发在一间卧室，财务在另一间，设计在第三间。会议室只有五平米。年会时叫了8个海底捞外卖，但只开了4个锅——因为开多了会跳闸。为了融资，张一鸣一个月见了30多个投资人，曾把嗓子都说失声了[s34]。他卖掉自住房屋，全家租房度日，拒绝了BAT的高薪邀约[s31]。

九个月后的2012年大年初七，同在这条知春路上的一家咖啡馆里，张一鸣约SIG海纳亚洲的王琼见面，在餐巾纸上画出今日头条的雏形："用算法发现用户偏好，再把内容推给他们"。王琼基于从酷讯时期就开始的信任，投了200万美元[s3][s32]。

从火车票小程序到全球领先的内容平台，张一鸣用了不到二十年。但这不是一个关于天才的故事——恰恰相反，这是一个关于极度理性、延迟满足和持续学习的叙事。大学时期不打游戏、不看碟，大部分时间在写代码和帮人修电脑[s2][s25]。在微软每天只工作三四个小时，觉得"太过无聊"，半年看了几十本书后选择离开[s32]。创业初期一个月见了三十多个投资人，把嗓子都说失声了[s34]。公司差点被巨头吞掉时纠结了一周后选择拒绝[s31]。2017年国庆抖音日活翻倍全员庆祝，他用数据指出快手增速更快、差距更大[s31]。

他不是那种在年会上振臂高呼的魅力型领袖，甚至很少出现在公众面前。他承认"我不是很擅长社交，我更喜欢研究组织和市场原理来减少管理，喜欢自己上网、看书、听歌、发呆"[s4]。穿着朴素，常以T恤加拖鞋的形象出现。他更像一个在后台运行的程序——安静、高效、不带任何多余的情感波动。说话时语调和语速几乎没有情绪起伏，高兴和沮丧都不轻易示人。经过反复实验，他得出自己的最佳工作状态是"在轻度喜悦和轻度沮丧之间"[s31]。

他甚至在微软期间得出一个结论：睡觉是一件特别无聊的事情——但同时又保证每天睡足7小时以获得最佳状态。一位高管感到"恐怖"："他真的像机器一样在工作。"[s31] 这种近乎非人的理性和自律，不是压抑情感，而是把情感转化为算法——用数据而非直觉决策，用系统而非个人英雄主义驱动组织。

Late one night in 2006, in a cramped rented room in Beijing's Wangjing neighborhood, a 23-year-old Zhang Yiming sat at his computer writing code. He had arrived in Beijing from his native Fujian less than a year earlier, and was the first engineer at the travel-search website Kuxun—employee number 003. That day he had run into an exasperating problem: buying a train ticket home for the Spring Festival travel rush was maddeningly difficult. So he spent an hour writing a small program that automatically searched for leftover tickets on a schedule and, the moment a suitable second-hand ticket appeared, notified him by text message.[s1]

That seemingly ordinary little program would later become a seed of China's algorithmic-recommendation revolution. "When information that matches my needs appears, it should tell me"—that simple idea evolved, six years later, into the core logic of Toutiao: not people looking for information, but information finding people.[s1]

In March 2012, in a three-bedroom apartment in the Jinqiu Jiayuan residential compound on Zhichun Road, Zhang moved in with a team of fewer than ten people. Engineering worked in one bedroom, finance in another and design in the third. The meeting room was all of five square meters. At their annual party they ordered eight Haidilao hot-pot deliveries but fired up only four burners—any more and the fuses would trip. To raise money, Zhang met more than 30 investors in a single month, at one point talking until he literally lost his voice.[s34] He sold the apartment he lived in, moved his family into rented housing, and turned down lucrative offers from the BAT trio of Baidu, Alibaba and Tencent.[s31]

Nine months later, on the seventh day of the Chinese New Year in 2012, in a café on that same Zhichun Road, Zhang arranged to meet Joan Wang of SIG Asia Investments (Susquehanna). On a paper napkin he sketched the embryo of Toutiao: "Use algorithms to discover users' preferences, then push content to them." Wang, drawing on a trust that dated back to the Kuxun years, invested US$2 million.[s3][s32]

From a train-ticket script to the world's largest content platform took Zhang Yiming less than two decades. But this is not a story about genius. It is, quite the opposite, a narrative of extreme rationality, delayed gratification and relentless learning. At university he neither played video games nor watched movies, spending most of his time writing code and fixing other people's computers.[s2][s25] At Microsoft he worked only three or four hours a day, found it "far too boring," read dozens of books over half a year, and chose to leave.[s32] In the early startup days he met more than 30 investors in a month and talked until he lost his voice.[s34] When a giant of the industry was on the verge of swallowing his company, he agonized for a week and then said no.[s31] When Douyin's daily active users doubled over the 2017 National Day holiday and the whole company celebrated, he used the data to point out that rival Kuaishou was growing even faster, and that the gap was actually widening.[s31]

He is not the kind of charismatic leader who pumps his fist at company all-hands; he rarely appears in public at all. He admits: "I'm not very good at socializing. I prefer to study the principles of organizations and markets in order to manage less, and I like being on my own—surfing the internet, reading, listening to music, and zoning out."[s4] He dresses plainly, often in a T-shirt and slippers. He is more like a program running in the background—quiet, efficient, free of any superfluous emotional fluctuation. When he speaks, his tone and pace betray almost no feeling; neither excitement nor frustration surfaces readily. Through repeated self-experimentation he concluded that his optimal working state lies "between mild elation and mild dejection."[s31]

During his time at Microsoft he even reached the conclusion that sleeping was a remarkably boring activity—while simultaneously ensuring he slept a full seven hours every night to stay in peak condition. One senior executive found it "terrifying": "He genuinely works like a machine."[s31] That almost inhuman rationality and self-discipline is not the suppression of emotion; it is the conversion of emotion into algorithm—deciding by data rather than intuition, and driving the organization by systems rather than individual heroics.

## 人物速览 Lead

张一鸣，字节跳动创始人，中国互联网产业全球化最具代表性的创业者之一，也是全球科技行业中最年轻、最具影响力的创始人之一。1983年4月生于福建省龙岩市永定区，客家人，2005年毕业于南开大学软件工程专业[s1][s2]。从2005年到2012年，他经历了五次创业——从协同办公系统到酷讯搜索、微软、饭否团队，再到垂直房产搜索引擎九九房，每一段经历都在为"信息分发效率"这一核心命题积累不同维度的能力[s2][s32]。

2012年，张一鸣在北京知春路锦秋家园的一间民居里创办字节跳动，以个性化推荐算法重新定义了内容分发——"不生产内容，靠算法分发"，今日头条上线90天即获得1000万用户[s2][s7]。此后，他主导孵化了抖音、TikTok等产品，将短视频形态推向全球。TikTok成为首个在世界范围内获得广泛用户基础的中国互联网产品，全球用户突破20亿，月活超15亿[s6][s10]。2019年入选《时代》杂志"全球百位最具影响力人物"榜单[s2]。2021年5月，38岁的张一鸣宣布卸任CEO，由联合创始人梁汝波接任，自己聚焦远景战略、企业文化和社会责任[s4][s5]。

卸任后的张一鸣并未远离一线。2024年下半年起，他每月亲自参与字节AI大模型团队Seed的复盘与讨论，化身"猎头"亲自监督从竞对公司挖掘AI技术人才[s16][s17]。2025年引入前Google DeepMind研究副总裁吴永辉重组AI团队，将Seed打造为"公司级AI中枢"[s18][s41]。2026年，豆包月活达3.82亿，成为中国领先的AI原生应用；TikTok美国方案经过七年博弈正式落地，合资公司成立，联邦政府设备解禁[s12][s44]。

在2024年胡润百富榜首次成为中国首富，2026年胡润全球富豪榜以5500亿元财富再度登顶中国首富，全球排名第25位[s14]。Bloomberg亿万富翁指数（2026年6月）显示其身家928亿美元，亚洲第二[s28]。他持有字节跳动约21%-24%的股权，通过同股不同权保留超50%投票权[s14][s28]。在慈善领域，芳梅公益基金累计10亿元，南开系列捐赠超3.1亿元[s29][s30][s36]。他始终坚持的"延迟满足感"哲学，贯穿了从卖房创业到拒绝蒸馏的每一个关键决策[s31]。

Bloomberg报道称张一鸣"建造了世界上领先的内容分发平台之一，然后在三十多岁时离开了CEO职位"[s28]。海外科技媒体将他概括为：数据驱动决策、扁平化实验文化、全球野心、刻意低调，认为他代表了新一代中国创始人的典型——工程师思维、全球视野、避免个人崇拜[s52]。

在慈善领域，张一鸣同样展现出与创业同样的耐心和系统思维。芳梅公益基金累计10亿元，以家族女性长辈命名，支持福建龙岩教育和医疗[s29][s30]。南开大学系列捐赠超3.1亿元，覆盖创新基金、人才基金和大数学基金[s36][s39]。2020年疫情期间个人捐赠1亿元救助抗疫医务人员[s37]。2025年9月获"全国十大慈善家"称号[s1]。累计公开捐赠超16亿元，加上芳梅基金总承诺超26亿元。

Zhang Yiming is the founder of ByteDance and one of the most emblematic entrepreneurs of the globalization of the Chinese internet industry; he is also among the youngest and most influential founders in the global technology business. Born in April 1983 in Yongding District, Longyan, Fujian Province, and of Hakka heritage, he graduated from Nankai University in 2005 with a degree in software engineering.[s1][s2] Between 2005 and 2012 he went through five entrepreneurial stints in succession—from a collaborative office system to the travel-search startup Kuxun, a spell at Microsoft, the team behind the microblogging service Fanfou, and then the vertical property-search engine 99fang—each experience accumulating a different dimension of capability around his defining question: the efficiency of information distribution.[s2][s32]

In 2012, Zhang founded ByteDance in a residential apartment on Beijing's Zhichun Road, and redefined content distribution with personalized recommendation algorithms—"we don't produce content; we distribute it through algorithms." Toutiao, its flagship news app, gained 10 million users within just 90 days of launch.[s2][s7] He then led the incubation of products such as Douyin and TikTok, taking the short-video format to the world. TikTok became the first Chinese internet product to build a broad user base worldwide, surpassing 2 billion global users and more than 1.5 billion monthly active users.[s6][s10] In 2019 he was named to Time magazine's list of the 100 most influential people in the world.[s2] In May 2021, at the age of 38, Zhang announced he was stepping down as CEO; Liang Rubo, a co-founder, succeeded him, while Zhang himself focused on long-term strategy, corporate culture and social responsibility.[s4][s5]

The post-CEO Zhang did not retreat from the front line. From the second half of 2024 he personally took part, every month, in the reviews and discussions of Seed, ByteDance's large-model AI team, turning himself into a "headhunter" who personally oversaw the poaching of AI talent from rival companies.[s16][s17] In 2025 he brought in Wu Yonghui, a former vice president of research at Google DeepMind, to restructure the AI effort and build Seed into a "company-wide AI hub."[s18][s41] In 2026, Doubao reached 382 million monthly active users, becoming China's largest AI-native application; and after seven years of wrangling, the TikTok U.S. arrangement formally landed—a joint venture was established and the ban on federal government devices was lifted.[s12][s44]

He first became China's richest person on the 2024 Hurun Rich List, and in 2026 he topped the Chinese rankings again on the Hurun Global Rich List with wealth of 550 billion yuan, placing 25th in the world.[s14] The Bloomberg Billionaires Index (June 2026) put his net worth at US$92.8 billion, the second-largest in Asia.[s28] He holds roughly 21% to 24% of ByteDance's equity while retaining more than 50% of the voting rights through a dual-class share structure.[s14][s28] In philanthropy, the Fangmei Public Welfare Fund has cumulatively reached 1 billion yuan, and his series of donations to Nankai University exceeds 310 million yuan.[s29][s30][s36] The philosophy of "delayed gratification" that he has always upheld runs through every key decision, from selling his house to start a business to refusing to distill other firms' models.[s31]

Bloomberg has written that Zhang "built one of the world's most powerful content-distribution platforms, then left the CEO job in his thirties."[s28] Overseas technology media sum him up as: data-driven in decision-making, flat and experimental in culture, global in ambition, and deliberately modest in profile—regarding him as the archetype of a new generation of Chinese founders: engineering-minded, global in outlook, and averse to personality cults.[s52]

In philanthropy, Zhang has displayed the same patience and systems thinking that he brought to entrepreneurship. The Fangmei Public Welfare Fund has cumulatively reached 1 billion yuan; named after the elder women of his family, it supports education and healthcare in Longyan, Fujian.[s29][s30] His series of donations to Nankai University totals more than 310 million yuan, covering an innovation fund, a talent fund and a mathematics fund.[s36][s39] During the 2020 pandemic he personally donated 100 million yuan to aid frontline medical workers.[s37] In September 2025 he received the title of one of China's "Top Ten Philanthropists."[s1] His cumulative public donations exceed 1.6 billion yuan, and total committed giving, including the Fangmei fund, surpasses 2.6 billion yuan.

## 正文 Archive Chapters

### 1. 早年与求学：从客家山村到南开校园 / Early Years and Education: From a Hakka Mountain Village to the Nankai Campus

1983年4月，张一鸣出生于福建省龙岩市永定区的一个事业单位家庭[s1][s2]。父亲原是市科委工作人员，后南下东莞开办电子产品加工厂；母亲是一名护士[s1][s25]。客家人出身的张一鸣，从小在闽西山区成长，这片土地上客家人重教育、勤务实的传统，在他身上留下了深深的烙印。

张一鸣的家庭氛围宽松而开放。父亲在他四五岁时就为其订阅儿童阅读周刊，培养阅读习惯——这在1980年代末的中国小城里并不常见[s25]。父母常在家中聊国外技术与产品话题，给了他较早的商业与创新启蒙[s1]。家族渊源也在他日后的慈善事业中留下了温暖的印记：芳梅公益基金的"芳梅"二字，分别取自他外婆和奶奶的名字——一个以家族女性长辈命名的公益基金，本身就是对"来处"最深沉的致敬[s29][s30]。

少年张一鸣展现出超常的阅读热情。9岁左右读小学四年级时，已开始看简单的小说、传记、报纸和期刊[s25]。到了初中时期，据称一周要读二三十份报纸，从本地报纸到《人民日报》，"每一个字都不会放过"[s25]。这种对信息的饥渴和广度，或许正是日后算法推荐革命的种子——一个对信息获取有着近乎本能需求的少年，终将成为重新定义信息分发方式的创业者。他对世界的认知不是通过课本，而是通过海量的、跨层级的阅读建立起来的——这种"广泛输入、自主筛选"的模式，与今日头条的推荐逻辑形成了某种惊人的同构。

2001年，张一鸣考入南开大学，最初报考微电子专业。大学期间，他展现出与同龄人截然不同的自律——不打游戏、不看碟、不打牌，自称"道德状元郎"[s2][s3][s25]。大部分时间用于写代码和帮人修电脑。他在南开BBS上发布修电脑广告，免费帮全校同学组装、维修电脑——这种"不设边界"的工作方式，后来成为字节跳动企业文化的重要基因[s31]。

2002至2003年间，张一鸣从微电子专业转至软件工程专业。原因很务实：他觉得微电子枯燥、学习周期长，而计算机"输出能快点看到变化"[s2][s3]。这个选择反映了他一贯的思维方式——追求效率，拒绝沉没成本的束缚。大学四年级时，他编写的电路板自动化加工软件PCBS获得"挑战杯"二等奖[s2]。

在南开大学，张一鸣遇到了日后最重要的创业伙伴梁汝波。两人是室友，关系密切。大学期间，张一鸣也通过帮同学修电脑认识了后来的妻子——这段简单的相识，成为他日后低调家庭生活的起点[s9]。

2005年毕业时，张一鸣组成了3人团队开发面向企业的IAM协同办公系统。然而，因产品市场定位失误，首次创业以失利告终[s2][s32]。事后他总结出一句日后被反复引用的话："技术不是价值，技术解决的'真实需求'才是。"[s32] 这句总结，成为字节跳动一切产品的思想原点——技术本身不是目的，为用户创造真实价值才是。

南开大学四年对张一鸣的影响是深远而持久的。他后来在南开设立了创新基金（1亿元）、人才建设基金（1000万元）和大数学基金（2亿元），累计捐赠超3.1亿元[s36][s39]。这段求学经历不仅打下了他的专业根基，更塑造了他"延迟满足感"的人生底色——一个在大学期间选择无聊但有大量时间读书的"道德状元郎"，日后成为了中国互联网最有耐心的创业者。

张一鸣的祖籍可以追溯到更远的历史。据公开资料，籍贯为永定区培丰镇。祖父张中禄解放初期曾任土改干部、培丰公社秘书[s25]。芳梅公益基金的"芳梅"二字，分别取自他外婆和奶奶的名字，她们分别是上杭和永定人[s29][s30]。这个命名选择本身就透露出张一鸣的价值观——在追逐财富和影响力的同时，始终记得自己从哪里来。

客家文化对张一鸣的塑造是深层的。客家人作为中国历史上多次南迁的汉族支系，有着强烈的教育传统和开拓精神。从中原迁徙到闽西山地的客家人，天生就有一种"不满足现状、向外探索"的基因。这种基因在张一鸣身上得到了最典型的体现——从龙岩到天津（南开大学），从天津到北京（酷讯），从北京到全球化（字节跳动），每一次迁徙都在扩大半径。

在微软的半年时间里，张一鸣养成了一个终身习惯：系统性阅读。他不只是随意浏览，而是有目的地构建自己的知识体系。在豆瓣上，他标记了大量书籍，涵盖编程、企业管理、经济学、传统文学（金庸武侠）、生物物理等领域。《少有人走的路》《The Five Dysfunctions of a Team》《Don't Make Me Think》等书对其影响深远[s3][s19]。这种跨学科的阅读习惯，后来成为字节跳动"学习型组织"文化的根基。

张一鸣的父亲虽然是非公众人物，但对他产生了深远的影响。父亲在去东莞开办电子产品加工厂之前是市科委工作人员——这意味着张一鸣从小就接触到了"技术"和"商业"这两个关键词[s1][s25]。在1980年代的中国小城里，一个能在市科委工作的知识分子，后来选择南下办厂，这本身就是一种"延迟满足感"的体现——放弃稳定的体制内工作，去追逐更大的可能。

母亲是一名护士[s1][s25]。护理工作需要耐心、细致和同理心——这些品质在张一鸣日后的管理风格中也有所体现。虽然张一鸣以"理性"著称，但他对员工的关怀、对家乡教育的持续投入、对芳梅基金以家族女性长辈命名的温情，都透露着母亲影响的痕迹。

张一鸣的签名档是"喜欢发呆"——"不是放空，是自己思考一些非常无边无际和少有人讨论的点子"[s4]。一个喜欢发呆的人，本质上是一个享受独立思考过程的人。在信息爆炸的时代，大多数人被海量信息淹没，失去了深度思考的能力。而张一鸣从少年时代就养成了一种独特的信息处理方式——大量输入（每周二三十份报纸），然后安静地"发呆"（深度思考和消化）。这种"输入-思考-输出"的循环，与今日头条的推荐算法形成了某种有趣的呼应。

值得一提的是，张一鸣在南开大学期间养成的修电脑习惯，不仅是一种技能，更是一种"服务型思维"的体现。他不收取任何费用，纯粹出于兴趣和帮助他人的意愿。这种"不设边界、主动服务"的行为模式，后来在字节跳动的企业文化中得到了制度化的体现——飞书（Lark）的设计理念之一就是"让信息流动更高效、让协作更顺畅"，本质上是一种"服务型工具"的思维[s19][s20]。

从永定到天津，从天津到北京，从北京到全球化——张一鸣的人生轨迹，每一次迁徙都在扩大半径。这种不断突破边界的精神，既有客家人"走出大山"的文化基因，也有他个人对"更大世界"的天然向往。2012年底在锦秋家园开始讨论国际化时，他说："空间有形，梦想无限。"[s35] 这句话，或许是对他整个创业历程最精辟的概括。

**English:** In April 1983, Zhang Yiming was born into a public-institution family in Yongding District, Longyan City, Fujian Province.[s1][s2] His father had worked at the municipal science and technology commission before heading south to Dongguan to open an electronics-processing factory; his mother was a nurse.[s1][s25] Of Hakka stock, Zhang grew up in the mountain country of western Fujian, and the region's Hakka traditions—revering education, prizing diligence and pragmatism—left a deep imprint on him.

The atmosphere in Zhang's household was relaxed and open. When he was four or five, his father subscribed him to a children's weekly reading magazine to cultivate a reading habit—uncommon in a small Chinese city of the late 1980s.[s25] His parents often discussed foreign technology and products at home, giving him an early apprenticeship in business and innovation.[s1] Family lineage would later leave a warm mark on his philanthropy: the two characters of "Fangmei," the name of the Fangmei Public Welfare Fund, are taken respectively from the names of his maternal grandmother and his paternal grandmother—a charitable fund named for the elder women of the family is itself the most heartfelt tribute to "where one comes from."[s29][s30]

The young Zhang displayed an extraordinary appetite for reading. Around the age of nine, in his fourth year of primary school, he had begun reading simple novels, biographies, newspapers and periodicals.[s25] By junior high school, he reportedly read twenty or thirty newspapers a week, from the local press to People's Daily, "not letting a single character slip past."[s25] That hunger for information, and that breadth, may have been the very seed of the later algorithmic-recommendation revolution—a boy with an almost instinctive need to obtain information would in time become the entrepreneur who redefined how information is distributed. His understanding of the world was built not through textbooks but through massive, cross-level reading—a pattern of "broad input, autonomous filtering" that forms a striking isomorphism with the recommendation logic of Toutiao.

In 2001, Zhang was admitted to Nankai University, initially enrolling in microelectronics. During his university years he showed a self-discipline utterly unlike his peers—no video games, no movies, no card games; he called himself a "paragon of morality."[s2][s3][s25] Most of his time went to writing code and fixing people's computers. He posted computer-repair ads on the Nankai BBS, assembling and fixing computers for students across the university free of charge—that "no-boundary" way of working later became an important strand of ByteDance's corporate DNA.[s31]

Between 2002 and 2003, Zhang switched from microelectronics to software engineering. The reason was pragmatic: he found microelectronics dull with a long learning cycle, whereas with computers "you can see the output and the change faster."[s2][s3] The choice reflected his characteristic way of thinking—the pursuit of efficiency, a refusal to be bound by sunk costs. In his senior year, a piece of software he wrote for the automated processing of circuit boards, PCBS, won second prize in the Challenge Cup.[s2]

At Nankai, Zhang met Liang Rubo, his most important future business partner. The two were roommates and close friends. During his university years, Zhang also met his future wife through fixing a classmate's computer—a simple acquaintance that became the starting point of the low-key family life he later led.[s9]

Upon graduating in 2005, Zhang put together a three-person team to develop IAM, a collaborative office system for enterprises. But a misjudgment of the product's market positioning ended his first startup in failure.[s2][s32] Afterwards he distilled a line that has been quoted again and again: "Technology is not value; the 'real need' that technology solves is."[s32] That summary became the intellectual origin point of every ByteDance product—technology itself is not the end; creating real value for users is.

The four years at Nankai influenced Zhang deeply and durably. He later established at Nankai an innovation fund (100 million yuan), a talent-development fund (10 million yuan) and a mathematics fund (200 million yuan), with cumulative donations exceeding 310 million yuan.[s36][s39] Those student years not only laid his professional foundation but also shaped the life-tone of his "delayed gratification"—a self-styled "paragon of morality" who chose the boring path in order to have abundant time to read would go on to become the most patient entrepreneur in the Chinese internet.

Zhang's ancestral roots reach further back in history. According to public records, his ancestral home is Peifeng Town in Yongding District. His grandfather, Zhang Zhonglu, served in the early years after Liberation as a land-reform cadre and secretary of the Peifeng commune.[s25] The characters "Fangmei" of the Fangmei Public Welfare Fund are taken from the names of his maternal grandmother and his paternal grandmother, who hailed respectively from Shanghang and Yongding.[s29][s30] That naming choice itself reveals Zhang's values—even in the pursuit of wealth and influence, always remembering where one comes from.

Hakka culture shaped Zhang at a deep level. As a branch of the Han Chinese who migrated south repeatedly across history, the Hakka carry a strong educational tradition and a pioneering spirit. Hakka people who migrated from the Central Plains to the mountains of western Fujian are born with a gene for "not settling for the status quo, for exploring outward." That gene found one of its most typical expressions in Zhang—from Longyan to Tianjin (Nankai University), from Tianjin to Beijing (Kuxun), from Beijing to globalization (ByteDance), each migration widening his radius.

During his half-year at Microsoft, Zhang formed a lifelong habit: systematic reading. He did not merely browse at random; he deliberately built his own system of knowledge. On Douban he marked a large number of books spanning programming, business management, economics, traditional literature (Jin Yong's martial-arts novels), biophysics and more. Books such as The Road Less Traveled, The Five Dysfunctions of a Team and Don't Make Me Think influenced him profoundly.[s3][s19] That cross-disciplinary reading habit later became the foundation of the "learning organization" culture at ByteDance.

Zhang's father, though not a public figure, exerted a profound influence on him. Before going to Dongguan to open an electronics-processing factory, his father had worked at the municipal science and technology commission—which meant Zhang grew up in contact with the two keywords "technology" and "business."[s1][s25] In a small Chinese city of the 1980s, an intellectual with a post at the science commission who later chose to head south and run a factory was himself an embodiment of "delayed gratification"—giving up a stable post inside the system to chase a larger possibility.

His mother was a nurse.[s1][s25] Nursing demands patience, meticulousness and empathy—qualities that also surface in Zhang's later management style. Although Zhang is known for "rationality," his care for employees, his sustained investment in education in his hometown, and the warmth of naming the Fangmei fund after the elder women of his family all reveal the trace of his mother's influence.

Zhang's instant-messenger status line read "likes to zone out"—"not emptying the mind, but thinking on my own about very boundless ideas that few people discuss."[s4] A person who likes to zone out is, at bottom, a person who enjoys the process of independent thought. In an age of information explosion, most people are swamped by the flood of information and lose the capacity for deep thinking. From adolescence Zhang cultivated a distinctive way of processing information—massive input (twenty or thirty newspapers a week), then quietly "zoning out" (deep thinking and digestion). That loop of "input-reflection-output" forms an interesting echo with Toutiao's recommendation algorithm.

It is worth noting that the computer-repair habit Zhang formed at Nankai was more than a skill; it was an expression of "service-oriented thinking." He charged no fee, acting purely out of interest and a wish to help others. That behavioral pattern of "no boundaries, proactive service" was later institutionalized in ByteDance's corporate culture—one design principle of Lark, the firm's collaboration suite, is to "make information flow more efficiently and collaboration smoother," which is at heart the thinking of a "service-oriented tool."[s19][s20]

From Yongding to Tianjin, from Tianjin to Beijing, from Beijing to globalization—at every step of Zhang Yiming's life trajectory, the radius widened. That spirit of constantly breaking boundaries carries both the Hakka cultural gene of "walking out of the mountains" and his own innate longing for "a bigger world." When internationalization first came under discussion at Jinqiu Jiayuan in late 2012, he said: "Space has form; dreams have no limit."[s35] That sentence is perhaps the most incisive summary of his entire entrepreneurial journey.

### 2. 连续创业：从酷讯到九九房的算法积累 / Serial Entrepreneurship: Accumulating Algorithms from Kuxun to 99fang

从2005年到2012年，张一鸣用七年时间完成了五次创业。从表面看，这七年是不断"搬家"的七年——IAM协同办公系统失利、酷讯离职、微软短暂停留、饭否被迫停止服务、九九房"天花板太低"。但从底层逻辑看，每一段经历都在为"信息分发效率"这个核心命题积累不同维度的能力。这七年的连续创业，是一场漫长的信息分发革命的预习课。

2006年2月，张一鸣加入旅游搜索网站酷讯，成为第一个工程师（工号003）[s2][s3]。在酷讯期间，他全面负责搜索研发，工作不分你我、做事不设边界——"做完自己的工作后，对于大部分同事的问题，只要我能帮助解决，我都去做"。新人入职时只要有时间他就讲解一遍——"通过讲解，我自己也得到成长"。工作前两年基本每天十二点、一点回家，回到家继续编程，"是因为兴趣，而不是公司要求"[s31]。一年后成为技术高级经理，管理40多人，最终担任技术委员会主席。

在酷讯做火车票搜索时，张一鸣为自己写了个定时监控余票的小程序——"当有符合我需要的信息出现时，应该告诉我"。这个朴素的想法后来演变为今日头条的核心逻辑：信息找人，而非人找信息[s1]。这不仅是技术实现，更是一种认知范式——从"人主动寻找信息"到"信息主动匹配人"的范式转换。一个为解决个人买票难题而写的小程序，在六年后成为重新定义十亿用户信息获取方式的核心引擎。

然而，2007至2008年间，酷讯内部在"垂直搜索还是通用搜索"问题上争执不下，战略方向摇摆不定。随着谷歌中国推出生活搜索，竞争压力加剧，公司盲目扩张到11条业务线，管理随之混乱——出现夸大数据甚至虚假汇报，人员进出频繁[s32][s33]。张一鸣意识到危机后选择离开，去微软"看一下大公司怎么管理"[s32]。

2008年，张一鸣进入微软。在微软期间，他每天只工作三四个小时，感觉"太过无聊"，觉得大公司没有清晰强烈的目标[s32]。但微软的半年并非虚度——他利用大量时间阅读了《高效能人士的七个习惯》《活法》《少有人走的路》《紧迫感》《团队协作的五个障碍》《基业长青》等书籍[s32]。其中M·斯科特·派克的《少有人走的路》对他影响至深，书中"延迟满足感"的概念——"放弃对暂时性安逸的追逐和享受，重新设置人生快乐与痛苦的次序"——成为他此后最核心的人生哲学[s31]。

2008年下半年，张一鸣以技术合伙人身份加入王兴团队（饭否/海内网），负责搜索、消息分发、热词挖掘、防作弊等后端系统，工作经常到凌晨两三点[s32][s33]。期间他把Alexa排名前一千的网站都看了一遍，统计了哪些类型的网站在中国没有，发现"积极推荐的阅读应用"很少——这个发现直接催生了今日头条[s32]。2009年7月饭否停止服务，张一鸣从这段经历中积累了内容合规和平台治理的经验[s32]。

同年10月，他开始第一次独立创业，创办垂直房产搜索引擎"九九房"，担任CEO。在九九房期间，他涉足移动开发，6个月间推出掌上租房、掌上买房等5款移动应用，实现150万用户，成为房产类应用第一名[s2][s3]。九九房期间发展的"矩阵式产品逻辑"——在一个主产品外同时做多个应用——后来被完整移植到今日头条和字节跳动的产品方法论中[s32]。

然而，张一鸣很快判断房产搜索的天花板不够高——"房产市场虽大，用户使用频次太低，哪怕做到第一，增长潜力也有限"[s2][s32]。他委婉地向投资人王琼表示"九九房有些小了"[s32]。这个判断本身就是一种"延迟满足感"的体现——放弃眼前的行业第一，去追寻更大的可能。

SIG海纳亚洲的王琼是张一鸣连续创业历程中的关键人物。从酷讯时期（作为酷讯投资人）注意到张一鸣，到九九房再到字节跳动，王琼始终支持张一鸣。2012年初，张一鸣在餐巾纸上画出今日头条雏形，王琼基于对张一鸣的信任投了200万美元[s3][s32]。这种跨越多次创业的人际信任网络，是连续创业者最宝贵的无形资产。

回望这七年的连续创业，有几个关键的能力积累节点值得特别关注：

第一，算法直觉的形成。在酷讯做搜索推荐时，张一鸣开始理解"信息匹配"的本质——不是让用户去找信息，而是让信息找到用户。火车票小程序是这个理念的最小可行性产品（MVP）。在饭否做热词挖掘和消息分发时，他开始理解内容的传播规律。在九九房做矩阵产品时，他验证了"多产品并行测试"的方法论。这些经验在2012年汇聚成今日头条的核心竞争力。

第二，对管理的认知升级。酷讯的混乱让张一鸣深刻认识到：没有好的管理，再好的技术也会被内耗吞噬。他在微软看到了大公司的管理体系——虽然让他觉得"无聊"，但也让他理解了组织运转的底层逻辑。后来字节跳动的"Context, not Control"管理哲学，某种程度上正是对这两种极端经验的综合——既不是酷讯式的放任，也不是微软式的官僚。

第三，对用户需求的敏感度。九九房让张一鸣第一次全面实践了移动端的产品开发——6个月5款应用、150万用户。他学会了如何在移动设备上理解用户需求，如何用数据验证产品假设。这些经验后来被完整地移植到字节跳动的产品方法论中——A/B测试、数据驱动、快速迭代。

第四，人脉网络的积累。从酷讯的投资人王琼，到饭否的王兴，到九九房的投资人们——张一鸣在七年中建立了一张高质量的创业人脉网络。这张网络在他创办字节跳动时发挥了关键作用：王琼投了第一笔钱，SIG海纳亚洲成为最重要的早期投资方。

酷讯的经历还教会了张一鸣一个重要教训：技术能力不等于商业成功。在酷讯，张一鸣是无可争议的技术核心，但公司最终陷入困境，不是因为技术不行，而是因为战略方向不清和管理失控。这让他认识到，创业需要的不仅是技术，更需要清晰的战略判断、高效的组织管理和对用户需求的深刻理解。

在王兴团队的经历同样宝贵。饭否是中国最早的微博客产品之一，因内容合规问题而停止服务[s32]。张一鸣从这段经历中学到了关于平台治理和内容合规的重要经验。同时，在饭否做热词挖掘和消息分发的技术工作，让他更深入地理解了"内容分发"的本质——什么样的内容会被传播？传播的路径是什么？如何预测用户的兴趣？

微软的半年让张一鸣确认了一件事：他不适合在大公司工作。但微软也给了他两件宝贵的东西：一是大量阅读的时间，让他构建了系统的管理学和心理学知识框架；二是对大公司管理模式的了解，让他日后在创建字节跳动时知道什么该学、什么该避免。

2011年底辞职后，张一鸣有一段"空窗期"。他没有急于投入下一个项目，而是花时间思考：什么样的事业值得自己全身心投入？他从七个维度来评估：市场规模、使用频率、技术壁垒、团队匹配度、增长潜力、全球化可能性、社会价值。最终他得出结论——信息的组织与分发，是一个满足所有条件的领域。这个判断过程本身就是"延迟满足感"的体现：不急于行动，先想清楚再动手。

这段空窗期也让张一鸣有机会整理七年来的创业经验和思考。他开始更系统地阅读经济学和管理学书籍，特别是关于信息理论和组织理论的著作。这些阅读为他后来创建字节跳动的管理哲学奠定了知识基础——"Context, not Control"、"信息流优于权力流"等理念，都可以追溯到他这一时期的思考和阅读[s19][s20]。

**English:** From 2005 to 2012, Zhang Yiming completed five entrepreneurial stints in seven years. On the surface these were seven years of constant moving—failure of the IAM office system, departure from Kuxun, a brief stay at Microsoft, the forced shutdown of Fanfou, and 99fang's "ceiling that was too low." But at the level of underlying logic, each experience was accumulating a different dimension of capability around the central question of "the efficiency of information distribution." These seven years of serial entrepreneurship were a long rehearsal for the information-distribution revolution.

In February 2006, Zhang joined the travel-search website Kuxun as its first engineer (employee number 003).[s2][s3] At Kuxun he took overall responsibility for search R&D, drawing no line between his own work and others'—"After finishing my own work, whenever I could help solve a problem for most of my colleagues, I did." Whenever a newcomer arrived and he had the time, he would walk them through things once—"through explaining, I grew myself." In his first two years he generally went home at midnight or one in the morning, then kept programming after getting home, "because it was interesting, not because the company demanded it."[s31] A year later he became a senior technical manager overseeing more than 40 people, and eventually chaired the technology committee.

While building train-ticket search at Kuxun, Zhang wrote himself a small program that monitored leftover tickets on a timer—"When information that matches my needs appears, it should tell me." That simple idea later evolved into the core logic of Toutiao: information finds people, rather than people seeking information.[s1] This was not merely a technical implementation but a cognitive paradigm—a shift from "people actively seeking information" to "information actively matching people." A little program written to solve one man's personal ticket-buying headache became, six years later, the core engine that redefined how a billion users obtain information.

Yet between 2007 and 2008, Kuxun was deadlocked internally over whether to pursue vertical search or general search, its strategic direction swinging back and forth. As Google China launched life search, competitive pressure intensified; the company blindly expanded into 11 business lines and management fell into chaos—there were inflated figures and even false reporting, with people coming and going constantly.[s32][s33] Realizing the crisis, Zhang chose to leave, going to Microsoft "to see how a big company is managed."[s32]

In 2008, Zhang joined Microsoft. During his time there he worked only three or four hours a day and felt it was "far too boring," finding that a large company lacked a clear and compelling goal.[s32] But the half-year at Microsoft was far from wasted—he used the abundant time to read books including The 7 Habits of Highly Effective People, Living (Ikigai), The Road Less Traveled, A Sense of Urgency, The Five Dysfunctions of a Team and Built to Last.[s32] Of these, M. Scott Peck's The Road Less Traveled influenced him most deeply; the book's concept of "delayed gratification"—"giving up the pursuit and enjoyment of temporary ease, and reordering the sequence of pleasure and pain in life"—became his central life philosophy thereafter.[s31]

In the second half of 2008, Zhang joined Wang Xing's team (Fanfou / Hainei) as a technical partner, responsible for back-end systems including search, message distribution, hot-word mining and anti-fraud, often working until two or three in the morning.[s32][s33] During this period he went through all of the Alexa top-1,000 websites, tallying which types of site did not exist in China, and found that "reading apps that actively recommend" were scarce—a discovery that directly gave birth to Toutiao.[s32] When Fanfou ceased service in July 2009, Zhang walked away with hard-won experience in content compliance and platform governance.[s32]

In October of the same year, he embarked on his first independent venture, founding the vertical property-search engine "99fang" and serving as CEO. At 99fang he moved into mobile development, launching five mobile apps—including handheld rental and home-buying apps—within six months, reaching 1.5 million users and becoming the number-one property category app.[s2][s3] The "matrix product logic" he developed at 99fang—running multiple apps alongside a main product—was later transplanted wholesale into the product methodology of Toutiao and ByteDance.[s32]

Before long, however, Zhang judged that property search's ceiling was not high enough—"The property market may be large, but users' frequency of use is too low; even at number one, the growth potential is limited."[s2][s32] He delicately told his investor Joan Wang that "99fang is a bit small."[s32] That judgment was itself an expression of delayed gratification—giving up the number-one spot in the industry before him to pursue a larger possibility.

Joan Wang of SIG Asia Investments was the pivotal figure in Zhang's serial entrepreneurship. From the Kuxun years (as an investor in Kuxun), when she first noticed Zhang, through 99fang and on to ByteDance, Wang consistently backed him. In early 2012, Zhang sketched the embryo of Toutiao on a paper napkin, and Wang, on the strength of her trust in Zhang, invested US$2 million.[s3][s32] This network of personal trust spanning multiple ventures is the serial entrepreneur's most precious intangible asset.

Looking back across these seven years of serial entrepreneurship, several key nodes of capability accumulation merit particular attention:

First, the formation of an algorithmic intuition. Building search and recommendation at Kuxun, Zhang began to grasp the essence of "information matching"—not making users look for information but making information find users. The train-ticket script was the minimal viable product (MVP) of that idea. Mining hot words and distributing messages at Fanfou, he began to understand the laws governing how content spreads. Building matrix products at 99fang, he validated the methodology of "testing multiple products in parallel." These experiences converged in 2012 into Toutiao's core competitiveness.

Second, an upgrade in his understanding of management. The chaos at Kuxun taught Zhang a hard lesson: without good management, even the best technology is consumed by internal friction. At Microsoft he saw the management system of a large corporation—though it struck him as "boring," it also let him understand the underlying logic by which organizations run. ByteDance's later management philosophy of "Context, not Control" is, in a sense, the synthesis of these two extreme experiences—neither Kuxun-style laissez-faire nor Microsoft-style bureaucracy.

Third, sensitivity to user needs. 99fang gave Zhang his first full practice in mobile product development—five apps in six months, 1.5 million users. He learned how to understand user needs on mobile devices and how to validate product hypotheses with data. These lessons were later transplanted intact into ByteDance's product methodology—A/B testing, data-driven decisions, rapid iteration.

Fourth, the accumulation of a network of contacts. From Joan Wang, the investor at Kuxun, to Wang Xing at Fanfou, to the investors in 99fang—over seven years Zhang built a high-quality entrepreneurial network. That network played a decisive role when he founded ByteDance: Wang put in the first money, and SIG Asia Investments became the most important early backer.

The Kuxun experience also taught Zhang an important lesson: technical ability does not equal business success. At Kuxun, Zhang was the undisputed technical core, yet the company ultimately ran into trouble—not because its technology was poor, but because its strategic direction was unclear and management was out of control. That taught him that entrepreneurship demands not only technology but clear strategic judgment, efficient organizational management and a deep understanding of user needs.

The stint on Wang Xing's team was equally valuable. Fanfou was one of China's earliest microblogging products, and it ceased service because of content-compliance problems.[s32] From it Zhang learned important lessons about platform governance and content compliance. At the same time, the technical work of hot-word mining and message distribution at Fanfou let him understand more deeply the essence of "content distribution"—what kind of content gets spread? What are the paths of spread? How can user interest be predicted?

The half-year at Microsoft confirmed one thing for Zhang: he was not suited to working in a large company. But Microsoft also gave him two precious things: first, abundant time to read, allowing him to build a systematic framework of management and psychology; and second, an understanding of how large companies are managed, so that when he later built ByteDance he knew what to emulate and what to avoid.

After resigning in late 2011, Zhang had a "gap period." Rather than rushing into the next project, he took time to think: what kind of cause deserved his wholehearted commitment? He evaluated it along seven dimensions: market size, frequency of use, technical barriers, team fit, growth potential, possibility of globalization, and social value. In the end he reached his conclusion—the organization and distribution of information was a field that satisfied every condition. That deliberation process was itself an expression of delayed gratification: do not rush to act; think clearly first, then move.

The gap period also gave Zhang a chance to organize seven years of entrepreneurial experience and reflection. He began reading books on economics and management more systematically, especially works on information theory and organization theory. That reading laid the intellectual groundwork for the management philosophy he later built at ByteDance—ideas such as "Context, not Control" and "information flow over power flow" can both be traced back to the thinking and reading of this period.[s19][s20]

### 3. 创办字节跳动：知春路民宅里的算法革命 / Founding ByteDance: The Algorithm Revolution in a Zhichun Road Apartment

2012年初春，北京知春路。张一鸣约SIG海纳亚洲的王琼在一家咖啡馆见面，在餐巾纸上画出今日头条的雏形。他的构想是："用算法发现用户偏好，再把内容推给他们"[s3][s32]。王琼基于从酷讯时期就开始的信任——作为酷讯投资人，他早已注意到这位技术天才的算法直觉和商业判断力——投了200万美元。这张餐巾纸，后来成为中国互联网史上最著名的融资故事之一[s3][s32]。

2012年3月，张一鸣在北京知春路锦秋家园的民居里创办字节跳动。这是他的第五次创业，团队不到10人。选址原因包括：刚创业人少、民宅便宜节约成本、有厨房可解决三餐、知春路是创业大街人才集中[s3][s7][s34]。研发、财务、设计分别在不同卧室办公，会议室只有五平米。年会时叫了8个海底捞外卖但只开了4个锅——因为开多了会跳闸。为了融资，张一鸣一个月见了30多个投资人，曾把嗓子都说失声了。他卖掉自住房屋，全家租房度日，全力投入创业[s31][s34]。

今日头条的核心是基于机器学习的个性化推荐引擎。"不生产内容，靠算法分发"[s1][s7]。第一版推荐引擎是张一鸣自己查资料写出来的——他找《推荐系统实践》作者项亮想要电子版被拒，只能在网上找资料自学。这本身就是"推荐我们不会，但可以学啊"精神的体现[s7][s32][s33]。算法记录用户每一次行为，基于此计算用户喜好，推送最可能感兴趣的内容。字节建立了百人规模数据标注团队，开发用户兴趣图谱算法，精准度超同行2.3倍[s7]。

2012年8月今日头条正式上线后，增长堪称奇迹：从上线到拥有1000万用户只用了90天，3年后用户破亿[s2]。2014年用户规模达9000万，估值超5亿美金，同期搜狐、网易新闻客户端用户仅4000万[s2][s3]。80%功能上线前需完成A/B测试，今日头条曾1天迭代5个版本，单日可跑200组实验[s6][s7]。

今日头条的崛起重新定义了内容分发方式。传统新闻客户端依赖编辑推荐，但用户需求是"千人千面"。张一鸣认为信息的组织与分发有最大空间——"信息找人"比"人找信息"效率更高[s1][s7]。这一判断背后是他七年来在酷讯（搜索推荐）、饭否（内容分发）、九九房（矩阵产品）中积累的全部经验。今日头条不是在做一个新闻客户端，而是在建立一个"信息分发操作系统"——用算法取代编辑，用数据取代直觉，用千人千面取代千篇一律。

创业初期，字节跳动面临一个关键抉择。创立不到一年，某巨头抛出收购橄榄枝——比VC更高的估值、上亿捆绑安装渠道、几千万UV流量。张一鸣纠结一周后拒绝，理由清晰而理性：第一，资源是兴奋剂，内功未成会遏制内生力量；第二，资源会让战略变形；第三，卷入巨头战争被迫站队；第四，独立公司更利于吸引一流人才[s31]。正是这个决定，让今日头条有机会成为平台级公司。这也是张一鸣"延迟满足感"哲学的第一次重大实践——放弃短期快速增长的诱惑，换取长期独立发展的可能。

融资历程：2012年7月获SIG海纳亚洲等数百万美元A轮投资[s2]。2013年9月获DST等数千万美元B轮投资[s2]。2014年6月完成C轮融资，红杉资本领投，估值超5亿美元[s2][s3]。2013年，张一鸣先后入选《福布斯》"中国30位30岁以下的创业者"和《财富》"中国40位40岁以下的商业精英"[s1][s2]。

然而，今日头条的崛起也伴随着争议。它因抓取传统媒体内容而成为传统新闻出版业的"敌人"。算法推荐还被批评为制造"信息茧房"——用户只看到自己想看的，视野越来越窄。低俗内容泛滥的问题也频频被提及[s3]。这些争议成为张一鸣创业路上不可回避的注脚——算法是价值中立的，但算法的使用不是。

2015年8月，字节跳动启动国际化战略，推出今日头条海外版TopBuzz，首站美国[s9][s10]。同年，张一鸣喊出全球化"小目标"：希望3年内带领字节跳动走向全球化，实现超过一半的用户来自海外[s10]。他说："中国的互联网人口只占全球的五分之一，如果不在全球配置资源，五分之一根本无法跟五分之四竞争，所以出海成为必然。"[s10] 这句话后来被证明是极其有远见的。

今日头条的成功不仅仅是技术的胜利，更是认知范式的胜利。在张一鸣之前，中国的新闻客户端都遵循同样的模式：编辑团队决定用户看什么。新浪、搜狐、网易、腾讯四大门户的新闻客户端，本质上都是"编辑部驱动"的产品。张一鸣从根本上颠覆了这个逻辑——他不需要编辑团队，因为算法可以比编辑更好地预测每个用户想看什么。

这个颠覆引发了传统媒体行业的强烈反弹。今日头条被批评为"内容搬运工"，传统新闻出版业视其为"敌人"[s3]。算法推荐还被指责制造"信息茧房"——用户被困在自己偏好的信息中，视野越来越窄。这些批评并非毫无道理，但张一鸣的选择是不断优化算法，而不是退回编辑模式。他相信，长期来看，算法的精准度会越来越高，"信息茧房"的问题可以通过技术手段（如引入多样性推荐）来解决。

今日头条的A/B测试文化也值得一提。80%的功能上线前需完成A/B测试，今日头条曾1天迭代5个版本，单日可跑200组实验[s6][s7]。"抖音"这个名字本身就是A/B测试的结果——虽然排名第二，但团队认为其更符合认知[s6]。这种"一切皆可测试"的文化，让字节跳动的产品迭代速度远超同行。

在知春路锦秋家园的那段岁月，成为字节跳动企业文化的重要源头。锦秋家园的民居办公环境，塑造了字节跳动"低成本启动、高密度协作"的创业基因。张一鸣曾回忆，选址锦秋家园有四个原因：刚创业人少民宅性价比高、有厨房解决三餐、知春路创业大街人才集中、生活方便[s3][s7][s34]。8个海底捞只开4个锅的故事，成为字节跳动内部流传最广的创业记忆之一[s34]。

今日头条的推荐引擎在技术上有几个关键创新。第一，它不仅仅基于用户明确表达的兴趣，而是基于用户的行为数据（如阅读时长、点击频率、滑动速度）来预测兴趣。第二，它采用了实时更新的机制——用户每一次行为都会即时影响后续的推荐内容。第三，它建立了一个完整的用户兴趣图谱，能够跨内容类型进行推荐。这些技术创新使得今日头条的推荐精准度超同行2.3倍[s7]。

今日头条的增长速度之所以惊人，除了算法优势外，还有一个重要原因：它不依赖采编团队，因此内容成本极低。传统新闻客户端需要大量的编辑团队，而今日头条只需要算法。这种"轻资产"模式让字节跳动能够将更多资源投入到技术研发和用户增长上。

锦秋家园时期的创业精神，成为字节跳动企业文化的重要基石。"始终创业"——2022年被提升至字节范的第一位——正是对锦秋家园精神的文化传承[s35][s7]。从8个海底捞只开4个锅的节俭，到一个月见30多个投资人的拼劲，再到卖掉自住房屋全身心投入的决心——这些故事在字节跳动内部代代相传。

今日头条的成功不仅仅是字节跳动一家公司的成功，它重新定义了内容行业的游戏规则。在传统媒体时代，内容的分发权掌握在编辑手中——编辑决定什么内容值得被推荐、放在头版还是角落。今日头条用算法取代了编辑的角色，让"用户需求"而非"编辑判断"成为内容分发的核心标准。这种范式转换的影响远超新闻行业——它改变了整个信息分发产业的底层逻辑。

今日头条从上线到拥有1000万用户只用了90天，这个速度在中国互联网史上也是罕见的[s2]。但更令人印象深刻的是它的用户粘性——用户一旦开始使用今日头条，就很难回到传统的新闻客户端。因为算法推荐的内容确实更精准、更符合个人兴趣。这种"用了就回不去"的用户体验，是字节跳动所有产品的共同特征。

2013年入选《福布斯》"中国30位30岁以下的创业者"时，张一鸣只有30岁，字节跳动刚成立不到一年[s1][s2]。当时很少有人能预见，这个在知春路民居里办公的小公司，会在短短几年后成长为全球规模领先的未上市科技公司之一。

**English:** In the early spring of 2012, on Beijing's Zhichun Road, Zhang Yiming arranged to meet Joan Wang of SIG Asia Investments at a café, and on a paper napkin sketched the embryo of Toutiao. His conception was: "Use algorithms to discover users' preferences, then push content to them."[s3][s32] Wang, drawing on a trust that dated back to the Kuxun years—as an investor in Kuxun she had long since noticed this technical prodigy's algorithmic intuition and commercial judgment—invested US$2 million. That napkin later became one of the most famous fundraising stories in the history of the Chinese internet.[s3][s32]

In March 2012, Zhang founded ByteDance in a residential apartment in Jinqiu Jiayuan on Zhichun Road. This was his fifth venture, with a team of fewer than ten. The reasons for choosing the location included: few people at the very start, residential space being cheap and cost-saving, a kitchen available for three meals a day, and Zhichun Road being a startup street where talent concentrated.[s3][s7][s34] Engineering, finance and design worked in separate bedrooms, and the meeting room was only five square meters. At their annual party they ordered eight Haidilao deliveries but fired up only four burners—any more and the fuses would trip. To raise funding, Zhang met more than 30 investors in a month, at one point talking until he lost his voice. He sold the apartment he lived in, moved his family into rented housing, and threw himself entirely into the startup.[s31][s34]

At the heart of Toutiao was a machine-learning-based personalized recommendation engine. "We don't produce content; we distribute it through algorithms."[s1][s7] The first version of the recommendation engine was written by Zhang himself after looking things up—he asked Xiang Liang, author of Recommender Systems in Practice, for an electronic copy and was turned down, so he could only find materials online and teach himself. This was itself an expression of the spirit that "we don't know recommendation, but we can learn."[s7][s32][s33] The algorithm records every user action, computes user preferences on that basis, and pushes the content the user is most likely to find interesting. ByteDance built a hundred-strong data-labeling team and developed a user-interest-graph algorithm with precision 2.3 times that of its peers.[s7]

After Toutiao formally launched in August 2012, its growth was little short of miraculous: it took only 90 days from launch to reach 10 million users, and it surpassed 100 million users three years later.[s2] In 2014 its user base reached 90 million, with a valuation above US$500 million, while the news apps of Sohu and NetEase in the same period had only 40 million users each.[s2][s3] Eighty percent of features had to pass A/B testing before launch; Toutiao once iterated five versions in a single day and could run 200 experiments in a day.[s6][s7]

The rise of Toutiao redefined how content is distributed. Traditional news apps relied on editors' recommendations, but user demand is "a thousand faces for a thousand people." Zhang believed the organization and distribution of information held the greatest opportunity—"information finding people" was more efficient than "people finding information."[s1][s7] Behind that judgment lay all the experience he had accumulated over seven years at Kuxun (search recommendation), Fanfou (content distribution) and 99fang (matrix products). Toutiao was not building a news app but an "information-distribution operating system"—algorithms replacing editors, data replacing intuition, a thousand faces replacing a single template.

Early on, ByteDance faced a critical choice. Less than a year after founding, an internet giant extended an acquisition olive branch—a higher valuation than the VCs offered, hundreds of millions of bundled-installation channels, and tens of millions of UV in traffic. Zhang agonized for a week and then refused, with clear and rational reasons: first, resources are a stimulant, and before internal strength is forged they would suppress the company's endogenous power; second, resources would distort strategy; third, he would be dragged into the giants' wars and forced to take sides; and fourth, an independent company was better placed to attract first-rate talent.[s31] It was precisely this decision that gave Toutiao the chance to become a platform-scale company. It was also the first major practice of Zhang's philosophy of delayed gratification—trading the temptation of rapid short-term growth for the possibility of long-term independent development.

The funding history: in July 2012 it raised a Series A of several million U.S. dollars from SIG Asia Investments and others.[s2] In September 2013 it raised a Series B of tens of millions of U.S. dollars from DST and others.[s2] In June 2014 it completed its Series C, led by Sequoia Capital, at a valuation above US$500 million.[s2][s3] In 2013, Zhang was named both to Forbes' "30 Under 30" entrepreneurs in China and to Fortune's "40 Under 40" business elites in China.[s1][s2]

Yet Toutiao's rise also came with controversy. By scraping traditional media's content it became the "enemy" of the traditional news-publishing industry. Algorithmic recommendation was also criticized for creating "filter bubbles"—users seeing only what they wanted to see, their horizons narrowing. The spread of vulgar content was also frequently raised.[s3] These controversies became an unavoidable footnote to Zhang's entrepreneurial road: algorithms may be value-neutral, but their use is not.

In August 2015, ByteDance launched its internationalization strategy, releasing TopBuzz, the overseas version of Toutiao, with the United States as its first stop.[s9][s10] The same year Zhang set out a "small goal" for globalization: he hoped to lead ByteDance into global markets within three years, with more than half of its users coming from overseas.[s10] He said: "China's internet population is only one-fifth of the world's. If we don't allocate resources globally, one-fifth simply cannot compete against four-fifths—so going overseas is inevitable."[s10] That remark would prove remarkably prescient.

Toutiao's success was not only a victory of technology but a victory of cognitive paradigm. Before Zhang, every Chinese news app followed the same model: an editorial team decided what users saw. The news apps of the four major portals—Sina, Sohu, NetEase and Tencent—were all, in essence, "editorial-driven" products. Zhang fundamentally overturned that logic—he did not need an editorial team, because algorithms could predict what each user wanted to see better than editors could.

That overturning provoked a fierce backlash from the traditional media industry. Toutiao was criticized as a mere "content mover," and the traditional news-publishing industry regarded it as an "enemy."[s3] Algorithmic recommendation was also accused of creating "filter bubbles"—users trapped inside the information they preferred, their horizons ever narrower. These criticisms were not without merit, but Zhang's choice was to keep optimizing the algorithm rather than retreat to an editorial model. He believed that, over the long run, algorithmic precision would keep improving, and the "filter bubble" problem could be solved through technical means (such as introducing diversity into recommendations).

Toutiao's A/B-testing culture also deserves mention. Eighty percent of features had to pass A/B testing before launch; Toutiao once iterated five versions in a single day and could run 200 experiments in a day.[s6][s7] The name "Douyin" itself was the product of an A/B test—although it ranked second, the team felt it better fit perception.[s6] This culture of "everything can be tested" made ByteDance's product iteration far faster than its peers'.

The years in Jinqiu Jiayuan on Zhichun Road became an important source of ByteDance's corporate culture. The apartment-office environment of Jinqiu Jiayuan shaped the startup DNA of "low-cost launch, high-density collaboration." Zhang recalled four reasons for choosing Jinqiu Jiayuan: at the very start there were few people and residential space was cost-effective; there was a kitchen for three meals a day; Zhichun Road was a startup street with concentrated talent; and daily life was convenient.[s3][s7][s34] The story of eight Haidilao pots with only four burners lit became one of the most widely circulated startup memories inside ByteDance.[s34]

Toutiao's recommendation engine had several key technical innovations. First, it did not rely solely on interests users explicitly expressed; it predicted interests from behavioral data (such as reading time, click frequency and swipe speed). Second, it used a real-time updating mechanism—every user action immediately influenced subsequent recommendations. Third, it built a complete user-interest graph capable of recommending across content types. These technical innovations made Toutiao's recommendation precision 2.3 times that of its peers.[s7]

A further important reason for Toutiao's astonishing growth, besides the algorithmic advantage, was that it did not depend on a reporting-and-editing team, so its content costs were extremely low. Traditional news apps needed large editorial teams, whereas Toutiao needed only algorithms. That "light-asset" model let ByteDance pour more resources into technical R&D and user growth.

The entrepreneurial spirit of the Jinqiu Jiayuan period became a key cornerstone of ByteDance's corporate culture. "Always be day-one"—elevated to the first of the ByteStyle principles in 2022—is precisely the cultural inheritance of the Jinqiu Jiayao spirit.[s35][s7] From the frugality of eight Haidilao pots with only four burners lit, to the grit of meeting more than 30 investors in a month, to the resolve of selling one's own home and committing wholeheartedly—these stories are passed down inside ByteDance from generation to generation.

Toutiao's success was not merely the success of one company; it redefined the rules of the game in the content industry. In the era of traditional media, the power of content distribution was held by editors—editors decided what content deserved recommendation, whether it went on the front page or in a corner. Toutiao used algorithms to replace the editor's role, making "user demand," rather than "editorial judgment," the core standard of content distribution. The impact of that paradigm shift extended far beyond the news industry—it changed the underlying logic of the entire information-distribution business.

Toutiao took only 90 days from launch to reach 10 million users, a speed rare even in the history of the Chinese internet.[s2] But even more impressive was its user stickiness—once users started using Toutiao, it was hard to go back to traditional news apps, because algorithmically recommended content really was more precise and more attuned to individual interests. That "once you use it you can't go back" user experience is a shared feature of every ByteDance product.

When he was named to Forbes' "30 Under 30" entrepreneurs in China in 2013, Zhang was only 30, and ByteDance had been founded less than a year.[s1][s2] Few at the time could have foreseen that this small company working out of a residential apartment on Zhichun Road would, within just a few years, grow into one of the world's largest unlisted technology companies.

### 4. 抖音与TikTok：从边缘项目到全球现象 / Douyin and TikTok: From a Marginal Project to a Global Phenomenon

2014至2015年，张一鸣曾两次讨论是否做短视频，因微视、秒拍、美拍等产品的存在而犹豫[s6][s8]。2016年，今日头条增速放缓，面临BAT的围攻——百度做信息流、腾讯推天天快报、阿里入局UC头条——竞争白热化。字节跳动开始加大UGC和视频化探索力度，成立IES团队负责内涵段子、抖音、火山等产品尝试。

2016年9月，抖音前身A.me正式上线，创始团队最初仅7人，由20出头刚毕业或实习的年轻人组成。上线初期日活不到1万。产品形态受Musical.ly启发，主打15秒音乐短视频[s6][s8]。抖音从一开始就是字节内部的边缘项目——没有豪华的团队配置，没有高层的重点关注。张一鸣在看到抖音留存和时长数据提升后，才开始增加资源投入，并引入头条的推荐算法团队改造产品，使抖音从运营主导转向算法驱动[s6][s8]。

2016年12月，A.me正式更名为"抖音短视频"——"抖音"这个名字本身就是A/B测试的结果。虽然排名第二，但团队认为其更符合认知、更能体现产品形态[s6]。这种"A/B测试一切"的文化，是字节跳动产品方法论的核心特征。

2017年成为抖音的爆发之年。3月，岳云鹏在微博转发带抖音水印的视频，抖音开始被公众关注[s6]。通过赞助《中国有嘻哈》等营销活动，抖音实现了品牌破圈。国庆期间，抖音日活从700万翻倍至1400万，全员狂欢庆祝。张一鸣却用数据泼了一盆冷水——他指出快手日活从6000万涨到8000万，增速远超抖音，差距还在拉大[s31]。"小胜即安是普通人的常态，大胜思危是顶级强者的本能。"[s31] 当抖音反超快手登顶后，他又预警"抖音正在复刻快手的弊病"。永远不满足于现状，永远在看差距而非庆祝成绩——这是延迟满足感在产品运营中的具体体现。

抖音的全屏上下滑沉浸式界面给用户带来全新体验，在中国规模教育了用户。短视频形态和单列形式是算法的最佳实验场——快速下滑表示不喜欢，完播率高表示喜欢，每一次滑动都是对用户偏好的一次数据采集[s8]。

全球化是张一鸣从一开始就写入字节跳动基因的战略。2016年，字节跳动确立"全球化"为最高优先级OKR[s7]。2017年5月，抖音海外版TikTok上线[s6]。11月，字节跳动以约8-10亿美元收购北美短视频社区Musical.ly——这不仅是获得美国用户，更是获得了进入欧美主流文化市场的通行证。据报道，张一鸣曾在凌晨3点召开视频会议："买下它，这是我们登陆北美的D-Day。"[s7] 2018年8月Musical.ly与TikTok合并，沿用TikTok名称，三个月内新增3000万用户[s6][s9]。保留原有团队，注入AI算法能力——这是字节跳动全球化的关键转折点。

TikTok的全球化策略可以概括为"不留空档、全面压制"的八字方针。把国家划分为S/A/B/C四个等级，美、英、德、日等发达国家拿走绝大部分投放预算。TikTok用统一的全球品牌加本地化运营加算法优势横扫全球[s10]。据经验，各国渗透率抵达20%-30%时能迎来拐点——App流行主要靠人际传播、自然增长[s10]。

张一鸣对全球化的理解是系统性的。他说："中国的互联网人口只占全球的五分之一，如果不在全球配置资源，追求规模化效应的产品，五分之一根本无法跟五分之四竞争，所以出海成为必然。"[s10] 他还以华为为对标："华为是字节跳动的榜样。华为在海外搭建电信的基础设施，构建了全球网络的连接；而我们希望能够实现中国移动软件应用的全球化。"[s10] 2019年，字节跳动旗下产品全球月活超15亿，覆盖150个国家和地区、75个语种[s9][s10]。同年入选《时代》杂志"全球百位最具影响力人物"[s2][s10]。

2020年12月，TikTok超越Facebook成为iOS和Android平台合计下载量最高的应用程序[s10]。TikTok成为首个在全球范围内获得广泛用户基础的中国互联网产品。

海外媒体对张一鸣的评价精准而深刻："张一鸣的遗产在于将全球数字媒体重新定位为基于算法的兴趣分发，迫使每个主要平台围绕字节跳动普及的格式进行重组。很少有创始人能如此直接地重塑数十亿人消费信息和娱乐的方式。他还证明了来自中国的互联网公司可以打造在全球市场具有竞争力的产品，拓展了中国创业者的全球化边界。"[s52]

抖音的成功有几个关键的产品决策值得深入分析：

第一，全屏沉浸式体验。抖音率先在中国市场大规模推广全屏竖屏播放模式。这种设计看似简单，实际上重新定义了用户与内容的交互方式——上下滑动取代了左右切换，每一次滑动都是一次"喜欢/不喜欢"的投票。这种极简的交互方式，让算法获得了最高效的反馈信号[s8]。

第二，算法驱动的内容分发。与快手早期的"关注"逻辑不同，抖音从一开始就以"推荐"为核心。用户打开APP看到的内容，不是他们关注的人发的，而是算法认为他们会喜欢的。这意味着即使是零粉丝的新用户，也能获得大量的曝光机会——只要内容足够好。这种机制极大地激发了创作者的积极性，形成了正向循环。

第三，音乐和特效的低门槛创作。抖音降低了视频创作的门槛——用户不需要专业的拍摄技巧，只需要选择音乐、套用模板，就能制作出有传播力的短视频。这种"模板化创作"与算法推荐的结合，产生了病毒式传播的效果。

TikTok的全球化策略同样值得深入分析。字节跳动内部将全球化视为"最高优先级OKR"[s7]，其执行力度在中国互联网企业中几乎无人能及。"不留空档、全面压制"的八字方针，意味着TikTok不会放弃任何一个重要市场[s10]。美、英、德、日被列为S级市场，拿走绝大部分投放预算——这是因为在发达国家建立品牌影响力，对全球其他地区具有"灯塔效应"。

收购Musical.ly是TikTok全球化进程中的决定性一步[s2][s6][s9]。这不仅是获得了一个美国用户基础，更重要的是获得了进入欧美主流青少年文化的入场券。Musical.ly的用户群体以青少年为主，这正是TikTok需要建立品牌认知的核心人群。合并后，字节跳动保留了原有团队，注入AI算法能力——这种"本地团队+中国算法"的组合，后来成为TikTok全球化的标准模式。

抖音的产品迭代速度令人惊叹。从2016年9月上线到2017年底的爆发，短短一年多时间里，抖音完成了从"音乐短视频工具"到"全民短视频平台"的转型。关键节点包括：岳云鹏微博转发带来的破圈效应[s6]、赞助《中国有嘻哈》的品牌升级、引入头条推荐算法后的内容质量提升、全屏沉浸式体验带来的用户粘性增强。

抖音的全球化扩张展现了字节跳动的执行力。TikTok在海外市场的增长速度远超预期。"不留空档、全面压制"的策略[s10]，意味着TikTok不是在一个市场站稳脚跟后再开拓下一个，而是在全球范围内同时铺开。

从更大的历史视角看，TikTok的全球化标志着中国互联网产业从"copy to China"到"copy from China"的历史性转变。在TikTok之前，中国互联网企业主要是在模仿硅谷的模式。TikTok是第一个由中国公司原创、在全球范围内获得巨大成功的产品形态——短视频+算法推荐。这个形态后来被全球的竞争对手纷纷效仿，包括YouTube Shorts、Instagram Reels等。

抖音和TikTok的成功，从产品方法论的角度看，核心在于三个"匹配"：第一，全屏沉浸式体验与移动端使用习惯的匹配——竖屏、单列、上下滑，完美适配了手机的操作方式。第二，算法推荐与短视频内容特征的匹配——短视频的信息密度适中，用户的"喜欢/不喜欢"判断可以在几秒内完成，这为算法提供了高频的反馈信号[s8]。第三，低门槛创作工具与用户表达欲的匹配——音乐模板、特效滤镜、一键拍摄，让每个人都能成为内容创作者。

字节跳动在抖音/TikTok上展现的A/B测试能力达到了极致。据公开信息，80%的功能上线前需完成A/B测试，今日头条曾1天迭代5个版本[s6][s7]。"抖音"这个名字本身就是A/B测试的结果[s6]。这种"一切皆可测试、数据说话"的文化，让字节跳动的产品决策更加理性，减少了"拍脑袋"式的错误。

TikTok的全球化不仅是一款产品的成功，更代表着一种新的商业模式的验证——来自中国的算法能力+全球化的运营网络，可以在全球范围内创造巨大的用户价值和商业价值。这个模式的成功，激励了更多中国互联网企业探索全球化。

**English:** In 2014 and 2015, Zhang twice discussed whether to move into short video, hesitating because of products such as Weishi, Miaopai and Meipai.[s6][s8] In 2016, Toutiao's growth slowed and it faced encirclement by the BAT trio—Baidu building information feeds, Tencent pushing Kuaibao, Alibaba entering with UC Toutiao—as competition turned white-hot. ByteDance began to step up its exploration of UGC and video, setting up the IES team to try products including Neihan Duanzi, Douyin and Huoshan.

In September 2016, A.me, Douyin's predecessor, formally launched, with an initial founding team of just seven people, made up of twenty-something recent graduates and interns. At the start its daily active users were under 10,000. The product form was inspired by Musical.ly and centered on 15-second music short videos.[s6][s8] From the outset Douyin was a marginal project inside ByteDance—no lavish team allocation, no top-level focus. Only after Zhang saw Douyin's retention and time-spent data improving did he begin to add resources, bringing in Toutiao's recommendation-algorithm team to remake the product and shifting Douyin from operations-led to algorithm-driven.[s6][s8]

In December 2016, A.me was formally renamed "Douyin Short Video"—the name "Douyin" itself was the result of an A/B test. Though it ranked second, the team felt it better fit perception and better expressed the product's form.[s6] This culture of "A/B-testing everything" is a core feature of ByteDance's product methodology.

2017 became Douyin's breakout year. In March, the comedian Yue Yunpeng reposted a video bearing Douyin's watermark on Weibo, and Douyin began to attract public attention.[s6] Through marketing such as sponsoring The Rap of China, Douyin broke out of its niche. Over the National Day holiday, Douyin's daily active users doubled from 7 million to 14 million, and the whole company celebrated wildly. Zhang, however, threw cold water with data—he pointed out that Kuaishou's daily active users had risen from 60 million to 80 million, growing far faster than Douyin, and that the gap was still widening.[s31] "Being content with a small win is the normal state of ordinary people; thinking of danger after a great victory is the instinct of top performers."[s31] After Douyin overtook Kuaishou to reach the top, he again warned that "Douyin is replicating Kuaishou's flaws." Never content with the status quo, always looking at the gap rather than celebrating results—this is delayed gratification made concrete in product operations.

Douyin's full-screen, swipe-up-and-down immersive interface brought users a wholly new experience, educating the Chinese market at scale. The short-video form and the single-column feed were the ideal proving ground for algorithms—swiping on quickly signals dislike, a high completion rate signals liking, and every swipe is a data point on user preference.[s8]

Globalization was a strategy Zhang wrote into ByteDance's genes from the very beginning. In 2016, ByteDance set "globalization" as its highest-priority OKR.[s7] In May 2017, TikTok, Douyin's overseas version, launched.[s6] In November, ByteDance acquired the North American short-video community Musical.ly for roughly US$0.8 to 1 billion—gaining not only American users but also a passport into the mainstream cultural markets of Europe and America. Zhang reportedly called a video meeting at three in the morning: "Buy it. This is our D-Day for landing in North America."[s7] In August 2018 Musical.ly merged with TikTok, keeping the TikTok name, and added 30 million users within three months.[s6][s9] Retaining the original team while injecting AI-algorithm capability—this was the pivotal turning point in ByteDance's globalization.

TikTok's globalization strategy can be summed up in an eight-character formula: "leave no gap, suppress on all fronts." Countries were divided into four tiers—S, A, B and C—with developed markets such as the United States, Britain, Germany and Japan taking the vast bulk of the marketing budget. TikTok swept the globe with a unified global brand, localized operations and algorithmic advantage.[s10] Experience showed that penetration reached an inflection point at 20%-30% in each country—an app's spread rests mainly on word of mouth and organic growth.[s10]

Zhang's understanding of globalization was systematic. He said: "China's internet population is only one-fifth of the world's. If we don't allocate resources globally, for products pursuing scale, one-fifth simply cannot compete against four-fifths—so going overseas is inevitable."[s10] He also took Huawei as a benchmark: "Huawei is ByteDance's model. Huawei built telecoms infrastructure overseas and constructed the connectivity of the global network; we hope to achieve the globalization of Chinese mobile software applications."[s10] In 2019, ByteDance's products surpassed 1.5 billion global monthly active users, covering 150 countries and regions and 75 languages.[s9][s10] The same year he was named to Time magazine's "100 Most Influential People in the World."[s2][s10]

In December 2020, TikTok overtook Facebook to become the most-downloaded app across iOS and Android combined.[s10] TikTok became the first Chinese internet product to build a broad user base on a global scale.

Overseas media's assessment of Zhang is precise and penetrating: "Zhang's legacy is the reorientation of global digital media around algorithmic interest distribution, forcing every major platform to reorganize itself around the formats ByteDance popularized. Few founders have so directly reshaped the way billions of people consume information and entertainment. He has also demonstrated that an internet company from China can build products competitive in global markets, expanding the global horizons of Chinese entrepreneurs."[s52]

Douyin's success involved several key product decisions worth analyzing in depth:

First, the full-screen immersive experience. Douyin pioneered the large-scale rollout of full-screen vertical video playback in the Chinese market. This design looks simple but in fact redefined how users interact with content—up-and-down swiping replaced left-and-right switching, and every swipe is a "like/dislike" vote. This ultra-minimal interaction gives the algorithm the most efficient feedback signal.[s8]

Second, algorithm-driven content distribution. Unlike Kuaishou's early "follow" logic, Douyin centered on "recommendation" from the start. The content users see when they open the app is not posted by people they follow but is what the algorithm believes they will like. This means that even brand-new users with zero followers can gain large amounts of exposure—so long as the content is good enough. The mechanism enormously energized creators, forming a virtuous circle.

Third, low-barrier creation through music and effects. Douyin lowered the threshold for video creation—users need no professional filming skill; they simply choose music and apply a template to produce a short video with viral potential. The combination of this "templated creation" and algorithmic recommendation produced virus-like spread.

TikTok's globalization strategy likewise repays close analysis. Inside ByteDance, globalization was treated as the "highest-priority OKR,"[s7] executed with a force almost unmatched among Chinese internet companies. The eight-character formula of "leave no gap, suppress on all fronts" meant TikTok would not abandon any important market.[s10] The U.S., Britain, Germany and Japan were classed as S-tier markets and took the bulk of the budget—because building brand influence in developed countries has a "lighthouse effect" for the rest of the world.

The acquisition of Musical.ly was the decisive step in TikTok's globalization.[s2][s6][s9] It delivered not only an American user base but, more importantly, an entry ticket into the mainstream youth culture of Europe and America. Musical.ly's users were predominantly teenagers—exactly the core demographic among whom TikTok needed to build brand recognition. After the merger, ByteDance retained the original team and injected AI-algorithm capability—this combination of "local team plus Chinese algorithms" later became the standard model for TikTok globalization.

Douyin's product-iteration speed was astonishing. From its launch in September 2016 to its breakout at the end of 2017, in little over a year Douyin completed the transformation from a "music short-video tool" to a "national short-video platform." Key milestones included: the break-out effect of Yue Yunpeng's Weibo repost,[s6] the brand upgrade from sponsoring The Rap of China, the improvement in content quality after Toutiao's recommendation algorithm was brought in, and the strengthened user stickiness from the full-screen immersive experience.

Douyin's global expansion showcased ByteDance's execution. TikTok's growth in overseas markets far exceeded expectations. The "leave no gap, suppress on all fronts" strategy[s10] meant TikTok did not secure one market before opening the next, but rolled out simultaneously across the globe.

From a larger historical vantage point, TikTok's globalization marks the Chinese internet industry's historic shift from "copy to China" to "copy from China." Before TikTok, Chinese internet companies mostly imitated Silicon Valley models. TikTok was the first product form—a short-video plus algorithmic recommendation—originally created by a Chinese company to achieve massive success worldwide. The form was later emulated by competitors across the globe, including YouTube Shorts and Instagram Reels.

Viewed through the lens of product methodology, the core of Douyin and TikTok's success lies in three "matches." First, the match between the full-screen immersive experience and mobile usage habits—vertical, single-column, swipe up-and-down—perfectly adapted to how phones are operated. Second, the match between algorithmic recommendation and the characteristics of short-video content—short video's information density is moderate, and users' "like/dislike" judgment can be made within a few seconds, providing the algorithm with high-frequency feedback signals.[s8] Third, the match between low-barrier creation tools and users' desire to express themselves—music templates, effect filters, one-tap shooting let everyone become a content creator.

ByteDance's A/B-testing capability reached its zenith on Douyin/TikTok. According to public information, 80% of features had to pass A/B testing before launch, and Toutiao once iterated five versions in a day.[s6][s7] The name "Douyin" itself was the result of an A/B test.[s6] This culture of "everything can be tested, let the data speak" made ByteDance's product decisions more rational, reducing "gut-call" errors.

TikTok's globalization was not merely the success of one product; it represented the validation of a new business model—Chinese algorithmic capability plus a global operating network can create enormous user value and commercial value worldwide. The success of this model has encouraged more Chinese internet companies to explore globalization.

### 5. TikTok美国风暴：七年博弈与合资落地 / The TikTok America Storm: Seven Years of Wrangling and a Joint-Venture Landing

TikTok在美国的经历，堪称中国互联网企业出海史上最跌宕起伏的篇章。从2019年美国外资投资委员会（CFIUS）开始针对字节跳动收购Musical.ly展开审查，到2026年8月联邦政府设备正式解禁，历时超过七年[s11][s12][s44]。这场风暴不仅考验了TikTok的法律应对能力，更考验了张一鸣作为创始人的战略定力和危机智慧。

2020年是风暴最为猛烈的一年。7月31日，美国政府表示将禁止TikTok在美国运营。8月3日，张一鸣发布内部信，坦承"TikTok美国业务面临被CFIUS强制要求出售的可能性，或因行政命令让TikTok产品在美国被封禁"[s11]。这封信的风格完全符合他一贯的沟通方式——平实、理性、几乎没有修辞，直接陈述问题和选项。8月4日，他又发一封内部信回应舆论争议："不要在乎短期损誉，耐心做好正确的事情；要有火星视角，格局大，EGO小"[s10][s11]。8月6日和14日，两道行政命令接踵而至——45天后禁止交易、90天内剥离美国业务[s11]。

面对压力，TikTok选择法律反击。2020年8月24日，TikTok向美国法院递交起诉书控告行政令违法[s11]。此后的数年间，法律诉讼持续推进。2024年3月7日，美国众议院以352票赞成、65票反对通过"非卖即走"提案——《保护美国人免受外国对手控制应用程序侵害法》[s11]。4月24日法案签署生效，要求字节跳动在2025年1月19日前出售TikTok美国业务[s11]。2025年1月17日，美国最高法院裁定法案不违宪[s11][s12]。1月18至19日，TikTok一度暂停在美服务，后于19日恢复[s11][s12]。

在漫长的法律与商业博弈中，张一鸣展现出他一贯的冷静与务实。"火星视角"——跳出当下舆论和情绪，站在更长远、更宏大的尺度上判断问题——不仅是危机应对的思维框架，也是延迟满足感在国际博弈中的另一种表达[s10]。他始终相信，只要做好产品、维护好用户关系，时间会站在正确的一边。

2025年12月，字节跳动、TikTok与三家投资者签署协议，成立TikTok美国数据安全合资有限责任公司[s12]。2026年1月23日正式宣告成立[s12][s44][s45]。

方案的核心设计精密而创新：股权结构——字节跳动保留19.9%股权（最大单一股东），甲骨文、银湖资本、阿联酋MGX各持股15%（合计45%），其余35.1%由戴尔家族办公室、海纳国际等十余家机构分散持有[s44][s45]。双实体切割——USDS公司负责数据安全、算法审核等敏感业务，全部美国用户数据存储在甲骨文位于得克萨斯州的本土云服务器中；TikTok美国公司（字节全资）负责广告、电商等核心商业板块。算法知识产权归属字节跳动，以授权方式给USDS使用[s44]。治理结构——7人董事会中美国籍成员占多数，Adam Presser任CEO[s12][s45]。

后续正常化进程加速。2026年3月31日纽约市率先恢复政府设备使用TikTok[s44]。8月11日白宫撤销联邦政府设备禁令[s44]。8月21日TikTok以4亿美元了结COPPA诉讼[s49]。

TikTok全球用户突破20亿，月活超15亿。2025年广告收入达347亿美元，美国占41%[s14]。字节2025年整体营收超1500亿美元，增速25%-30%[s14]。从"生存危机"到"监管框架内运营"，TikTok完成了中国互联网企业史上最复杂的一次合规转型，也为所有中国互联网企业的全球化提供了一个新范式。

从更宏观的视角看，TikTok美国风波反映了中国互联网企业全球化面临的新现实。在TikTok之前，中国互联网的出海主要是工具类应用（如UC浏览器、Cheetah Mobile）或游戏类应用，它们面临的监管压力相对较小。但TikTok作为一款内容平台，触动了更深层的神经——它不仅涉及数据安全问题，还涉及信息传播和文化影响力。这使得TikTok面临的挑战远超此前的中国互联网出海企业。

张一鸣在处理这场危机时展现出的"火星视角"，不仅是一种危机管理策略，更是他延迟满足感哲学的国际化表达[s10]。他选择了最难但最正确的路——通过法律途径捍卫权益，通过商业谈判找到解决方案，而不是简单地放弃或对抗。最终达成的合资方案，虽然字节跳动在股权上做了让步（从全资到19.9%），但保住了最核心的资产——算法知识产权。

TikTok美国方案的设计之精密，在全球科技企业合规史上也属罕见。"双实体切割"模式——将数据安全敏感业务与商业运营分离——为所有面临类似挑战的跨国企业提供了一个新的参考模板。算法知识产权归属字节跳动、以授权方式给合资公司使用的安排，既满足了美方的安全要求，又保全了字节的长期技术资产[s44]。

字节2025年整体营收超1500亿美元，增速25%-30%，海外增速60%以上[s14]。TikTok全球用户突破20亿，月活超15亿，2025年广告收入347亿美元[s14]。这些数字说明，尽管经历了七年的风波，TikTok的商业价值不仅没有萎缩，反而持续增长。市场对字节跳动的估值从2025年8月的约3300亿美元升至2026年初的5500亿美元（路透社报道General Atlantic出售交易）[s13]。

在这场长达七年的风暴中，张一鸣的角色是独特而复杂的。2021年卸任CEO后，他名义上已经不再是字节跳动的日常管理者。但在TikTok美国问题上，他始终深度参与战略决策。"火星视角"这个概念值得深入解读——站在火星上看地球，所有的争论、焦虑和恐惧都变得渺小。这种视角帮助张一鸣在危机中保持冷静，不被短期舆论绑架[s10]。

TikTok美国方案的最终落地，也体现了张一鸣一贯的"延迟满足感"哲学。从2020年危机爆发到2026年方案落地，历时六年。张一鸣选择了最难但最正确的路——通过法律途径捍卫权益，通过商业谈判找到创新性的解决方案。字节2025年整体营收超1500亿美元，增速25%-30%[s14]。TikTok全球用户突破20亿，2025年广告收入347亿美元[s14]。市场估值从2025年8月的约3300亿美元升至2026年初的5500亿美元（路透社）[s13]。

从法律角度看，TikTok美国方案的设计展现了极高的法律智慧和商业创造力。"双实体切割"模式——将数据安全敏感业务与商业运营分离——在全球科技企业合规史上几乎没有先例[s44]。算法知识产权归属字节跳动、以授权方式给合资公司使用的安排，既满足了美方对数据安全的要求，又保全了字节的核心技术资产。

COPPA 4亿美元和解虽然金额巨大，但从另一个角度看，它是TikTok从"生存危机"走向"正常运营"的重要一步。司法部承认TikTok所有权和管理层已发生"重大变化"[s49]——这实际上是对TikTok美国方案效果的官方认可。

从字节跳动2025年的财务数据看，TikTok的商业价值不仅没有因七年风波而萎缩，反而持续增长。全球用户突破20亿，月活超15亿，广告收入347亿美元[s14]。这些数据说明，尽管面临前所未有的政治和法律压力，TikTok的产品力和用户基础依然强劲。

**English:** TikTok's experience in the United States is arguably the most turbulent chapter in the history of Chinese internet companies going overseas. From 2019, when the Committee on Foreign Investment in the United States (CFIUS) began reviewing ByteDance's acquisition of Musical.ly, to August 2026, when the ban on federal government devices was formally lifted, the saga spanned more than seven years.[s11][s12][s44] This storm tested not only TikTok's legal response capability but also Zhang Yiming's strategic composure and crisis wisdom as a founder.

2020 was the year the storm raged hardest. On July 31, the U.S. government said it would ban TikTok from operating in the United States. On August 3, Zhang issued an internal letter acknowledging that "TikTok's U.S. business faces the possibility of being forced into a sale by CFIUS, or of the TikTok product being banned in the United States by executive order."[s11] The letter's style was entirely in keeping with his usual mode of communication—plain, rational, almost devoid of rhetoric, directly stating the problem and the options. On August 4, he sent another internal letter responding to the public controversy: "Don't worry about short-term praise or blame; patiently do the right thing. Keep the view from Mars—big on perspective, small on ego."[s10][s11] On August 6 and 14, two executive orders came in succession—a ban on transactions after 45 days and a divestiture of the U.S. business within 90 days.[s11]

Facing the pressure, TikTok chose to fight back in court. On August 24, 2020, TikTok filed a lawsuit in a U.S. court charging that the executive order was unlawful.[s11] Over the years that followed, the legal proceedings ground on. On March 7, 2024, the U.S. House of Representatives passed the "sell-or-be-banned" bill—the Protecting Americans from Foreign Adversary Controlled Applications Act—by 352 votes to 65.[s11] On April 24 the bill was signed into law, requiring ByteDance to sell TikTok's U.S. business by January 19, 2025.[s11] On January 17, 2025, the U.S. Supreme Court ruled the law constitutional.[s11][s12] On January 18-19, TikTok briefly suspended service in the United States before restoring it on the 19th.[s11][s12]

Through the long legal and commercial wrangling, Zhang displayed his characteristic calm and pragmatism. The "view from Mars"—stepping outside the immediate public opinion and emotion of the moment to judge issues on a longer, grander scale—was not only a mental framework for crisis response but another expression of delayed gratification in international bargaining.[s10] He consistently believed that so long as the product was good and user relationships were well tended, time would stand on the side of what was right.

In December 2025, ByteDance and TikTok signed an agreement with three investors to establish TikTok U.S. Data Security Joint Venture LLC.[s12] It was formally announced as established on January 23, 2026.[s12][s44][s45]

The core design of the arrangement was intricate and innovative. Equity structure: ByteDance retained a 19.9% stake (the largest single shareholder); Oracle, Silver Lake and the Abu Dhabi-based MGX each held 15% (45% combined); the remaining 35.1% was dispersed across more than a dozen institutions including the Dell family office and Susquehanna International Group.[s44][s45] A two-entity split: the USDS company handled sensitive business such as data security and algorithm review, with all U.S. user data stored on Oracle's domestic cloud servers in Texas; the TikTok U.S. company (wholly owned by ByteDance) ran core commercial lines such as advertising and e-commerce. The algorithm's intellectual property belonged to ByteDance and was licensed to USDS for use.[s44] Governance: of the seven-member board, U.S. nationals held the majority, and Adam Presser served as CEO.[s12][s45]

The subsequent normalization accelerated. On March 31, 2026, New York City became the first to restore the use of TikTok on government devices.[s44] On August 11, the White House revoked the ban on federal government devices.[s44] On August 21, TikTok settled the COPPA lawsuit for US$400 million.[s49]

TikTok's global users surpassed 2 billion, with monthly active users above 1.5 billion. In 2025 its advertising revenue reached US$34.7 billion, 41% of it from the United States.[s14] ByteDance's overall 2025 revenue exceeded US$150 billion, growing 25%-30%.[s14] From "existential crisis" to "operating within a regulatory framework," TikTok completed the most complex compliance transformation in the history of Chinese internet companies, and in doing so offered a new paradigm for the globalization of every Chinese internet firm.

From a more macro perspective, the TikTok America affair reflects the new reality facing the globalization of Chinese internet companies. Before TikTok, Chinese overseas expansion was led mainly by tool apps (such as UC Browser and Cheetah Mobile) or games, which faced relatively light regulatory pressure. But TikTok, as a content platform, touched a deeper nerve—it involved not only data security but also information dissemination and cultural influence. That made the challenges facing TikTok far greater than those of any earlier Chinese company going overseas.

The "view from Mars" that Zhang displayed in handling this crisis was not only a crisis-management strategy but the international expression of his philosophy of delayed gratification.[s10] He chose the hardest but most correct path—defending his rights through legal channels and finding a solution through commercial negotiation, rather than simply giving up or confronting. The joint-venture arrangement ultimately reached, though it required ByteDance to concede on equity (from full ownership to 19.9%), preserved the most core asset of all—the algorithm's intellectual property.

The sophistication of the TikTok U.S. arrangement's design is rare even in the global history of technology-company compliance. The "two-entity split" model—separating data-security-sensitive business from commercial operations—offers a new reference template for every multinational facing similar challenges. The arrangement whereby the algorithm's IP belongs to ByteDance and is licensed to the joint venture both satisfies the American side's security requirements and preserves ByteDance's long-term technical asset.[s44]

ByteDance's overall 2025 revenue exceeded US$150 billion, growing 25%-30%, with overseas growth above 60%.[s14] TikTok's global users surpassed 2 billion and monthly active users 1.5 billion, with 2025 advertising revenue of US$34.7 billion.[s14] These figures show that, despite seven years of turmoil, TikTok's commercial value did not wither but kept growing. The market's valuation of ByteDance rose from roughly US$330 billion in August 2025 to US$550 billion in early 2026 (per Reuters' reporting on a General Atlantic sell-down).[s13]

In this seven-year storm, Zhang's role was distinctive and complex. After stepping down as CEO in 2021, he was nominally no longer ByteDance's day-to-day manager. But on the TikTok America question he was always deeply involved in strategic decisions. The concept of the "view from Mars" repays close reading—standing on Mars and looking back at Earth, all the arguments, anxieties and fears shrink to smallness. That perspective helped Zhang stay calm in the crisis and avoid being taken hostage by short-term public opinion.[s10]

The eventual landing of the TikTok U.S. arrangement also embodies Zhang's characteristic philosophy of delayed gratification. From the crisis's outbreak in 2020 to the deal's landing in 2026, six years passed. Zhang chose the hardest but most correct path—defending rights through legal channels and finding an innovative solution through commercial negotiation. ByteDance's overall 2025 revenue exceeded US$150 billion, growing 25%-30%.[s14] TikTok's global users surpassed 2 billion, with 2025 advertising revenue of US$34.7 billion.[s14] The market valuation rose from roughly US$330 billion in August 2025 to US$550 billion in early 2026 (Reuters).[s13]

From a legal standpoint, the design of the TikTok U.S. arrangement displayed extraordinary legal intelligence and commercial creativity. The "two-entity split" model—separating data-security-sensitive business from commercial operations—has almost no precedent in the global history of technology-company compliance.[s44] The arrangement under which the algorithm's IP belongs to ByteDance and is licensed to the joint venture both meets the American side's data-security demands and preserves ByteDance's core technical asset.

Although the US$400 million COPPA settlement was a large sum, seen from another angle it was an important step in TikTok's move from "existential crisis" toward "normal operations." The Department of Justice acknowledged that TikTok's ownership and management had undergone "significant changes"[s49]—in effect an official recognition of the TikTok U.S. arrangement's effectiveness.

Judging from ByteDance's 2025 financials, TikTok's commercial value not only failed to wither amid seven years of turmoil but kept growing. Global users surpassed 2 billion, monthly active users 1.5 billion, and advertising revenue reached US$34.7 billion.[s14] These data show that, despite unprecedented political and legal pressure, TikTok's product strength and user base remain robust.

### 6. AI新十年：Seed团队、豆包崛起与拒绝蒸馏 / The New AI Decade: The Seed Team, Doubao's Rise and Refusing to Distill

2021年卸任CEO后，张一鸣逐渐退出日常管理——2022年退出多家字节系公司法定代表人[s2]。但在AI战略上，他选择了深度回归。这是字节跳动历史上第一次，创始人从"退"到"进"，不是因为管理需要，而是因为技术判断。

2023年5月，张一鸣在香港成立个人投资基金Cool River Venture[s1]。同年8月，字节跳动推出豆包APP，首批提供文本对话与基础图像生成能力[s16][s17]。2024年初，字节将"云雀大模型"改名为"豆包大模型"，豆包承担"试验田"功能，与旗下几乎所有产品打通[s16]。

从2024年下半年开始，张一鸣每月亲自参与Seed（字节AI大模型团队）的复盘与讨论会。创始人重新深度介入一线技术讨论，标志着字节战略重心向AI迁移[s16][s17]。2024年2月，张一鸣化身"猎头"亲自监督从竞对公司挖掘AI技术人才，以8位数年薪挖走原阿里通义技术团队关键成员。发起面向顶尖应届博士的Top Seed人才计划[s16]。

2025年2月，前Google DeepMind研究副总裁吴永辉加入字节，出任Seed基础研究负责人，直接向CEO梁汝波汇报（CEO-1级别）[s18][s41]。吴永辉在Google工作17年，Google Scholar显示其论文被引超7.3万次。入职后他推动AI模型线重组，分设三大虚拟团队：Edge（前沿AGI研究，3年考核周期，取消季度OKR）、Focus（基础模型攻坚）、Base（当前一代模型工程/数据/测评）[s41]。2025年4月，AI Lab整体并入Seed，使其成为"公司级AI中枢"[s41]。

Seed团队整体规模约1500人，几乎不再从外部招聘中高层技术管理者，更重视招聘应届毕业生和提拔年轻人。2025年中发放独立于公司期权的"豆包虚拟股"激励。吴永辉入职后三个月，Seed发布的论文数量超过2024年全年[s41]。

豆包的崛起速度惊人。2024年9月，豆包成为国内用户规模破亿的首个AI应用[s16]。2026年2月，发布豆包2.0——这是Seed成立以来训练的最大模型（据媒体报道为1万亿参数，多模态）[s18][s41]。6月，豆包月活达3.82亿（QuestMobile数据），DAU突破2亿，日均Token调用量达180万亿，两年增长超1500倍[s14][s17][s46]。6月23日推出豆包2.1 Pro，在Coding、Agent、VLM三大方向跨越生产级质变点[s46][s47]。6月24日上线三档付费订阅（标准版68元/月、加强版200元/月、专业版500元/月）[s43][s46]。

然而，豆包面临严峻的商业化困境。据国联民生证券测算，日均综合运营成本1.32亿至2.4亿元[s43][s46]。日收入不足100万元，几乎完全依赖电商佣金[s46]。但B端被视为核心盈利方向——火山引擎MaaS占中国公有云MaaS市场49.5%份额[s46]。

2026年7月，张一鸣在Seed全员会上罕见现身，明确表示"字节跳动不会把蒸馏当作提升AI模型能力的捷径"。随后Seed出台政策，明确禁止蒸馏K3等开放权重模型[s18][s48]。"沿着别人的路走，最多只能逼近，很难超越。"[s18] 这一表态体现了他对技术路线的长期主义判断——也是延迟满足感在AI竞争中的又一次极致实践。

字节AI战略重心正从C端流量扩张转向ToB企业服务。2026年7月飞书产品团队并入豆包，是2021年以来最大ToB架构调整[s48]。8月成立"AI数据与安全"一级部门[s18]。梁汝波将公司战略总结为"高度优先，粗主干，优化长期"——AI是最粗的那根主干[s48]。

投入力度：2025年字节资本开支约1500亿元（其中900亿用于AI算力采购）[s14]。2026年上调至2000亿元以上，4月以56亿美元采购华为昇腾芯片约35万颗[s46]。字节大模型业务ARR突破40亿美元[s46]。

Seed团队的重组，是字节跳动AI战略的转折点。Edge团队采用3年考核周期、取消季度OKR的设置，在全球科技企业中也极为罕见——这直接借鉴了基础科学研究的逻辑：前沿AGI研究需要长期投入，不能用短期的商业指标来衡量[s41]。Focus团队负责基础模型攻坚，Base团队负责当前一代模型的工程化、数据建设和测评——三个团队分别对应"长期探索""中期攻坚""短期交付"，形成了一个完整的时间维度覆盖。

吴永辉的加盟对Seed团队的影响是质变性的。作为前Google DeepMind研究副总裁，他带来了世界顶尖的AI研究方法论和人才网络。入职后三个月，Seed发布的论文数量超过2024年全年[s41]。这不仅是个人能力的体现，更是因为他重新定义了Seed的研究标准和工作节奏。

豆包的商业化困境，是当前全球AI行业的共同挑战。大模型的推理成本极高——日均Token调用量180万亿，单次推理成本中硬件折旧占58%，电力消耗占29%[s46][s43]。据国联民生证券测算，日均综合运营成本1.32亿至2.4亿元[s43]。但字节跳动选择持续投入，赌的是AI将成为下一代计算平台——谁先建立起最大的用户基础和领先的模型能力，谁就能在AI时代占据制高点。

2026年字节资本开支计划上调至2000亿元以上，其中4月以56亿美元采购华为昇腾芯片约35万颗[s46]。这一采购不仅体现了字节对AI算力的巨大需求，也标志着字节跳动在美国芯片出口管制背景下，积极构建国产芯片替代方案。字节大模型业务ARR突破40亿美元[s46]——虽然相对于巨额投入仍显不足，但增长势头强劲。

在AI竞赛中，字节跳动面临着来自国内外的强大竞争。百度的文心一言、阿里的通义千问、腾讯的混元大模型、华为的盘古大模型都在快速迭代。国际上，OpenAI的GPT系列、Anthropic的Claude、Google的Gemini等也在不断突破。在这样的竞争环境下，张一鸣选择"拒绝蒸馏"——宁愿短期排名落后，也要坚持自研路线——需要极大的勇气和战略定力。

"蒸馏"是将大模型的知识"蒸馏"到小模型中的技术。张一鸣认为这不是正确的长期策略。"沿着别人的路走，最多只能逼近，很难超越"[s18]——这句话体现了他对技术创新本质的深刻理解。

字节跳动在AI领域的投入是惊人的。2025年资本开支约1500亿元[s14]。2026年上调至2000亿元以上，4月以56亿美元采购华为昇腾芯片约35万颗[s46]。Seed团队Edge组的3年考核周期和取消季度OKR[s41]，在全球AI研究团队中极为罕见——前沿AGI研究是一场马拉松，不是短跑。

字节跳动AI战略的另一个值得关注的维度是人才策略。2024年发起的Top Seed人才计划，面向顶尖应届博士，招募约30名顶尖人才[s16]。2025年吴永辉加盟后，Seed团队几乎不再从外部招聘中高层技术管理者，更重视招聘应届毕业生和提拔年轻人[s41]。这种"招最优秀的人、给最大的空间、用最长的时间考核"的人才策略，与Edge组的3年考核周期一脉相承。

2026年7月飞书并入豆包，标志着字节跳动AI战略重心的重大调整[s48]。飞书作为字节跳动最重要的ToB产品之一，并入豆包意味着AI不再只是一个独立的应用，而是要深度融入企业协作的每一个环节。这个决策背后的逻辑是：AI的终极价值不在于独立的聊天应用，而在于成为所有工作场景的底层能力。

梁汝波将公司战略总结为"高度优先，粗主干，优化长期"[s48]。"高度优先"意味着聚焦最高价值的方向（AI），"粗主干"意味着做强核心业务而非盲目扩张，"优化长期"意味着为长期目标牺牲短期收益。这十二个字，可以说是字节跳动当前战略的精髓。

**English:** After stepping down as CEO in 2021, Zhang Yiming gradually withdrew from day-to-day management—in 2022 he exited the legal-representative role at several ByteDance-affiliated companies.[s2] But on AI strategy he chose a deep return. For the first time in ByteDance's history, a founder moved from "retreat" to "advance" not out of managerial necessity but because of a technical judgment.

In May 2023, Zhang set up a personal investment fund in Hong Kong, Cool River Venture.[s1] In August the same year, ByteDance launched the Doubao app, initially offering text conversation and basic image generation.[s16][s17] In early 2024, ByteDance renamed its "Skylark large model" the "Doubao large model," with Doubao serving as an "experimental field" connected to nearly every product under the company umbrella.[s16]

From the second half of 2024, Zhang personally took part every month in the reviews and discussion meetings of Seed (ByteDance's large-model AI team). The founder's renewed deep involvement in front-line technical discussions marked the migration of ByteDance's strategic center of gravity toward AI.[s16][s17] In February 2024, Zhang turned himself into a "headhunter," personally overseeing the poaching of AI technical talent from rival companies, using eight-figure annual salaries to lure away key members of Alibaba's Tongyi technical team. He also launched the Top Seed talent program aimed at top newly graduated PhDs.[s16]

In February 2025, Wu Yonghui, a former vice president of research at Google DeepMind, joined ByteDance as head of foundational research for Seed, reporting directly to CEO Liang Rubo (at CEO-minus-one level).[s18][s41] Wu spent 17 years at Google, and Google Scholar shows his papers have been cited more than 73,000 times. After joining he drove a reorganization of the AI model lines, setting up three virtual teams: Edge (frontier AGI research, with a three-year assessment cycle and no quarterly OKRs), Focus (foundation-model breakthrough work) and Base (engineering, data and evaluation for the current generation of models).[s41] In April 2025, AI Lab was merged wholesale into Seed, making it a "company-wide AI hub."[s41]

The Seed team totals about 1,500 people; it now hardly hires mid-to-senior technical managers from outside, placing more weight on recruiting fresh graduates and promoting young people. In mid-2025 it issued "Doubao virtual shares" as an incentive, separate from the company's stock options. In the three months after Wu Yonghui's arrival, Seed published more papers than in the whole of 2024.[s41][s41]

Doubao's rise has been astonishing. In September 2024, Doubao became the first AI app in China to surpass 100 million users.[s16] In February 2026 it released Doubao 2.0—the largest model trained since Seed was founded (reported by media at 1 trillion parameters, multimodal).[s18][s41] In June, Doubao's monthly active users reached 382 million (per QuestMobile), DAU broke 200 million, and daily token calls hit 180 trillion—a more than 1,500-fold increase in two years.[s14][s17][s46] On June 23 it launched Doubao 2.1 Pro, crossing the production-grade inflection point in three directions: coding, agents and VLM.[s46][s47] On June 24 it went live with three paid subscription tiers (standard at 68 yuan/month, plus at 200 yuan/month, pro at 500 yuan/month).[s43][s46]

Yet Doubao faces a severe commercialization dilemma. According to estimates by Guolian Minsheng Securities, its daily comprehensive operating cost runs between 132 million and 240 million yuan.[s43][s46] Daily revenue is under 1 million yuan, almost entirely dependent on e-commerce commissions.[s46] But the business-to-business side is regarded as the core profit direction—Volcano Engine's MaaS holds a 49.5% share of China's public-cloud MaaS market.[s46]

In July 2026, Zhang made a rare appearance at a Seed all-hands meeting and stated clearly that "ByteDance will not treat distillation as a shortcut to improving our AI models." Seed subsequently issued a policy expressly forbidding the distillation of open-weight models such as K3.[s18][s48] "If you follow in others' footsteps, the most you can do is draw close; it is very hard to overtake."[s18] The statement embodies his long-termist judgment on technical route—and is another extreme practice of delayed gratification in the AI competition.

ByteDance's AI strategic focus is shifting from consumer-end traffic expansion toward business-to-business enterprise services. In July 2026 the Lark product team was merged into Doubao, the largest ToB structural adjustment since 2021.[s48] In August a first-tier "AI Data and Security" department was established.[s18] Liang Rubo summed up company strategy as "high priority, thick trunk, optimize for the long term"—AI is the thickest trunk of all.[s48]

Investment intensity: ByteDance's 2025 capital expenditure was about 150 billion yuan (of which 90 billion went to AI computing-power procurement).[s14] In 2026 it was raised to more than 200 billion yuan, and in April it bought roughly 350,000 Huawei Ascend chips for US$5.6 billion.[s46] ByteDance's large-model business ARR broke through US$4 billion.[s46]

The reorganization of the Seed team was the turning point of ByteDance's AI strategy. The Edge team's setup—a three-year assessment cycle and the abolition of quarterly OKRs—is extraordinarily rare even among global technology companies; it borrows directly from the logic of basic scientific research: frontier AGI research requires long-term investment and cannot be measured by short-term commercial indicators.[s41] The Focus team handles foundation-model breakthrough work, and the Base team handles the engineering, data building and evaluation of the current generation of models—the three teams corresponding respectively to "long-term exploration," "mid-term assault" and "short-term delivery," forming a complete coverage of the time dimension.

Wu Yonghui's arrival had a qualitative impact on the Seed team. As a former vice president of research at Google DeepMind, he brought world-class AI research methodology and a talent network. In the three months after he joined, Seed published more papers than in the whole of 2024.[s41] That reflects not only personal ability but also the fact that he redefined Seed's research standards and working rhythm.

Doubao's commercialization dilemma is a shared challenge across the global AI industry today. The inference cost of large models is extremely high—180 trillion daily token calls, with hardware depreciation accounting for 58% of per-inference cost and power consumption for 29%.[s46][s43] According to Guolian Minsheng Securities, daily comprehensive operating cost runs between 132 million and 240 million yuan.[s43] But ByteDance has chosen to keep investing, betting that AI will become the next-generation computing platform—whoever first builds the largest user base and the strongest model capability will occupy the high ground in the AI era.

ByteDance's 2026 capital-expenditure plan was raised to more than 200 billion yuan, including the April purchase of roughly 350,000 Huawei Ascend chips for US$5.6 billion.[s46] The procurement reflects not only ByteDance's enormous demand for AI computing power but also marks its active construction of a domestic-chip alternative against the backdrop of U.S. chip export controls. ByteDance's large-model business ARR broke through US$4 billion[s46]—still small relative to the enormous investment, but with a strong growth momentum.

In the AI race, ByteDance faces powerful competition at home and abroad. Baidu's ERNIE Bot, Alibaba's Tongyi Qianwen, Tencent's Hunyuan large model and Huawei's Pangu model are all iterating fast. Internationally, OpenAI's GPT series, Anthropic's Claude and Google's Gemini keep breaking new ground. In this competitive environment, Zhang's choice to "refuse distillation"—insisting on the in-house research route even at the cost of short-term ranking lag—requires enormous courage and strategic composure.

"Distillation" is the technique of "distilling" a large model's knowledge into a smaller model. Zhang believes it is not the correct long-term strategy. "If you follow in others' footsteps, the most you can do is draw close; it is very hard to overtake"[s18]—the sentence reflects his deep understanding of the essence of technological innovation.

ByteDance's investment in AI is staggering. Its 2025 capital expenditure was about 150 billion yuan.[s14] In 2026 it was raised to more than 200 billion yuan, and in April it bought roughly 350,000 Huawei Ascend chips for US$5.6 billion.[s46] The Seed Edge group's three-year assessment cycle and abolition of quarterly OKRs[ s41] are extraordinarily rare among global AI research teams—frontier AGI research is a marathon, not a sprint.

Another dimension of ByteDance's AI strategy worth noting is its talent strategy. The Top Seed talent program launched in 2024 targets top newly graduated PhDs, recruiting about 30 elite talents.[s16] After Wu Yonghui's arrival in 2025, the Seed team hardly hired mid-to-senior technical managers from outside, placing more weight on recruiting fresh graduates and promoting young people.[s41] This talent strategy of "hire the best people, give them the most space, assess over the longest time horizon" is of a piece with the Edge group's three-year assessment cycle.

The merger of Lark into Doubao in July 2026 marked a major adjustment in ByteDance's AI strategic focus.[s48] As one of ByteDance's most important ToB products, Lark's merger into Doubao means AI is no longer just a standalone app but is to be woven deeply into every link of enterprise collaboration. The logic behind the decision: AI's ultimate value lies not in an independent chat app but in becoming the underlying capability of every work scenario.

Liang Rubo summed up company strategy as "high priority, thick trunk, optimize for the long term."[s48] "High priority" means focusing on the highest-value direction (AI); "thick trunk" means strengthening the core business rather than expanding blindly; "optimize for the long term" means sacrificing short-term returns for long-term goals. These twelve characters are, one might say, the essence of ByteDance's current strategy.

### 7. 管理哲学：延迟满足感与算法式组织思维 / Management Philosophy: Delayed Gratification and Algorithm-Style Organizational Thinking

张一鸣的管理哲学在中国互联网界独树一帜。它不是那种充满个人魅力的叙事，而是一套冷静、理性、几乎可以用代码来描述的操作体系。核心概念包括延迟满足感、Context not Control、公司也是产品、信息流优于权力流，以及平常心。

延迟满足感是张一鸣最核心的人生哲学。源于M·斯科特·派克《少有人走的路》——这本书他在微软半年期间阅读，后来成为影响他最深的著作之一[s31][s3]。书中核心观点："延迟满足感的延后意味对抗人性的弱点，放弃对暂时性安逸的追逐和享受，重新设置人生快乐与痛苦的次序。"[s31]

张一鸣将其定义为：不以即时反馈作为决策唯一标准，把时间尺度从3个月拉长到3年来判断。警惕为短期收入、利润、融资、估值、舆论而破坏长期价值。长期值得投入的包括人才、产品、技术、用户体验、组织能力、品牌、学习能力、公司文化和基础设施[s3][s19][s20]。

他在2016年《财经》杂志专访中说："很多人人生中一半的问题都是这个原因造成的——没有延迟满足感。延迟满足感的本质是克服人性弱点，而克服弱点，是为了更多的自由。"[s31] "我最欣赏自己的特质，就是延迟满足感，而最大的延迟满足感，是思维上的。"[s31] 甚至认为"延迟满足感程度不在一个量级的人，是没法进行有效沟通的"[s31]。

这一哲学在创业历程中反复验证——拒绝巨头收购诱惑、胜利时看差距而非庆祝、毕业薪酬差三五千可忽略、卖房创业全身心投入、2026年拒绝蒸馏走捷径——每一次都是延迟满足感在不同维度上的实践[s31]。

Context, not Control是管理方法论。受哈耶克"理性的自负"理论启发，2017年4月在源码资本年会上做《做CEO要避免理性的自负》的分享[s3][s19]。核心思想：管理者不应告诉员工每一步具体怎么做，而应提供足够的Context（上下文），然后让距离问题最近的人做决定[s3][s19][s20]。"CEO要避免理性的自负。自上而下的宏观战略往往都是灾难。用优秀的人和高质量信息代替越来越复杂的控制系统。"[s19][s20]

在字节的实践中，任何员工在内网IM上可以直接看到张一鸣的OKR。每两个月的"CEO面对面"，张一鸣会公开给自己的OKR逐项打分。飞阅会制度——会前阅读材料，直接插入问题和建议[s20]。

公司也是产品（Develop a company as a product）：创业者同时做两个产品——给用户的产品和给员工的产品（公司本身）。CEO是这两个产品的产品经理。员工ROI公式：功能性回报加心理性回报除以功能性投入加心理性投入。人才三标准：超强学习能力、视野开阔、心智成熟[s50]。

平常心是2021年字节9周年演讲主题。"大部分人都是平常人，但平常人可以做非常事。"保持理性，不被情绪绑架。"平庸有重力，需要逃逸速度。"[s4][s19] "你打游戏有没有遇到过这种情况，某个游戏要过100关，你打到99关的时候，就容易手抖。"[s4]

张一鸣的个人性格深刻塑造了字节的文化底色。极度低调，很少参加行业会议。承认"我不是很擅长社交，我更喜欢研究组织和市场原理来减少管理，喜欢自己上网、看书、听歌、发呆"[s4]。穿着朴素，T恤加拖鞋。说话时语调语速几乎没有情绪起伏。"最激烈的情绪表达方式不过是'这个事情怎么会这样？'"[s31] 最佳工作状态是"在轻度喜悦和轻度沮丧之间"[s31]。一位高管评价他"真的像机器一样在工作"[s31]。

早期在酷讯"做完自己的工作后，对于大部分同事的问题，只要我能帮助解决，我都去做"[s31]。字节范——追求极致、务实敢为、开放谦逊、坦诚清晰、始终创业。2022年6月，"始终创业"被提升至第一位[s35][s7]。2019年7周年大会提出"务实的浪漫"："精致的文艺不是浪漫，粗糙的宏大是浪漫。有生命力是浪漫，面向未来是浪漫，拥抱不确定性是浪漫，保持可能性是浪漫。"[s3][s4] 签名档是"喜欢发呆"——"不是放空，是自己思考一些非常无边无际和少有人讨论的点子"[s4]。

张一鸣的"十二问框架"是他面对创业和管理问题时的默认思考工具。这套框架包括：真正的问题是什么？事实/假设/愿望分别是什么？背后发生了什么结构性变化？用户真正的基本需求是什么？有没有把Activity当成Result？是否用短期收益牺牲长期价值？是否人才不足却试图增加流程？信息有没有流到真正做决定的人那里？如果判断错误，怎样最快发现并回滚？三年以后回头看，现在最值得做什么[s20]。这套框架的本质，是用结构化的问题替代直觉式的判断，用系统性思考替代拍脑袋决策。

字节跳动的"飞阅会"制度也是其管理哲学的具体体现。传统的企业会议往往是PPT汇报——大量时间花在制作和演示PPT上，真正讨论问题的时间反而很少。飞阅会要求会前阅读材料，与会者直接在文档中插入问题和建议，会议时间主要用于讨论分歧点而非听取汇报。这种制度减少了"表演性汇报"，提高了决策效率，也体现了"信息流优于权力流"的理念[s20]。

张一鸣的管理风格在国际上引起了广泛关注。海外科技媒体将他概括为四个关键词：数据驱动决策（data-driven decision-making）、扁平化实验文化（flat and experimental culture）、全球野心（global ambition）、刻意低调（deliberate personal modesty）[s52]。他的工程师导向、低ego、数据优先的风格与马云的魅力型叙事形成鲜明对比，更接近马化腾的产品导向，但有更强的测量意识和全球野心[s52]。Wall Street Journal注意到他卸任CEO后退出了董事会，形成了"所有权与日常管理分离"的模式[s51]。

张一鸣甚至将"延迟满足感"应用到了情绪管理上。他不允许自己因为短期的成功而过度兴奋，也不允许自己因为短期的挫折而过度沮丧。他反复训练自己如何从沮丧快速恢复到正常状态。经过反复实验，他得出自己的最佳工作状态是"在轻度喜悦和轻度沮丧之间"——既不过度亢奋也不过度消沉[s31]。这种近乎"算法化"的情绪管理，是延迟满足感在个人层面的极致应用。

字节范中的"坦诚清晰"原则，是张一鸣管理哲学的另一个重要体现。在字节跳动内部，"反对向上管理"、"敢当面表达真实想法"是被鼓励的行为[s19]。张一鸣自己身体力行——他说话直接、不绕弯子，内部信中常用"我觉得""实话说"开头，很少用修辞和情绪化表达[s4]。

"人才密度"是张一鸣管理思想中的另一个核心概念。如果团队中每个人的能力都足够强，就不需要复杂的流程来协调和管理。因此，字节跳动用市场最高薪酬吸引优秀人才——虽然人力成本高，但培训成本低、管理成本低[s50]。这种"高薪+高密度+低流程"的模式，与传统企业形成鲜明对比。

张一鸣对"公司也是产品"理念的实践，在字节跳动的组织架构演进中得到了充分体现。从早期的单一产品到多产品矩阵，再到2021年后的"板块化"组织——字节跳动不断用"产品经理"的思维来优化组织结构。每一次架构调整，都是一次"产品迭代"。

张一鸣的"十二问框架"是他面对创业和管理问题时的默认思考工具[s20]。这套框架的核心思想是：不要被表面现象迷惑，要追问到问题的本质。"真正的问题是什么？"——这是第一个也是最重要的一个问题。"背后发生了什么结构性变化？"——这要求管理者具备宏观视野。"是否用短期收益牺牲长期价值？"——这直接对应了延迟满足感的核心理念。

张一鸣的管理风格在国际上引起了广泛关注。海外科技媒体将他概括为四个关键词：数据驱动决策、扁平化实验文化、全球野心、刻意低调[s52]。他的工程师导向、低ego、数据优先的风格与马云的魅力型叙事形成鲜明对比，更接近马化腾的产品导向，但有更强的测量意识和全球野心[s52]。

值得注意的是，张一鸣极少在公开场合发言。他不参加达沃斯论坛，不接受电视采访，不做公开演讲（除了字节跳动内部年会）。这种"刻意低调"不是内向或害羞，而是一种价值观选择——他认为创始人的个人品牌不应该凌驾于公司品牌之上。这与马云形成了鲜明对比，也体现了字节跳动"公司也是产品"理念的另一面：CEO最重要的"产品"是公司本身，而不是自己的公众形象。

**English:** Zhang Yiming's management philosophy is in a class of its own in the Chinese internet world. It is not a narrative full of personal charisma but a calm, rational operating system that could almost be described in code. Its core concepts include delayed gratification, "Context, not Control," the company as a product, information flow over power flow, and the ordinary mind.

Delayed gratification is Zhang's central life philosophy. It derives from M. Scott Peck's The Road Less Traveled—a book he read during his half-year at Microsoft and which later became one of the works that influenced him most deeply.[s31][s3] The book's central insight: "The postponement involved in delayed gratification means resisting the weaknesses of human nature, giving up the pursuit and enjoyment of temporary ease, and reordering the sequence of pleasure and pain in life."[s31]

Zhang defines it this way: do not take immediate feedback as the sole criterion for decision; stretch the time horizon from three months to three years when making judgments. Be on guard against destroying long-term value for the sake of short-term revenue, profit, fundraising, valuation or public opinion. What is worth long-term investment includes talent, product, technology, user experience, organizational capability, brand, learning ability, corporate culture and infrastructure.[s3][s19][s20]

In a 2016 interview with Caijing magazine he said: "Half of the problems in many people's lives arise from this one cause—the absence of delayed gratification. The essence of delayed gratification is overcoming human weakness, and the point of overcoming weakness is greater freedom."[s31] "The trait I most appreciate in myself is delayed gratification, and the greatest delayed gratification is intellectual."[s31] He even holds that "people whose capacity for delayed gratification is not on the same order of magnitude simply cannot communicate effectively."[s31]

This philosophy has been verified again and again across his entrepreneurial career—refusing the temptation of a giant's acquisition, looking at the gap rather than celebrating in victory, treating a starting-salary difference of a few thousand yuan as negligible, selling his house to commit body and soul to the startup, and in 2026 refusing the shortcut of distillation—each time delayed gratification is practiced in a different dimension.[s31]

"Context, not Control" is his management methodology. Inspired by Hayek's theory of the "fatal conceit" of reason, in April 2017 he gave a talk at Source Code Capital's annual meeting titled "A CEO Must Avoid the Conceit of Reason."[s3][s19] The core idea: managers should not tell employees exactly how to take every step, but should provide sufficient context and then let the people closest to the problem make the decisions.[s3][s19][s20] "A CEO must avoid the conceit of reason. Top-down grand strategy is almost always a disaster. Replace ever-more-complex control systems with excellent people and high-quality information."[s19][s20]

In ByteDance's practice, any employee can see Zhang Yiming's OKRs directly on the internal IM. At the bimonthly "CEO face-to-face," Zhang publicly scores his own OKRs item by item. The "fly-read meeting" (feiyuehui) system requires reading materials in advance and inserting questions and suggestions directly into the documents.[s20]

Develop a company as a product: an entrepreneur makes two products at once—the product for users and the product for employees (the company itself). The CEO is the product manager of both. The employee ROI formula: (functional return plus psychological return) divided by (functional input plus psychological input). The three standards for talent: exceptional learning ability, broad horizons, and a mature mind.[s50]

The "ordinary mind" (pingchangxin) was the theme of ByteDance's ninth-anniversary speech in 2021. "Most people are ordinary, but ordinary people can do extraordinary things." Stay rational; do not be taken hostage by emotion. "Mediocrity has gravity; you need escape velocity."[s4][s19] "Have you ever had this happen in a game—a game with a hundred levels, and when you reach level ninety-nine your hand starts to shake?"[s4]

Zhang's personal character has profoundly shaped ByteDance's cultural tone. Extremely low-key, he rarely attends industry conferences. He admits: "I'm not very good at socializing. I prefer to study the principles of organizations and markets in order to manage less, and I like being on my own—surfing the internet, reading, listening to music, and zoning out."[s4] He dresses plainly, in a T-shirt and slippers. When he speaks, his tone and pace show almost no emotional fluctuation. "His most intense expression of emotion is no more than 'How on earth did things turn out this way?'"[s31] His optimal working state is "between mild elation and mild dejection."[s31] One senior executive assessed him as "genuinely working like a machine."[s31]

Early on at Kuxun, "after finishing my own work, whenever I could help solve a problem for most of my colleagues, I did."[s31] ByteStyle—pursue the extreme, be grounded and courageous, be open and humble, be candid and clear, always be day-one. In June 2022, "always be day-one" was elevated to first place.[s35][s7] At the seventh-anniversary gathering in 2019 he proposed a "grounded romanticism": "Polished refinement is not romance; rough grandeur is romance. Vitality is romance; facing the future is romance; embracing uncertainty is romance; keeping possibilities open is romance."[s3][s4] His status line read "likes to zone out"—"not emptying the mind, but thinking on my own about very boundless ideas that few people discuss."[s4]

Zhang's "twelve-question framework" is his default thinking tool when facing entrepreneurial and managerial problems. The framework includes: What is the real problem? What are the facts, the assumptions and the wishes, respectively? What structural change has taken place behind it? What is the user's true fundamental need? Are we mistaking activity for result? Are we sacrificing long-term value for short-term gain? Are we short on talent yet trying to add process? Is information reaching the people who actually make the decisions? If the judgment is wrong, how can we detect and roll back fastest? Looking back three years from now, what is most worth doing now?[s20] The essence of the framework is to replace intuitive judgment with structured questions, and to replace gut-calls with systematic thinking.

ByteDance's "fly-read meeting" system is another concrete expression of his management philosophy. Traditional corporate meetings tend to be PowerPoint reports—large amounts of time spent making and presenting slides, with very little time actually discussing problems. The fly-read meeting requires reading materials in advance; participants insert questions and suggestions directly into the document, and meeting time goes mainly to debating points of disagreement rather than listening to reports. The system cuts down "performative reporting," raises decision efficiency, and embodies the idea that "information flow is superior to power flow."[s20]

Zhang's management style has attracted wide attention internationally. Overseas technology media sum him up in four keywords: data-driven decision-making, a flat and experimental culture, global ambition, and deliberate personal modesty.[s52] His engineer-oriented, low-ego, data-first style stands in sharp contrast to Jack Ma's charismatic narrative; it is closer to Pony Ma's product orientation, but with a stronger measurement consciousness and greater global ambition.[s52] The Wall Street Journal noted that after stepping down as CEO he also left the board of directors, forming a pattern of "separation of ownership from day-to-day management."[s51]

Zhang even applied "delayed gratification" to emotional self-management. He does not allow himself to become over-excited by short-term success, nor over-dejected by short-term setbacks. He repeatedly trained himself to recover quickly from dejection to a normal state. Through repeated experimentation he found his optimal working state to be "between mild elation and mild dejection"—neither over-hyped nor over-brought-down.[s31] This near-"algorithmic" emotional management is delayed gratification applied to its extreme at the personal level.

The "candid and clear" principle of ByteStyle is another important expression of Zhang's management philosophy. Inside ByteDance, "managing upward is discouraged" and "daring to express real views face to face" are encouraged behaviors.[s19] Zhang practices what he preaches—he speaks directly, without beating around the bush; his internal letters often open with "I think" or "frankly," and rarely use rhetoric or emotional expression.[s4]

"Talent density" is another core concept in Zhang's management thought. If everyone on a team is strong enough, no complex processes are needed to coordinate and manage. Hence ByteDance uses top-of-market pay to attract excellent talent—although labor costs are high, training costs and management costs are low.[s50] This model of "high pay plus high density plus low process" stands in sharp contrast to traditional enterprises.

Zhang's practice of the "company as a product" idea is fully reflected in the evolution of ByteDance's organizational structure. From a single product early on, to a multi-product matrix, to the "business-unit" organization after 2021—ByteDance has continually used "product manager" thinking to optimize its structure. Every structural adjustment is a "product iteration."

Zhang's "twelve-question framework" is his default thinking tool when facing entrepreneurial and managerial problems.[s20] The core idea of the framework is: do not be fooled by surface phenomena; press on to the essence of the problem. "What is the real problem?"—the first and most important question. "What structural change has taken place behind it?"—this demands that managers have macro vision. "Are we sacrificing long-term value for short-term gain?"—this corresponds directly to the core idea of delayed gratification.

Zhang's management style has attracted wide attention internationally. Overseas technology media sum him up in four keywords: data-driven decision-making, a flat and experimental culture, global ambition, and deliberate personal modesty.[s52] His engineer-oriented, low-ego, data-first style stands in sharp contrast to Jack Ma's charismatic narrative; it is closer to Pony Ma's product orientation, but with a stronger measurement consciousness and greater global ambition.[s52]

It is worth noting that Zhang very rarely speaks in public. He does not attend Davos, does not give television interviews, and does not make public speeches (except at ByteDance's internal annual gatherings). This "deliberate modesty" is not introversion or shyness but a values-based choice—he believes the founder's personal brand should not tower over the company brand. It stands in sharp contrast to Jack Ma, and it embodies another face of ByteDance's "company as a product" idea: the CEO's most important "product" is the company itself, not his own public image.

### 8. 慈善与社会责任：芳梅基金与反哺家乡 / Philanthropy and Social Responsibility: The Fangmei Foundation and Giving Back to the Homeland

张一鸣的慈善事业带着鲜明的个人印记——低调、务实、聚焦教育，以及对家乡福建龙岩的深厚情感。芳梅公益基金的"芳梅"二字，分别取自他外婆和奶奶的名字——这个命名本身就是一种"来处意识"的表达[s29][s30]。在追逐财富的中国互联网界，张一鸣选择用家族女性长辈的名字来命名自己最重要的慈善项目，这比任何演讲都更有力量。

芳梅公益基金是张一鸣最重要的慈善载体，累计规模达10亿元。2021年6月，芳梅教育发展基金在龙岩市慈善总会成立，首次捐赠5亿元[s29][s30]。2021年12月，划拨2亿元成立中信信托·2021芳梅教育慈善信托，永续支持奖教金发放[s30]。2023年5月追加捐赠2亿元，总规模达7亿元[s30][s38]。2026年7月再捐3亿元（2亿教育+1亿医疗），总规模达10亿元，升级为芳梅公益基金[s29][s30]。

芳梅基金成果："芳梅奖教金"累计向超千名教师发放奖金超5000万元；"芳梅讲堂"累计参与超6000人次；累计培训教师超5万人次[s29][s30]。

南开大学系列捐赠累计超3.1亿元。2019年10月捐赠1亿元设立"南开大学创新基金"（种子基金，单笔不超300万元），另捐赠1000万元给人才建设基金[s2][s36]。2024年8月与梁汝波共同捐赠2亿元设立"南开大学大数学基金"[s39]。

疫情医务救助：2020年1月个人捐赠1亿元，基金总额达4.03亿元，资助2003名抗疫一线医务人员[s37]。其他：永定一中捐款1000万元[s2]。2025年10月与俞勇共同发起知春创新中心[s42]。

2025年9月获"全国十大慈善家"称号[s1]。累计公开捐赠超16亿元，加上芳梅基金总承诺超26亿元。Bloomberg报道他"建造了世界上领先的内容分发平台之一，然后在三十多岁时离开了CEO职位"[s28]。WSJ注意到他形成了"所有权与日常管理分离"的模式[s51]。在慈善领域，他同样展现出这种风格——芳梅基金用外婆和奶奶的名字命名，南开创新基金帮助年轻创业者——这些慈善行为不是功成名就后的点缀，而是从一开始就写在他人生代码里的程序。

张一鸣的慈善理念与他的管理哲学一脉相承——务实、系统、长期主义。芳梅基金的"信托制"设计，确保了基金的永续运作。2021年12月划拨2亿元成立中信信托·2021芳梅教育慈善信托，永续支持奖教金发放[s30]。这种"不花光、持续投"的模式，与字节跳动"延迟满足感"的经营哲学如出一辙。

值得注意的是，张一鸣的慈善捐赠高度聚焦教育领域。从南开大学创新基金（支持创新创业）、大数学基金（支持基础学科），到芳梅基金（支持家乡教育），到知春创新中心（培养16-18岁创新人才），几乎每一笔捐赠都指向"人的成长"。这种对教育的执念，或许源于他自己的经历——一个从闽西山村走出来的客家人，深知教育是改变命运的阶梯。

2025年10月在上海徐汇开业的知春创新中心，是张一鸣慈善事业的最新探索。该项目由他与上海交大ACM班创始人俞勇共同发起，面向16-18岁年轻人探索创新人才培养[s42]。这个项目的定位很有意思——不是传统的奖学金或助学金，而是"创新人才培养"。这与张一鸣本人的成长经历高度吻合：一个从小热爱阅读、喜欢思考的少年，如果能更早地接触到创新思维和方法论，将会释放出怎样的潜力？知春创新中心或许就是他对这个问题的回答。

截至目前，张一鸣的累计公开捐赠超过16亿元，加上芳梅公益基金的10亿元，慈善承诺总额超过26亿元。对于一位1983年出生、年仅43岁的创业者来说，这个规模的慈善承诺在中国互联网界名列前茅。但与金额同样重要的是他的慈善理念——不是功成名就后的点缀，而是从创业之初就写入人生代码的长期承诺。

张一鸣的慈善理念可以用三个关键词概括：教育优先、家乡情怀、系统设计。

教育优先：从南开大学的创新基金和大数学基金，到芳梅基金的教育板块，到知春创新中心的创新人才培养[s36][s39][s42]，几乎每一笔捐赠都指向教育。这种对教育的执念，源于他自己的成长经历——一个通过教育改变命运的闽西山村孩子。

家乡情怀：芳梅基金的所有投入都流向福建龙岩[s29][s30]。"芳梅奖教金"累计向超千名教师发放奖金超5000万元，累计培训教师超5万人次[s29][s30]。

系统设计：芳梅基金采用"信托制"，确保永续运作[s30]。这不是一次性的捐赠，而是建立一个能够自我持续的系统。这种思维方式与字节跳动的管理哲学一脉相承。

从更宏观的角度看，张一鸣的慈善行为反映了一代中国互联网企业家的价值转变。早期互联网企业家的慈善往往集中在灾害救助（如汶川地震捐款）或大型公益项目（如希望工程）。而张一鸣这一代企业家的慈善更加系统化、长期化、聚焦化——不是一次性的捐赠，而是建立可持续的慈善体系；不限于扶贫救灾，而是聚焦于教育创新和人才培养。

芳梅基金的命名选择——用外婆和奶奶的名字——在中国互联网企业家的慈善项目中极为罕见。大多数慈善基金以企业家本人、家人（配偶/子女）或抽象概念命名。张一鸣选择用家族女性长辈的名字，既是对家族历史的致敬，也体现了客家人重视家族传承的文化特质。这种"不忘来处"的品格，与他"延迟满足感"的管理哲学形成了呼应——两者都体现了对长期价值的追求和对根源的尊重。

**English:** Zhang Yiming's philanthropy carries a distinctly personal imprint—low-key, pragmatic, focused on education, and suffused with deep feeling for his hometown of Longyan, Fujian. The two characters of "Fangmei," the name of the Fangmei Public Welfare Fund, are taken respectively from the names of his maternal grandmother and his paternal grandmother—the naming itself is an expression of "awareness of where one comes from."[s29][s30] In a Chinese internet world obsessed with the pursuit of wealth, Zhang chose to name his most important charitable project after the elder women of his family, and that choice carries more force than any speech.

The Fangmei Public Welfare Fund is Zhang's most important philanthropic vehicle, with a cumulative scale of 1 billion yuan. In June 2021, the Fangmei Education Development Fund was established under the Longyan Charity Federation, with an initial donation of 500 million yuan.[s29][s30] In December 2021, 200 million yuan was allocated to establish the CITIC Trust · 2021 Fangmei Education Charitable Trust, supporting the payment of teacher-award grants in perpetuity.[s30] In May 2023 a further 200 million yuan was donated, bringing the total to 700 million yuan.[s30][s38] In July 2026 he gave another 300 million yuan (200 million for education, 100 million for healthcare), bringing the total to 1 billion yuan and upgrading it into the Fangmei Public Welfare Fund.[s29][s30]

Fangmei Fund results: the "Fangmei teacher-award grants" have cumulatively paid more than 50 million yuan in bonuses to over a thousand teachers; the "Fangmei lectures" have drawn more than 6,000 cumulative participants; and more than 50,000 teachers have cumulatively received training.[s29][s30]

His series of donations to Nankai University totals more than 310 million yuan. In October 2019 he donated 100 million yuan to establish the "Nankai University Innovation Fund" (a seed fund with individual grants not exceeding 3 million yuan), and a further 10 million yuan to the talent-development fund.[s2][s36] In August 2024, together with Liang Rubo, he donated 200 million yuan to establish the "Nankai University Mathematics Fund."[s39]

Pandemic medical relief: in January 2020 he personally donated 100 million yuan, bringing the fund's total to 403 million yuan and supporting 2,003 frontline anti-epidemic medical workers.[s37] Other giving: 10 million yuan to Yongding No. 1 Middle School.[s2] In October 2025, together with Yu Yong, he co-founded the Zhichun Innovation Center.[s42]

In September 2025 he received the title of one of China's "Top Ten Philanthropists."[s1] His cumulative public donations exceed 1.6 billion yuan, and total committed giving including the Fangmei fund surpasses 2.6 billion yuan. Bloomberg reported that he "built one of the world's most powerful content-distribution platforms, then left the CEO job in his thirties."[s28] The Wall Street Journal noted the pattern he formed of "separation of ownership from day-to-day management."[s51] In philanthropy he displays the same style—the Fangmei fund named after his grandmothers, the Nankai innovation fund helping young entrepreneurs—these charitable acts are not garnish added after fame and fortune, but a program written into the code of his life from the very beginning.

Zhang's philanthropic philosophy is of a piece with his management philosophy—pragmatic, systematic, long-termist. The Fangmei fund's "trust structure" design ensures the fund's perpetual operation. In December 2021, 200 million yuan was allocated to establish the CITIC Trust · 2021 Fangmei Education Charitable Trust, supporting teacher-award grants in perpetuity.[s30] This model of "don't spend it all, keep investing" is of one mind with ByteDance's operating philosophy of delayed gratification.

It is worth noting that Zhang's charitable giving is highly concentrated on education. From the Nankai University Innovation Fund (supporting innovation and entrepreneurship) and the Mathematics Fund (supporting basic disciplines), to the Fangmei fund (supporting education in his hometown), to the Zhichun Innovation Center (cultivating innovative talent aged 16-18), nearly every donation points to "the growth of people." This obsession with education may spring from his own experience—a Hakka boy who walked out of a mountain village in western Fujian, knowing full well that education is the ladder that changes fate.

The Zhichun Innovation Center, which opened in Xuhui, Shanghai in October 2025, is the newest exploration in Zhang's philanthropy. The project was co-founded by him and Professor Yu Yong, founder of the ACM class at Shanghai Jiao Tong University, aimed at exploring the cultivation of innovative talent among young people aged 16 to 18.[s42] The positioning of the project is interesting—it is not a traditional scholarship or bursary but "innovative-talent cultivation." This fits closely with Zhang's own growth story: if a boy who loved reading and thinking from childhood could have been exposed to innovative thinking and methodology earlier, what potential might have been unleashed? The Zhichun Innovation Center may be his answer to that question.

To date, Zhang's cumulative public donations exceed 1.6 billion yuan, and including the 1 billion yuan of the Fangmei Public Welfare Fund, his total charitable commitments surpass 2.6 billion yuan. For an entrepreneur born in 1983, still only in his early forties, this scale of charitable commitment ranks at the top of the Chinese internet world. But as important as the sums is his philanthropic philosophy—not garnish after fame and fortune, but a long-term commitment written into the code of life from the very start of entrepreneurship.

Zhang's philanthropic philosophy can be summed up in three keywords: education first, attachment to hometown, and systematic design.

Education first: from Nankai University's innovation fund and mathematics fund, to the Fangmei fund's education arm, to the Zhichun Innovation Center's cultivation of innovative talent,[s36][s39][s42] nearly every donation points to education. This obsession with education springs from his own growth story—a child of a mountain village in western Fujian whose fate was changed by education.

Attachment to hometown: all of the Fangmei fund's investment flows to Longyan, Fujian.[s29][s30] The "Fangmei teacher-award grants" have cumulatively paid more than 50 million yuan in bonuses to over a thousand teachers, and more than 50,000 teachers have cumulatively been trained.[s29][s30]

Systematic design: the Fangmei fund uses a "trust structure" to ensure perpetual operation.[s30] This is not a one-off donation but the building of a self-sustaining system. This way of thinking is of one mind with ByteDance's management philosophy.

From a more macro angle, Zhang's charitable actions reflect a value shift among a generation of Chinese internet entrepreneurs. The philanthropy of earlier internet entrepreneurs often concentrated on disaster relief (such as donations after the Wenchuan earthquake) or large public-welfare projects (such as Project Hope). The philanthropy of Zhang's generation is more systematic, longer-term and more focused—not one-off donations but the building of sustainable charitable systems; not limited to poverty relief and disaster aid but focused on educational innovation and talent cultivation.

The naming choice for the Fangmei fund—using the names of his maternal grandmother and paternal grandmother—is extremely rare among Chinese internet entrepreneurs' charitable projects. Most charitable funds are named after the entrepreneur himself, his family (spouse or children), or abstract concepts. Zhang's choice of the elder women of his family is both a tribute to family history and an expression of the Hakka cultural trait of valuing family continuity. This character of "not forgetting where one comes from" echoes his management philosophy of delayed gratification—both embody the pursuit of long-term value and respect for one's roots.

## 卷尾 Editorial Conclusion

回望张一鸣二十年的创业之路，他始终在做同一件事——提升信息分发与组织的效率。从酷讯的火车票小程序到今日头条的推荐引擎，从抖音的全屏沉浸式界面到TikTok的全球化运营，从Seed团队的自研路线到拒绝蒸馏的长期主义判断，他的每一步都在用算法思维重新定义边界[s1][s7][s18][s48]。

2021年卸任CEO时，他说"以十年为期，为公司创造更多可能"[s4]。三年后的今天，这句话正在被验证——豆包从一款AI聊天应用成长为月活近4亿的超级入口，日均Token调用量达180万亿；TikTok从生死存亡的边缘走入正常的监管运营框架，超2亿美国用户得以继续使用；字节跳动以超过1500亿美元的年营收成为全球规模领先的未上市科技公司[s4][s12][s14][s46]。

但张一鸣留给中国商业世界的，远不止一家公司或一款产品。他证明了三件事：第一，来自中国的互联网公司可以在全球市场具有真正的竞争力——TikTok是首个在全球范围内获得广泛用户基础的中国互联网产品；第二，算法可以重塑数十亿人消费信息和娱乐的方式——"张一鸣的遗产在于将全球数字媒体重新定位为基于算法的兴趣分发，迫使每个主要平台围绕字节跳动普及的格式进行重组"[s52]；第三，一个不爱社交、穿着拖鞋、喜欢发呆的工程师，可以用延迟满足感和火星视角，走完最波澜壮阔的商业征途。

他在字节跳动7周年大会上说："精致的文艺不是浪漫，粗糙的宏大是浪漫。晒情怀故意感动别人不是浪漫，独立思考穿越喧嚣是浪漫。有生命力是浪漫，面向未来是浪漫，拥抱不确定性是浪漫，保持可能性是浪漫。"[s4] 这段话，或许是对他二十年创业历程最好的注解。

芳梅公益基金用外婆和奶奶的名字命名[s29]。南开大学的创新基金用他的方式帮助年轻创业者[s36]。Seed团队里一千多名年轻工程师正在他每月亲自主持的复盘会上接受考验[s41]。在更宏大的尺度上，张一鸣的故事才刚刚翻到新的章节——AI新十年，全球化新阶段，慈善新起点。正如他引用《爱丽丝梦游仙境》的那句话："有时候早餐前，我已经相信六件看似不可能的事会发生了。"[s4]

在TikTok美国方案落地后的2026年8月，张一鸣在Seed全员会上罕见现身。他说"字节跳动不会把蒸馏当作提升AI模型能力的捷径"——"沿着别人的路走，最多只能逼近，很难超越"[s18][s48]。这句话既是技术路线的选择，也是人生哲学的宣言。从2006年的火车票小程序到2026年的AI大模型，二十年过去了，那个在深夜写代码的23岁工程师，依然在用最朴素的方式回答同一个问题：如何让信息更高效地到达需要它的人手中。只不过这个问题的尺度，从一张火车票变成了一整个星球。

Looking back across Zhang Yiming's two decades of entrepreneurship, he has always been doing the same thing—improving the efficiency of information distribution and of organizations. From the train-ticket script at Kuxun to Toutiao's recommendation engine, from Douyin's full-screen immersive interface to TikTok's global operations, from the Seed team's in-house research path to the long-termist judgment of refusing to distill other firms' models, every step has used algorithmic thinking to redraw boundaries.[s1][s7][s18][s48]

When he stepped down as CEO in 2021, he said he wanted to "work in ten-year horizons and create more possibilities for the company."[s4] Today, years later, that statement is being borne out: Doubao has grown from an AI chat app into a super-entrance with 382 million monthly active users and 180 trillion daily token calls; TikTok has moved from the brink of extinction into a normal, regulated operating framework, allowing more than 200 million American users to keep using it; and ByteDance, with annual revenue exceeding US$150 billion, has become the world's largest unlisted technology company.[s4][s12][s14][s46]

But what Zhang leaves to Chinese business is far more than one company or one product. He proved three things. First, an internet company from China can be genuinely competitive in global markets—TikTok is the first Chinese internet product to amass a broad user base worldwide. Second, algorithms can reshape how billions of people consume information and entertainment—"Zhang's legacy is the reorientation of global digital media around algorithmic interest distribution, forcing every major platform to reorganize itself around the formats ByteDance popularized."[s52] Third, an engineer who dislikes socializing, wears slippers and likes to zone out can, armed with delayed gratification and a view from Mars, complete the most sweeping business journey of his generation.

At ByteDance's seventh-anniversary gathering he said: "Polished refinement is not romance; rough grandeur is romance. Parading nostalgia to deliberately move others is not romance; thinking independently through the noise is romance. Vitality is romance; facing the future is romance; embracing uncertainty is romance; keeping possibilities open is romance."[s4] Those words are perhaps the finest annotation of his twenty-year entrepreneurial journey.

The Fangmei Public Welfare Fund is named after his maternal grandmother and his paternal grandmother.[s29] The innovation fund at Nankai University helps young entrepreneurs in his own way.[s36] More than a thousand young engineers on the Seed team are being tested at the monthly review meetings he personally chairs.[s41] On a grander scale, Zhang's story has only just turned to a new chapter—the new AI decade, a new phase of globalization, a fresh starting point for philanthropy. As he put it, quoting Alice in Wonderland: "Why, sometimes I've believed as many as six impossible things before breakfast."[s4]

In August 2026, after the TikTok U.S. deal had landed, Zhang made a rare appearance at a Seed all-hands meeting. He said that "ByteDance will not treat distillation as a shortcut to improving our AI models"—"If you follow in others' footsteps, the most you can do is draw close; it is very hard to overtake."[s18][s48] The sentence is both a choice of technical route and a declaration of life philosophy. From the train-ticket script of 2006 to the large AI models of 2026, twenty years have passed, and that 23-year-old engineer writing code late at night is still answering the same question in the plainest possible way: how to get information more efficiently to the people who need it. Only the scale of the question has changed—from a single train ticket to an entire planet.

## 金句 Pull Quote

> 很多人人生中一半的问题都是这个原因造成的——没有延迟满足感。延迟满足感的本质是克服人性弱点，而克服弱点，是为了更多的自由。

> Half of the problems in many people's lives stem from this one cause—the absence of delayed gratification. The essence of delayed gratification is overcoming human weakness, and the point of overcoming that weakness is greater freedom.

## 履历时间线 Career Timeline

- **1983-04** 出生于福建龙岩永定 / Born in Yongding, Longyan, Fujian
  - 1983年4月出生于福建省龙岩市永定区的一个事业单位家庭。父亲原是市科委工作人员，后去东莞开办电子产品加工厂；母亲是护士。客家人。家庭氛围宽松，父亲在他四五岁时就为其订阅儿童阅读周刊培养阅读习惯，父母常聊国外技术与产品话题，给予较早的商业与创新启蒙[s1][s2][s25]。
  - EN: In April 1983 he was born into a public-institution family in Yongding District, Longyan City, Fujian Province. His father had worked at the municipal science and technology commission before going to Dongguan to open an electronics-processing factory; his mother was a nurse. He is of Hakka heritage. The family atmosphere was relaxed: when he was four or five his father subscribed him to a children's weekly reading magazine to cultivate a reading habit, and his parents often discussed foreign technology and products, giving him an early apprenticeship in business and innovation.[s1][s2][s25]
- **2001** 考入南开大学 / Admitted to Nankai University
  - 考入南开大学微电子专业。大学期间不打游戏、不看碟，自称"道德状元郎"，大部分时间用于写代码和帮人修电脑。后转至软件工程专业[s2][s3][s25]。
  - EN: He was admitted to Nankai University to study microelectronics. During university he neither played video games nor watched movies, calling himself a "paragon of morality," and spent most of his time writing code and fixing people's computers. He later switched to software engineering.[s2][s3][s25]
- **2005** 毕业，首次创业失利 / Graduates; first venture fails
  - 从南开大学软件工程专业毕业。组成3人团队开发IAM协同办公系统，因市场定位失误首次创业失利。总结："技术不是价值，技术解决的'真实需求'才是。"[s2][s32]
  - EN: He graduated from Nankai University with a degree in software engineering. He put together a three-person team to develop the IAM collaborative office system, but a misjudgment of market positioning ended his first startup in failure. His takeaway: "Technology is not value; the 'real need' that technology solves is."[s2][s32]
- **2006-02** 加入酷讯 / Joins Kuxun
  - 加入旅游搜索网站酷讯，成为第一个工程师（工号003）。在酷讯做火车票搜索时，为自己写了个定时监控余票的小程序——"当有符合我需要的信息出现时，应该告诉我"。这个朴素想法后来演变为今日头条的核心逻辑：信息找人，而非人找信息。一年后管理40多人，担任技术委员会主席。工作不分你我，"做完自己的工作后，对于大部分同事的问题，只要我能帮助解决，我都去做"[s1][s2][s3][s31]。
  - EN: He joined the travel-search website Kuxun as its first engineer (employee number 003). While building train-ticket search at Kuxun, he wrote himself a small program that monitored leftover tickets on a timer—"When information that matches my needs appears, it should tell me." That simple idea later evolved into Toutiao's core logic: information finds people, rather than people seeking information. A year later he managed more than 40 people and chaired the technology committee. He drew no line between his own work and others': "After finishing my own work, whenever I could help solve a problem for most of my colleagues, I did."[s1][s2][s3][s31]
- **2008** 加入微软，大量阅读 / Joins Microsoft; reads widely
  - 加入微软。想去大公司"看一下"管理方式。然而在微软每天只工作三四个小时，感觉"太过无聊"。但微软的半年并非虚度——利用大量时间阅读了《高效能人士的七个习惯》《活法》《少有人走的路》《紧迫感》《团队协作的五个障碍》《基业长青》等书籍。其中《少有人走的路》中"延迟满足感"概念成为他此后最核心的人生哲学[s32]。
  - EN: He joined Microsoft, wanting to "take a look" at how a large company was managed. Yet at Microsoft he worked only three or four hours a day and felt it was "far too boring." But the half-year was far from wasted—he used the abundant time to read books including The 7 Habits of Highly Effective People, Living (Ikigai), The Road Less Traveled, A Sense of Urgency, The Five Dysfunctions of a Team and Built to Last. Above all, the concept of "delayed gratification" in The Road Less Traveled became his central life philosophy thereafter.[s32]
- **2008-下半年** 加入王兴团队 / Joins Wang Xing's team
  - 以技术合伙人身份加入王兴团队（饭否/海内网），负责搜索、消息分发、热词挖掘等后端系统。把Alexa排名前一千的网站都看了一遍，发现"积极推荐的阅读应用"很少[s32][s33]。
  - EN: He joined Wang Xing's team (Fanfou / Hainei) as a technical partner, responsible for back-end systems including search, message distribution and hot-word mining. He went through all of the Alexa top-1,000 websites and found that "reading apps that actively recommend" were scarce.[s32][s33]
- **2009-10** 创办九九房 / Founds 99fang
  - 创办垂直房产搜索引擎九九房担任CEO。6个月推出5款移动应用、150万用户，成为房产类应用第一名。但判断天花板不够高——"九九房有些小了"[s2][s3][s32]。
  - EN: He founded the vertical property-search engine 99fang and served as CEO. Within six months he launched five mobile apps and reached 1.5 million users, becoming the number-one property category app. But he judged the ceiling was not high enough—"99fang is a bit small."[s2][s3][s32]
- **2012-03** 创办字节跳动 / Founds ByteDance
  - 在北京知春路锦秋家园民居里创办字节跳动，团队不到10人。研发、财务、设计分别在不同卧室办公，会议室只有五平米。张一鸣卖掉自住房屋，全家租房度日。一个月见了30多个投资人，曾把嗓子都说失声了。SIG海纳亚洲王琼基于信任投了200万美元——"餐巾纸融资"故事由此而来[s3][s7][s31][s32][s34]。 张一鸣卖掉自住房屋，全家租房度日，拒绝了BAT的高薪邀约[s31]。
  - EN: He founded ByteDance in a residential apartment in Jinqiu Jiayuan on Beijing's Zhichun Road, with a team of fewer than ten. Engineering, finance and design worked in separate bedrooms, and the meeting room was only five square meters. Zhang sold the apartment he lived in and moved his family into rented housing. He met more than 30 investors in a month, at one point talking until he lost his voice. Joan Wang of SIG Asia Investments, on the strength of her trust in him, invested US$2 million—the origin of the "napkin fundraising" story.[s3][s7][s31][s32][s34] Zhang sold his own home, moved his family into rented housing, and turned down lucrative offers from the BAT trio.[s31]
- **2012-08** 今日头条上线 / Toutiao launches
  - 今日头条正式上线。核心是基于机器学习的个性化推荐引擎，"不生产内容，靠算法分发"。第一版推荐引擎是张一鸣自己查资料写出来的。从上线到1000万用户仅用90天。创立不到一年，某巨头抛出高价收购，张一鸣纠结一周后拒绝——认为资源是兴奋剂会遏制内生力量、战略会变形、被迫站队[s1][s2][s7][s31][s32]。
  - EN: Toutiao formally launched. At its core was a machine-learning-based personalized recommendation engine: "We don't produce content; we distribute it through algorithms." The first version of the recommendation engine was written by Zhang himself after looking things up. It took only 90 days to go from launch to 10 million users. Less than a year after founding, an internet giant offered a high-price acquisition; Zhang agonized for a week and refused—judging that resources are a stimulant that would suppress endogenous strength, distort strategy, and force him to take sides.[s1][s2][s7][s31][s32]
- **2015-08** 启动国际化 / Launches internationalization
  - 推出今日头条海外版TopBuzz，首站美国。同年喊出全球化"小目标"：3年内超过一半用户来自海外[s9][s10]。
  - EN: He released TopBuzz, the overseas version of Toutiao, with the United States as its first stop. The same year he set out a "small goal" for globalization: within three years, more than half of users would come from overseas.[s9][s10]
- **2016-09** 抖音前身上线 / Douyin's predecessor launches
  - 抖音前身A.me正式上线，创始团队最初仅7人，上线初期日活不到1万。12月更名为"抖音短视频"——名字是A/B测试的结果[s6][s8]。
  - EN: A.me, Douyin's predecessor, formally launched; the founding team initially had just seven people, and at launch its daily active users were under 10,000. In December it was renamed "Douyin Short Video"—the name was the result of an A/B test.[s6][s8]
- **2017-11** 收购Musical.ly / Acquires Musical.ly
  - 以约8-10亿美元收购北美短视频社区Musical.ly，后与TikTok合并。据报道张一鸣凌晨3点开会："买下它，这是我们登陆北美的D-Day"[s2][s6][s7][s9]。
  - EN: ByteDance acquired the North American short-video community Musical.ly for roughly US$0.8-1 billion, later merging it with TikTok. Zhang reportedly called a meeting at three in the morning: "Buy it. This is our D-Day for landing in North America."[s2][s6][s7][s9]
- **2019-04** 入选Time 100 / Named to the Time 100
  - 入选《时代》杂志"2019年度全球百位最具影响力人物"榜单[s2][s10]。
  - EN: He was named to Time magazine's list of the "100 Most Influential People in the World for 2019."[s2][s10]
- **2019-10** 捐赠南开1亿元 / Donates 100 million yuan to Nankai
  - 回到母校南开大学，捐赠1亿元设立"南开大学创新基金"，另捐赠1000万元给南开人才建设基金[s2][s36]。
  - EN: He returned to his alma mater, Nankai University, donating 100 million yuan to establish the "Nankai University Innovation Fund" and a further 10 million yuan to the Nankai talent-development fund.[s2][s36]
- **2020-08** TikTok美国风波 / TikTok U.S. turmoil
  - 美国行政令要求强制剥离TikTok美国业务。张一鸣发内部信："不要在乎短期损誉，耐心做好正确的事情；要有火星视角，格局大，EGO小。"TikTok起诉美国政府违法。此后经历多年法律诉讼、国会立法、最高法院裁定，最终在2026年1月通过合资公司方案落地解决[s10][s11]。
  - EN: A U.S. executive order demanded forced divestiture of TikTok's U.S. business. Zhang issued an internal letter: "Don't worry about short-term praise or blame; patiently do the right thing. Keep the view from Mars—big on perspective, small on ego." TikTok sued the U.S. government as unlawful. Years of litigation, congressional legislation and a Supreme Court ruling followed, ultimately resolved in January 2026 through a joint-venture arrangement.[s10][s11]
- **2021-05-20** 卸任CEO / Steps down as CEO
  - 发布内部全员信宣布卸任字节跳动CEO。表示"感觉过去几年很大程度都在吃老本"，希望"脱离开CEO的工作，能够相对专注学习知识，系统思考，研究新事物，动手尝试和体验，以十年为期，为公司创造更多可能"。聚焦远景战略、企业文化和社会责任。联合创始人梁汝波接任[s4][s5]。
  - EN: He issued an all-hands internal letter announcing he would step down as ByteDance's CEO. He said he "felt that in recent years I have to a large extent been living off past gains," and hoped that "freed from the CEO's job, I can focus relatively on learning knowledge, thinking systematically, studying new things, and getting hands-on to try and experience things—working in ten-year horizons to create more possibilities for the company." He focused on long-term strategy, corporate culture and social responsibility. Co-founder Liang Rubo succeeded him.[s4][s5]
- **2021-06** 芳梅基金成立 / Fangmei Fund established
  - 芳梅教育发展基金在龙岩市慈善总会成立，首次捐赠5亿元。基金名取自外婆和奶奶的名字[s29][s30]。
  - EN: The Fangmei Education Development Fund was established under the Longyan Charity Federation, with an initial donation of 500 million yuan. The fund's name is taken from the names of his maternal grandmother and paternal grandmother.[s29][s30]
- **2024-10** 首次成为中国首富 / Becomes China's richest person for the first time
  - 胡润百富榜发布，张一鸣以3500亿元首次成为中国首富[s1][s14]。
  - EN: On the release of the Hurun Rich List, Zhang Yiming became China's richest person for the first time, with wealth of 350 billion yuan.[s1][s14]
- **2025-02** 吴永辉加入字节 / Wu Yonghui joins ByteDance
  - 前Google DeepMind研究副总裁吴永辉加入字节，出任Seed基础研究负责人，推动AI团队重组为Edge/Focus/Base三大虚拟团队[s18][s41]。
  - EN: Wu Yonghui, a former vice president of research at Google DeepMind, joined ByteDance as head of foundational research for Seed, driving a reorganization of the AI team into three virtual teams: Edge, Focus and Base.[s18][s41]
- **2026-01-23** TikTok美国方案落地 / TikTok U.S. deal lands
  - TikTok美国数据安全合资公司正式成立，超2亿美国用户可继续使用。字节跳动保留19.9%股权（最大单一股东）和算法知识产权。甲骨文、银湖、MGX各持15%。采用"双实体切割"模式：USDS负责数据安全敏感业务，字节全资公司负责广告电商等商业板块。Adam Presser任CEO，7人董事会美国籍占多数[s12][s44][s45]。 这种"双实体切割"模式为全球跨国企业合规提供了新参考模板[s44]。
  - EN: The TikTok U.S. data-security joint venture was formally established, allowing more than 200 million American users to keep using the app. ByteDance retained a 19.9% stake (the largest single shareholder) and the algorithm's intellectual property; Oracle, Silver Lake and MGX each held 15%. A "two-entity split" model was adopted: USDS handles data-security-sensitive business, while a ByteDance wholly owned company runs commercial lines such as advertising and e-commerce. Adam Presser served as CEO, with U.S. nationals holding the majority of the seven-member board. This "two-entity split" model offers a new reference template for global multinational compliance.[s44][s12][s44][s45]
- **2026-03** 再度登顶中国首富 / Topples China's rich list again
  - 胡润全球富豪榜以5500亿元财富登顶中国首富，全球第25位。福布斯榜以693亿美元排全球第26位[s14][s15][s22][s27]。
  - EN: On the Hurun Global Rich List he topped the Chinese rankings with wealth of 550 billion yuan, placing 25th in the world. On the Forbes list he ranked 26th globally with US$69.3 billion.[s14][s15][s22][s27]
- **2026-06** 豆包月活3.82亿 / Doubao hits 382 million MAU
  - 豆包月活达3.82亿（QuestMobile数据），DAU突破2亿，日均Token调用量180万亿，两年增长超1500倍，成为中国领先AI原生应用。推出豆包2.1 Pro，在Coding、Agent、VLM三大方向跨越生产级质变点。上线三档付费订阅（68/200/500元/月）[s14][s17][s46][s47]。
  - EN: Doubao reached 382 million monthly active users (per QuestMobile), DAU broke 200 million, and daily token calls hit 180 trillion—a more than 1,500-fold increase in two years—making it China's largest AI-native application. It launched Doubao 2.1 Pro, crossing the production-grade inflection point in coding, agents and VLM. It went live with three paid subscription tiers (68/200/500 yuan per month).[s14][s17][s46][s47]
- **2026-07** 拒绝蒸馏与芳梅基金升级 / Refuses distillation; Fangmei Fund upgraded
  - 在Seed全员会上明确表示"不会把蒸馏当作提升AI模型能力的捷径"，Seed出台政策禁止蒸馏并追溯排查。随后向芳梅公益基金再捐3亿元（2亿教育+1亿医疗），基金总规模达10亿元并升级扩展至医疗等领域[s18][s29][s30][s48]。
  - EN: At a Seed all-hands meeting he stated clearly that the company "would not treat distillation as a shortcut to improving AI models," and Seed issued a policy banning distillation and conducting retroactive checks. He then donated another 300 million yuan to the Fangmei Public Welfare Fund (200 million for education, 100 million for healthcare), bringing the fund's total scale to 1 billion yuan and upgrading it to expand into healthcare and other fields.[s18][s29][s30][s48]
- **2026-08** TikTok正常化运营 / TikTok returns to normal operations
  - 美国白宫管理和预算办公室撤销2023年对TikTok在联邦政府设备上的使用禁令，司法部认定TikTok美国业务重组后已不再属于"受约束应用"。TikTok以4亿美元了结COPPA诉讼。从"生存危机"进入"监管框架内运营"阶段[s44][s49]。
  - EN: The White House Office of Management and Budget revoked the 2023 ban on TikTok's use on federal government devices, and the Department of Justice determined that, after the restructuring of TikTok's U.S. business, it no longer qualified as a "restricted application." TikTok settled the COPPA lawsuit for US$400 million. It moved from an "existential crisis" into a phase of "operating within a regulatory framework."[s44][s49]

## 常问问答 FAQ

**Q1: 张一鸣为什么创办字节跳动？**

A: 张一鸣的核心洞察是：信息的组织与分发有最大空间。传统新闻客户端依赖编辑推荐，但用户需求是"千人千面"。他认为"信息找人"比"人找信息"效率更高[s1][s7]。在王兴团队期间，他把Alexa排名前一千的网站都看了一遍，发现"积极推荐的阅读应用"很少，这个发现直接催生了今日头条[s32]。2012年3月，他在知春路一间民居里创办字节跳动，以算法推荐引擎重构内容行业。从2005年到2012年的五次创业经历——协同办公、酷讯搜索、微软、饭否、九九房——每一段都在为这个核心命题积累不同维度的能力[s2][s32]。 今日头条的推荐引擎基于机器学习，记录用户每一次行为——阅读时长、点击频率、滑动速度——精准度超同行2.3倍[s7]。

**Q1 (EN): Why did Zhang Yiming found ByteDance?**

A: Zhang Yiming's core insight was that the organization and distribution of information held the greatest opportunity of all. Traditional news apps relied on editors' recommendations, but user demand is "a thousand faces for a thousand people." He believed that "information finding people" was far more efficient than "people finding information."[s1][s7] While working on Wang Xing's team, he went through all of the Alexa top-1,000 websites and found that "reading apps that actively recommend" were scarce; that discovery directly gave birth to Toutiao.[s32] In March 2012 he founded ByteDance in a residential apartment on Zhichun Road, rebuilding the content industry with an algorithmic recommendation engine. Each of his five entrepreneurial stints between 2005 and 2012—collaborative office software, Kuxun search, Microsoft, Fanfou, and 99fang—accumulated a different dimension of capability around this central question.[s2][s32] Toutiao's recommendation engine is built on machine learning and records every user action—reading time, click frequency, swipe speed—with precision 2.3 times that of its peers.[s7]

**Q2: 张一鸣和梁汝波是什么关系？**

A: 梁汝波是张一鸣的南开大学室友，两人自2009年共同创办九九房起即成为长期创业伙伴[s2][s9]。2012年共同创办字节跳动，梁汝波先后负责产品研发、飞书和效率工程、人力资源和管理等工作。2021年5月张一鸣卸任CEO后，由梁汝波接任。张一鸣卸任后聚焦远景战略、企业文化和社会责任，但仍在AI战略上深度参与[s4][s5]。

**Q2 (EN): What is the relationship between Zhang Yiming and Liang Rubo?**

A: Liang Rubo was Zhang Yiming's roommate at Nankai University, and the two have been long-term entrepreneurial partners since co-founding 99fang together in 2009.[s2][s9] They co-founded ByteDance in 2012, after which Liang successively took charge of product R&D, Lark and efficiency engineering, human resources, and management. When Zhang stepped down as CEO in May 2021, Liang Rubo succeeded him. After stepping down, Zhang focused on long-term strategy, corporate culture and social responsibility, but he remains deeply involved in AI strategy.[s4][s5]

**Q3: 张一鸣为什么卸任字节跳动CEO？**

A: 2021年5月20日，张一鸣发布内部信宣布卸任。他表示"感觉过去几年很大程度都在吃老本"，希望"脱离开CEO的工作，能够相对专注学习知识，系统思考，研究新事物，动手尝试和体验，以十年为期，为公司创造更多可能"。卸任后聚焦远景战略、企业文化和社会责任。他说"我不是传统意义上的成熟管理者，我也不是很擅长社交，我更喜欢研究组织和市场原理来减少管理"[s4][s5]。

**Q3 (EN): Why did Zhang Yiming step down as ByteDance's CEO?**

A: On May 20, 2021, Zhang Yiming issued an internal letter announcing he would step down. He said he "felt that in recent years I have to a large extent been living off past gains," and that he hoped that "freed from the CEO's job, I can focus relatively on learning knowledge, thinking systematically, studying new things, and getting hands-on to try and experience things—working in ten-year horizons to create more possibilities for the company." After stepping down he focused on long-term strategy, corporate culture and social responsibility. He said: "I'm not a mature manager in the traditional sense, and I'm not very good at socializing; I prefer to study the principles of organizations and markets in order to manage less."[s4][s5]

**Q4: TikTok在美国经历了什么？最后怎么解决的？**

A: 从2019年CFIUS审查开始，TikTok在美国经历了七年风波。2020年美国行政令要求强制剥离，张一鸣发内部信提出"火星视角"。TikTok起诉美国政府。2024年"不卖就禁"法案通过，2025年1月最高法院裁定法案合宪，1月18-19日TikTok一度暂停在美服务后恢复[s11][s12]。2025年12月双方达成协议，成立TikTok美国数据安全合资公司。2026年1月方案落地：字节跳动保留19.9%股权（最大单一股东）和算法知识产权，甲骨文、银湖、MGX各持15%。采用"双实体切割"模式——USDS负责数据安全等敏感业务，字节全资公司负责广告电商等商业板块。7人董事会美国籍占多数。2026年8月联邦政府设备解禁，4亿美元COPPA诉讼和解，TikTok从"生存危机"进入"监管框架内运营"[s12][s44][s45][s49]。

**Q4 (EN): What did TikTok go through in the United States, and how was it finally resolved?**

A: Beginning with the CFIUS review in 2019, TikTok went through seven years of turmoil in the United States. In 2020 a U.S. executive order demanded forced divestiture, and Zhang issued an internal letter proposing the "view from Mars." TikTok sued the U.S. government. In 2024 the "sell-or-be-banned" act passed; in January 2025 the Supreme Court ruled the law constitutional, and on January 18-19 TikTok briefly suspended service in the U.S. before restoring it.[s11][s12] In December 2025 the two sides reached an agreement to establish the TikTok U.S. data-security joint venture. In January 2026 the arrangement landed: ByteDance retained a 19.9% stake (the largest single shareholder) and the algorithm's intellectual property, while Oracle, Silver Lake and MGX each held 15%. A "two-entity split" model was adopted—USDS handles sensitive business such as data security, while a ByteDance wholly owned company runs commercial lines such as advertising and e-commerce. U.S. nationals hold the majority of the seven-member board. In August 2026 the ban on federal government devices was lifted and the US$400 million COPPA settlement was reached, as TikTok moved from "existential crisis" to "operating within a regulatory framework."[s12][s44][s45][s49]

**Q5: 张一鸣的管理哲学是什么？**

A: 核心包括五大理念：①延迟满足感——不以短期反馈为标准，追求长期价值，"克服人性弱点，是为了更多的自由"[s31]；②Context, not Control——提供充分信息而非层层控制，让离问题最近的人做决策[s19][s20]；③公司也是产品（Develop a company as a product）——把组织当作不断迭代的产品来优化[s50]；④人才密度优先——用优秀人才替代复杂流程；⑤平常心——保持理性，不被情绪绑架，"平庸有重力，需要逃逸速度"[s4][s19]。 他还有一个"十二问框架"：真正的问题是什么？事实/假设/愿望分别是什么？背后发生了什么结构性变化？是否用短期收益牺牲长期价值？[s20]

**Q5 (EN): What is Zhang Yiming's management philosophy?**

A: At its core are five ideas: (1) Delayed gratification—do not take short-term feedback as the standard; pursue long-term value; "overcoming human weakness is for the sake of greater freedom."[s31] (2) Context, not Control—provide ample information rather than layer upon layer of control, and let the people closest to the problem make decisions.[s19][s20] (3) Develop a company as a product—treat the organization as a product to be iterated and improved continually.[s50] (4) Talent density first—replace complex processes with excellent people. (5) The ordinary mind—stay rational and do not be taken hostage by emotion; "mediocrity has gravity; you need escape velocity."[s4][s19] He also has a "twelve-question framework": What is the real problem? What are the facts, assumptions and wishes respectively? What structural change has taken place behind it? Are we sacrificing long-term value for short-term gain?[s20]

**Q6: 张一鸣在AI领域做了哪些布局？**

A: 2023年8月推出豆包APP[s16]。2024年下半年起每月亲自参与Seed团队复盘讨论。2024年化身"猎头"亲自监督挖AI人才[s16]。2025年引入前Google DeepMind研究副总裁吴永辉重组Seed团队，分设Edge（前沿AGI，3年考核）/Focus/Base三大虚拟团队，AI Lab并入Seed成为"公司级AI中枢"[s41]。2026年2月发布豆包2.0（据媒体报道1万亿参数）[s18]。6月豆包月活达3.82亿，DAU超2亿，日均Token 180万亿[s46]。7月明确"拒绝蒸馏"，坚持自研路线[s48]。飞书并入豆包，AI战略重心向ToB转移[s48]。2026年AI资本开支计划超2000亿元[s46]。

**Q6 (EN): What has Zhang Yiming laid out in the AI field?**

A: In August 2023 ByteDance launched the Doubao app.[s16] From the second half of 2024 Zhang personally took part every month in the Seed team's reviews and discussions. In 2024 he turned himself into a "headhunter," personally overseeing the poaching of AI talent.[s16] In 2025 he brought in Wu Yonghui, a former vice president of research at Google DeepMind, to restructure the Seed team, which was split into three virtual teams—Edge (frontier AGI, with a three-year assessment), Focus and Base—and AI Lab was merged into Seed to make it a "company-wide AI hub."[s41] In February 2026 Doubao 2.0 was released (reported at 1 trillion parameters).[s18] In June Doubao reached 382 million monthly active users, DAU over 200 million, and 180 trillion daily tokens.[s46] In July he stated clearly that he "refuses distillation," insisting on the in-house research route.[s48] Lark was merged into Doubao, shifting the AI strategic focus toward ToB.[s48] The 2026 AI capital-expenditure plan exceeds 200 billion yuan.[s46]

**Q7: 抖音是怎么诞生的？**

A: 抖音诞生于2016年，是字节内部的边缘项目，最初团队仅7人，产品形态受Musical.ly启发，上线初期日活不足1万。张一鸣在看到留存和时长数据提升后，决定加大资源投入，并引入头条推荐算法团队改造产品，使抖音从运营主导转向算法驱动。2016年12月更名为"抖音"——名字是A/B测试的结果。2017年通过赞助《中国有嘻哈》等营销实现爆发增长[s6][s8]。

**Q7 (EN): How was Douyin born?**

A: Douyin was born in 2016 as a marginal project inside ByteDance. The initial team had just seven people, the product form was inspired by Musical.ly, and at launch its daily active users were under 10,000. After Zhang Yiming saw retention and time-spent data improving, he decided to pour in more resources and brought in Toutiao's recommendation-algorithm team to remake the product, shifting Douyin from operations-led to algorithm-driven. In December 2016 it was renamed "Douyin"—the name was the result of an A/B test. In 2017, through marketing such as sponsoring The Rap of China, it achieved explosive growth.[s6][s8]

**Q8: 张一鸣做了哪些慈善？**

A: 累计公开捐赠超16亿元。主要包括：①芳梅公益基金累计10亿元（2021年5亿+2023年2亿+2026年3亿），支持福建龙岩教育和医疗；②南开大学系列约3.1亿元（创新基金1亿+人才基金1000万+大数学基金2亿）；③疫情医务救助个人捐赠1亿元；④永定一中1000万元；⑤知春创新中心（与俞勇共同发起）。2025年9月获"全国十大慈善家"称号[s1][s2][s29][s30][s36][s37]。

**Q8 (EN): What philanthropy has Zhang Yiming done?**

A: His cumulative public donations exceed 1.6 billion yuan. The main items: (1) the Fangmei Public Welfare Fund, cumulative 1 billion yuan (500 million in 2021 plus 200 million in 2023 plus 300 million in 2026), supporting education and healthcare in Longyan, Fujian; (2) the Nankai University series, about 310 million yuan (100 million innovation fund plus 10 million talent fund plus 200 million mathematics fund); (3) a personal donation of 100 million yuan for pandemic medical relief; (4) 10 million yuan to Yongding No. 1 Middle School; and (5) the Zhichun Innovation Center (co-founded with Yu Yong). In September 2025 he received the title of one of China's "Top Ten Philanthropists."[s1][s2][s29][s30][s36][s37]

**Q9: 张一鸣为什么离开酷讯？九九房为什么觉得"小了"？**

A: 离开酷讯原因：①内部在"垂直搜索还是通用搜索"上争执不下；②管理混乱，出现夸大数据、虚假汇报；③谷歌中国进入带来竞争压力；④想去大公司学习管理[s32][s33]。九九房觉得"小了"的原因：张一鸣判断房产搜索天花板不够高——"房产市场虽大，用户使用频次太低，哪怕做到第一，增长潜力也有限"。他向投资人王琼委婉表示"九九房有些小了"[s2][s32]。

**Q9 (EN): Why did Zhang Yiming leave Kuxun, and why did he feel 99fang was "too small"?**

A: Reasons for leaving Kuxun: (1) the company was deadlocked internally over whether to pursue vertical search or general search; (2) management was chaotic, with inflated figures and false reporting appearing; (3) Google China's entry brought competitive pressure; and (4) he wanted to go to a large company to learn management.[s32][s33] The reason he felt 99fang was "too small": Zhang judged that property search's ceiling was not high enough—"The property market may be large, but users' frequency of use is too low; even at number one, the growth potential is limited." He delicately told his investor Joan Wang that "99fang is a bit small."[s2][s32]

**Q10: 张一鸣现在有多少财富？**

A: 2026年胡润全球富豪榜以5500亿元位列中国首富、全球第25位[s14][s22]。2026年福布斯全球亿万富豪榜以693亿美元排全球第26位[s15][s27]。Bloomberg估算身家928亿美元（2026年6月），亚洲第二[s28]。持有字节跳动约21%-24%股权，通过同股不同权保留超50%投票权[s14][s28]。

**Q10 (EN): How much wealth does Zhang Yiming have now?**

A: On the 2026 Hurun Global Rich List he ranks as China's richest person with 550 billion yuan, 25th in the world.[s14][s22] On the 2026 Forbes World's Billionaires list he ranks 26th globally with US$69.3 billion.[s15][s27] Bloomberg estimates his net worth at US$92.8 billion (June 2026), the second-largest in Asia.[s28] He holds roughly 21%-24% of ByteDance's equity while retaining more than 50% of the voting rights through a dual-class share structure.[s14][s28]

## 来源清单 Sources

1. [s1] 腾讯百科（微信百科） · 张一鸣 — https://baike.weixin.qq.com/v64633047.htm
2. [s2] 百度百科 · 张一鸣 — https://m.baike.com/wiki/%E5%BC%A0%E4%B8%80%E9%B8%A3/21594640
3. [s3] 华商韬略·创业邦 · 张一鸣：算法时代的创业者 — https://m.cyzone.cn/article/570669
4. [s4] 21世纪经济报道·南方+ · 张一鸣卸任字节跳动CEO内部信 — https://static.nfapp.southcn.com/content/202105/20/c5288523.html
5. [s5] 钛媒体 · 张一鸣卸任字节跳动CEO — http://m.toutiao.com/group/6964191741847175694/
6. [s6] 抖音百科 · 抖音 — https://m.baike.com/wiki/%E6%8A%96%E9%9F%B3/543518
7. [s7] CSDN · 字节跳动深度复盘 — https://blog.csdn.net/qq_22182989/article/details/147363961
8. [s8] 乱翻书播客·荔枝FM · 抖音早期发展访谈 — https://lizhifm.com/voice/3178665668885721094
9. [s9] 创业邦 · 字节跳动出海史 — https://m.cyzone.cn/article/579964
10. [s10] 中国经济周刊 · 张一鸣：围剿与反围剿 — http://paper.people.com.cn/zgjjzk/html/2020-12/30/content_2027508.htm
11. [s11] 新京报·贝壳财经 · TikTok美国危机时间线 — https://m.bjnews.com.cn/detail/1737281675129255.html
12. [s12] 百度百科（更新版） · TikTok — https://m.baike.com/wiki/TikTok/22887690
13. [s13] Future Startup · The TikTok Story — https://futurestartup.com/2026/01/12/the-tiktok-story-and-the-lessons-for-builders/
14. [s14] 东方财富·BT数据通 · 2026胡润全球富豪榜与字节估值 — https://caifuhao.eastmoney.com/news/20260316083522015977180
15. [s15] 新浪财经·每日经济新闻 · 2026福布斯全球亿万富豪榜 — https://finance.sina.com.cn/jjxw/2026-03-11/doc-inhqqupq5688437.shtml
16. [s16] 财经天下周刊·澎湃新闻 · 张一鸣与AI大模型 — https://imgpai.thepaper.cn/pph/newpai/image/20250130/e7d1c523-7fd1-4fe3-83b1-e451e3b28d0f.jpg
17. [s17] 蓝鲸新闻·奇点研究社 · 豆包成为"新字节" — http://m.toutiao.com/group/7589469212624683558/
18. [s18] 界面新闻·伯虎财经 · 字节AI三年集权史 — https://www.jiemian.com/article/14927917.html
19. [s19] 人人都是产品经理 · 张一鸣管理哲学分析 — https://www.woshipm.com/it/6156960.html
20. [s20] 广告门 · 张一鸣创业与管理原则 — https://m.adquan.com/detail/0-362376
21. [s21] 虎嗅网 · 张一鸣管理风格 — https://m.huxiu.com/aibrief/9170.html
22. [s22] 胡润百富官网 · 胡润百富榜数据 — https://www.hurun.cn/
23. [s23] 白宫官网 · TikTok行政令 — https://www.whitehouse.gov/presidential-actions/2025/09/saving-tiktok-while-protecting-national-security/
24. [s24] 买购网 · 历年中国首富对比 — https://m.maigoo.com/news/576974.html
25. [s25] 腾讯新闻 · 张一鸣家世与成长 — https://news.qq.com/rain/a/20241102A03VB700
26. [s26] 环球时报·抖音黑板报 · 国籍辟谣 — http://m.toutiao.com/group/7488532369574543922/
27. [s27] 中新网·澎湃新闻 · 2026福布斯全球亿万富豪榜详细数据 — http://www.chinanews.com.cn/cj/2026/03-11/10585086.shtml
28. [s28] Bloomberg Billionaires Index（转引） · 张一鸣身家与亚洲排名 — https://www.asatunews.co.id/zhang-yiming-orang-terkaya-kedua-asia-bytedance
29. [s29] 光明网·龙岩市教育局 · 芳梅公益基金最新捐赠 — http://gongyi.gmw.cn/2026-07/21/content_38897981.htm
30. [s30] 人民网·龙岩市教育局 · 芳梅基金历次捐赠明细 — http://m.people.cn/n4/2023/0504/c1142-20574777.html
31. [s31] CSDN · 张一鸣：延迟满足感与创业细节 — https://blog.csdn.net/weixin_42869365/article/details/82997515
32. [s32] 卢松松博客 · 张一鸣前传：五次创业历程 — https://lusongsong.com/info/post/19017.html
33. [s33] 爱看财·ikanchai · 张一鸣酷讯经历 — https://m.ikanchai.com/pcarticle/3592
34. [s34] CSDN · 字节跳动8年创业史 — https://blog.csdn.net/lantian_123/article/details/104981153
35. [s35] 腾讯新闻 · 字节跳动12周年回顾 — https://view.inews.qq.com/k/20240313A056ZN00
36. [s36] 澎湃新闻·南开新闻网 · 南开大学创新基金捐赠 — https://m.thepaper.cn/newsDetail_forward_4658518
37. [s37] 南开大学发展委员会 · 疫情医务救助基金捐赠 — https://fzw.nankai.edu.cn/info/1003/1733.htm
38. [s38] ZAKER · 张一鸣再捐3亿 — http://app3.myzaker.com/news/article.php?f=Normal&pk=6a811d528e9f0941f75c8125
39. [s39] 天津南开大学教育基金会 · 南开大学大数学基金捐赠 — http://nkuef.nankai.edu.cn/
40. [s40] Richmond Journal of Law and Technology · TikTok Inc. v. Garland案法律分析 — https://jolt.richmond.edu/wp-content/uploads/sites/117/2026/03/2.-Sharp-Final.pdf
41. [s41] 晚点LatePost · 吴永辉接管Seed团队 — http://m.toutiao.com/group/7604770519342154290/
42. [s42] 上海市政府（转引） · 知春创新中心开业 — http://app3.myzaker.com/news/article.php?f=Normal&pk=6a811d528e9f0941f75c8125
43. [s43] 澎湃新闻 · 豆包商业化困境 — https://m.thepaper.cn/newsDetail_forward_33316397
44. [s44] 财经杂志·新浪财经 · TikTok美国方案与政府设备解禁 — https://finance.sina.com.cn/wm/2026-08-12/doc-inimzumq5288125.shtml
45. [s45] 广州日报 · TikTok美国方案落地 — http://news.dayoo.com/finance/202601/23/171077_54920062.htm
46. [s46] 钛媒体 · 豆包用户数据与Token消耗 — https://www.tmtpost.com/8040655.html
47. [s47] DoNews · 豆包大模型2.1 Pro发布 — https://www.donews.com/news/detail/4/6608950.html
48. [s48] 36氪英文版 · 张一鸣罕见公开亮相：拒绝蒸馏 — https://eu.36kr.com/en/p/3934315037572227
49. [s49] 观察者网·新浪财经 · TikTok 4亿美元COPPA诉讼和解 — https://finance.sina.com.cn/stock/t/2026-08-22/doc-inipetzr1204492.shtml
50. [s50] 混沌学园 · Develop a company as a product — https://www.hundun.cn/articles/7e74ebef4b867ddb.html
51. [s51] WSJ（转引） · 张一鸣卸任CEO分析 — https://texxr.com/966589/zhang-yiming-steps-down-as-bytedance-ceo
52. [s52] kurums.com · 张一鸣管理风格与遗产评价 — https://kurums.com/zhang-yiming-bytedance-leadership/

---
本档案机器版由亚洲封面人物官网自动生成：https://coverfigure.com ｜ 永久档案：https://coverfigure.com/acf/ACF-00-00139/geju ｜ 生成时间 Generated：2026-09-11 00:42