Part IThe Story
We'll Use His English Name
On a smoggy August morning in 2009, outside the China World Hotel in Beijing, a crowd of teenagers and twentysomethings was being organized into a fan club. They held placards studded with LEDs. A wrangler moving among them had one instruction that mattered, and he shouted it over the noise: "We'll use his English name, okay?" A black Mercedes-Benz pulled up. The door opened. A boyishly handsome executive, forty years old and weeks short of forty-one, stepped onto the pavement. The crowd did as it had been told.
"Ro-bin! Ro-bin!"
The man they were screaming for was born Li Yanhong, the fourth child and only son of two factory workers in a coal city in Shanxi province. The name the crowd was told to use he had carried home from America, where he spent most of the 1990s writing software for Dow Jones and Disney and inventing, in a New Jersey office, one of the foundational ideas of web search. A few hours after the Mercedes arrived, at the company's annual Baidu World conference, Robin Li told an audience of thousands that the search engine he had built a decade earlier was becoming China's gateway to the world of information. Then he walked across the stage surrounded by smiling children while "Twinkle, Twinkle, Little Star" played in Mandarin and soap bubbles drifted through the hall.
Forbes, which was there to watch, called it Olympic-style marketing and did not say whether the young people on the sidewalk had been paid. Two years later Vanity Fair, compiling its annual list of the New Establishment, settled the matter in a single dry line filed under the heading "Evidence of Possibly Megalomaniacal Behavior": "Li has hired 'fans' to mob him at public appearances."
It is worth staying with that wrangler for a moment, because the instruction he gave holds most of what is strange and interesting about the man. We'll use his English name. Not his Chinese name, in Beijing, at a Chinese company's conference, before a Chinese crowd. The English name, the American name, the name that signaled Stanford lecterns and Nasdaq tickers and Silicon Valley picnics. In 2009 that name was the brand. Li's company was then known, to the point of cliché, as "China's Google." It had a market value of $12.8 billion and 7,000 employees, and it handled 8 billion searches a month for a country of 338 million internet users. Li's personal stake was worth around $2 billion. The year before he had been the seventh-richest person in China.
In a no-frills conference room afterward, Li told Forbes the sentence he seemed to want remembered. "A lot of Chinese people have wondered if knowledge really means power in today's market economy," he said. "I think I've proven that it does." By the end of that year the company would move into a new headquarters designed to look like an enormous rectangular search box, a building whose shape asks a question it cannot answer by itself.
What follows is an attempt to read the man the way his own algorithm reads a webpage: not by what he declares about himself, but by what points to him, from where, and with how much weight.
By the Numbers
Robin Li and Baidu
1996Year Li built RankDex, ranking pages by inbound links, two years before Google's PageRank
$1.2MSeed capital that took Li and Eric Xu from Silicon Valley to a Beijing hotel room
354%First-day gain at Baidu's Nasdaq IPO on August 5, 2005, from $27 to $122
CN¥133.1BBaidu revenue in 2024
59%Li's voting control of Baidu, on 18% of the equity
6 millionDriverless Apollo Go robotaxi rides completed across 11 cities by April 2024
30M+Entries in Baidu Baike, the encyclopedia Li launched in 2006
The Dimmest Candlelight
The company's name comes from a poem eight centuries old. Xin Qiji, a Song dynasty poet, wrote a lyric about the Lantern Festival called "Green Jade Table," and its last line is among the most quoted in classical Chinese: Having searched for her a hundred times in the crowd, suddenly turning back, she is there in the dimmest candlelight. "A hundred times" is bai du. Baidu.
The choice tells you something about the man who made it. Silicon Valley named its search companies after cartoon exclamations and misspelled numbers: Yahoo!, Google. Li reached for a twelfth-century poem about longing, a poem whose point is that the thing you have been looking for everywhere was behind you the whole time, in the place with the least light. A page later maintained under the RankDex name glosses the name as "a persistent search for the ideal." That is the corporate reading. The poem is lonelier than that. It is about a man in a festival crowd, among thousands of lanterns, who finds what he wants only when he stops looking forward.
Yangquan, where Li was born on November 17, 1968, is not a place associated with poetry. It is a mining and industrial city in Shanxi, the coal province, and Li grew up there during the Cultural Revolution, the fourth of five children and the only boy, his parents both factory workers. The official biographies, in the way of official biographies, describe an ordinary family. One detail survives that does not fit the arc of the dutiful scholar. As a boy, the English-language entry in his company's encyclopedia records, Li "was fascinated by Chinese opera and was admitted to the Shanxi Yangquan Jin Opera Troupe." Jin opera is the regional form, loud and stylized and Shanxi to the bone. The entry adds that during high school Li "refocused on his 'main studies,'" the quotation marks gently signaling that the opera had been a detour from the road a boy like him was supposed to walk.
He walked it. He got into Yangquan First High School with the second-highest score on the entrance exam, took computer classes, and entered programming competitions across the city. In 1987 he sat the national university entrance examination, the gaokao, and posted the top score in Yangquan. He chose Peking University and a major then called library and information science, later renamed information management, which is to say he studied cataloguing: how collections are ordered, indexed, and retrieved. It is hard to think of a better apprenticeship for a man who would spend his life ranking things.
Peking University sits in Haidian, next to the stretch of Beijing called Zhongguancun, which in the late 1980s was an electronics market crowded with stalls and early hardware merchants. Speaking at the university's 120th anniversary in 2018, Li recalled what caught his eye there. "I started to pay attention to people with beepers in Zhongguancun," he said. "I began to learn that messages could instantly send information to the other side of the ocean." These, he said, "were the ideal starting points for my further studies and entrepreneurship." The image is small and exact: a student from a coal town watching men in a street market receive signals from across the Pacific, and working out that the gap between here and there could be closed by a device on a belt.
From his junior year, by his own account, he set his sights on the United States and on computer science. He graduated in 1991. Two decades later he and his wife, Ma Dongmin, who goes by Melissa in English, would give RMB 660 million, roughly $104 million, to the university together with Baidu, for a fund supporting research compatible with Baidu's AI work. The beepers of Zhongguancun turned out to be worth a great deal of money.
Link Analysis, or the Logic of Reputation
In the fall of 1991, Li Yanhong arrived at the State University of New York at Buffalo to work toward a doctorate in computer science. Buffalo is a long way from Yangquan, though the two cities have more in common than either would admit: both industrial, both cold, both built on heavy things pulled from the earth and both a little behind the times. After a year Li won a three-month internship at Matsushita, the Japanese electronics giant now called Panasonic, an experience he later described, in a sentence his company's encyclopedia preserves only up to its first clause, as having "played a crucial role in my" (the rest of the thought lies past the edge of the available excerpt). By 1994 he had decided against finishing the PhD. He left Buffalo with a master's degree.
That May he took a job at IDD Information Services, a New Jersey division of Dow Jones & Company, where he helped build software for the online edition of The Wall Street Journal, a real-time financial information system for a newspaper just beginning to imagine life on a screen. He also worked, on the side and then not so much on the side, on how search engines should decide what to show first.
The search engines of the mid-1990s mostly judged a page by its own contents: how often a word appeared, where, in what font. That made them easy to fool, because a page could simply say what it wanted to be found for. Li's insight was that a page's own words were testimony and testimony is cheap. What mattered was what other pages said about it, and specifically whether they linked to it. A hyperlink was a vote. A page that many others pointed to was probably important; a page pointed to by important pages was probably more important still. He called it "link analysis." In 1996 he built a site-scoring algorithm and a search engine around it, RankDex, the first search engine to use hyperlinks to measure the quality of the sites it indexed.
On February 5, 1997, he filed for a patent titled "Hypertext document retrieval system and method." It issued on July 6, 1999, as U.S. Patent 5,920,859, inventor Yanhong Li. In the July/August 1998 issue of IEEE Internet Computing, he published a paper with a title that reads now almost like a manifesto: "Toward a Qualitative Search Engine." When Larry Page filed his own patent for a link-based ranking method, the system that became PageRank and Google, it cited Li's work.
The priority dispute, if it is one, is less interesting than the worldview the algorithm encodes. Link analysis is a theory of reputation. It holds that you cannot know what something is by listening to it describe itself; you know it by who vouches for it, and by who vouches for them. It is an epistemology of networks and endorsements, which happens to be a fairly good description of how influence works in a one-party state, and of how a young man from a coal town without connections might eventually be judged by investors who had never heard of him. Li would live inside that logic for the rest of his career, as both its beneficiary and its subject.
He left IDD in June 1997 for Infoseek, one of the pioneering search companies, as a staff engineer. Among his accomplishments there was the picture-search function used by Go.com, the portal Disney built after taking a stake in Infoseek. It was the Disney stake that soured things. Disney's commitment to search faded, and its attention drifted toward content and the logic of a media company. Li's did not drift. He was, by every account, frustrated.
In the summer of 1998, at a picnic in Silicon Valley, a friend named Eric Xu introduced him to someone who might understand. Xu was 34, a biochemist with a doctorate and, more usefully, good contacts across the Valley, and he was making a documentary about the place. He had gathered some friends together for it and thought two search people might get along. One was Li, then 30, whom Xu's acquaintances remembered as shy and reserved. The other was John Wu, who ran the search engine team at Yahoo, the most important internet company in the world at the time and one that, Wu admitted later, had decided search was not where the action was. Yahoo would eventually outsource its search entirely to a small start-up called Google, a decision the New York Times would describe as a colossal corporate blunder.
Wu later moved to China himself, as chief technology officer at Alibaba.com, which gives his memory of the picnic a certain retrospective ruefulness.
— John Wu, former head of Yahoo's search engine team, recalling the 1998 picnic where he met LiThe people at Yahoo didn't think search was all that important, and so neither did I. But Robin, he seemed very determined to stick with it. And you have to admire what he accomplished.
The shy man at the picnic, quietly refusing to give up on the unfashionable thing. Hold onto that image. It will come back, in less flattering lighting.
The Third-Best Search Engineer in the World
There is a story, told in the Chinese book Robin Li, Baidu World, that has the tidy shape of a corporate parable and a few details odd enough to suggest it is at least partly true.
Li and Xu were raising seed money. One of the prospective investors was Scott Walchek of Integrity Partners. During the meeting Walchek asked the young engineer across from him a question he presumably already had an answer to: who were the top three people in the world in search engine technology? Li named two men, identified in the book only as Mr. X and Mr. Y, and then a third: William Chang, chief technology officer of Infoseek, his own boss's boss. He did not name himself.
Walchek excused himself, stepped out of the room, and phoned William Chang, whom he knew. Was this Robin Li really that good? Chang's answer, as the book tells it: in search engine technology, Robin Li ranks in the top three in the world.
Walchek came back in, according to the account, and raised his offer. Integrity Partners and the second investor would put in $1.2 million instead of $1 million, split fifty-fifty. The second investor, per the New York Times, was Peninsula Capital.
Read as parable, it is about modesty, the book's author attributing Li's omission to "Eastern humility." Read through the logic of RankDex, it is about something else entirely. Li did not rank himself. He let a link do the work. The investor checked the inbound reference, found it came from a highly ranked node, and repriced the asset upward by 20 percent within the span of a phone call. Whether Li understood this as a tactic or simply as manners, the result was the same: his first fundraise was decided by the algorithm he had patented.
The name William Chang will come up again, attached to a Baidu executive speaking at a conference in 2008. The sources do not say plainly whether it is the same man. It is that kind of story, full of names that circle back.
Walchek, for his part, was candid about what he had not understood. "We were skeptical about search," he told the Times years later, by then on Baidu's board. "But we weren't as smart as Robin. Robin said he had a unique opportunity to build a brand around search. And he was right."
With the money in hand, Li and Xu flew to Beijing. They set up in a hotel room overlooking the Peking University campus, the same campus where Li had once watched the men with beepers. Baidu was incorporated on January 18, 2000. That September, as the dot-com bubble began to burst in America, Draper Fisher Jurvetson and IDG Technology Venture put in another $10 million. "So it was," the Times wrote, "that on the eve of the Internet bubble bursting in the United States, Baidu took off in China."
The early business was not glamorous. "We didn't have a feasible business model, let's put it this way," Li said later. "When we started, the business model is to provide the backend search technology to the portal companies in China." They built a strong Chinese-language engine and signed up, in short order, nearly every portal of consequence: Sina, Sohu, NetEase, Yahoo! China. Baidu was the plumbing behind other people's storefronts, invisible and paid accordingly.
Recruiting was a matter of belief more than money. In a letter to staff written sixteen years later, in the worst week of his professional life, Li recalled those days with an intensity that cuts through even a rough translation. Google, he wrote, could pay three times what Baidu could. People stayed anyway. To find them, the company made posters pairing famous faces with short appeals: the writer Lu Xun for those who wanted to make search meaningful for Chinese people; the rocket scientist Qian Xuesen, who returned from America to found China's missile program, for engineers; Mao Zedong over a line asking, more or less, whether you would surrender or become stronger than your enemy. Even now, Li wrote, when he repeated those slogans to people, they got choked up.
Lu Xun, Qian Xuesen, Mao. A writer, a returnee, a revolutionary. It is a revealing trinity for a company founded by a man fresh off seven years in America with an English first name and a U.S. patent. The posters announce what Baidu was selling to its first employees: not stock options so much as a homecoming.
A Brand Built Around Search
Somewhere in 2000 or 2001, Li noticed what a company in Pasadena called Overture was doing. Overture sold advertising alongside search results and auctioned the slots, so that, in the Times's example, ads for dental clinics would appear beside results for "cavities." The insight was that a search query is the purest statement of commercial intent ever invented: someone typing "cavities" is, at some level, asking to be sold a dentist.
Li moved fast. In 2001, Baidu let advertisers bid for placement and pay only when a user clicked, ahead of Google's own move to that model. The RankDex-branded "about" page, whatever its provenance, states the philosophy plainly: unlike Lycos and AltaVista, "which charge by flat fee," the products were "performance based." In September 2001, Baidu stopped being solely a back-end supplier and launched its own site, Baidu.com, with a homepage that looked almost exactly like Google's: a logo, a box, a button.
Small and medium-size Chinese businesses loved pay-for-performance advertising. Traffic poured in. In 2003 the company added news and picture search. In 2004 it turned a profit, and Li, who had been president and chairman since the founding, took the title of chief executive. He began pushing for an initial public offering in the United States, insisting, the Times reported, that it would be "a huge branding event."
He was right about that too. Baidu listed on Nasdaq on August 5, 2005, priced at $27 a share and structured, as Chinese internet companies are, through a variable interest entity incorporated in the Cayman Islands. By the close of trading it was at $122, up 354 percent, the biggest opening day on Nasdaq since the dot-com peak. Baidu was suddenly worth $4 billion and Li's shares more than $900 million. Many analysts called the stock absurdly overvalued, and it eventually slid to as low as $44 before recovering.
Before the IPO there had been a road not taken. Google had invested $5 million in Baidu early on and, according to Vanity Fair, came close to buying the company outright before it listed. Imagine the alternate history: Robin Li as a Google executive, Baidu as Google China, a single global search company with a Mandarin front end. It did not happen, and the reasons it did not are partly commercial and partly about everything else this profile is about.
The other thing critics noticed at the IPO was the advertising. A search on Baidu for the word "cancer," the Times reported, turned up ads for hospitals that had paid for top placement rather than information about cancer itself. Google and Yahoo put their ads in a column on the right; Baidu blended them into the results. Remember the word "cancer." It is the hinge of this story, and Baidu would be shown it many times.
By the following year Google had arrived in force, planning to spend hundreds of millions of dollars in China. Its China president, Kai-fu Lee, framed the contest as a matter of integrity. "People want information and they want global information," he said. "We can't be bought." The line was a direct shot at Baidu's paid rankings, and it carried the confidence of a company that believed the better product would win.
Analysts were less sure. "The globally dominant U.S. Internet companies have failed to take the No. 1 market share position in any category," Jason Brueschke of Citigroup said of China. "And they came with more money and major brand names. And so there's something fundamentally different about this market." Richard Ji of Morgan Stanley offered a demographic explanation: 70 percent of China's internet users were under 30, "most of them are single, only children. They're looking for entertainment." Baidu's MP3 search, which linked freely to music files and drew copyright lawsuits, was one of the things they were looking for.
Li's response to all of this was flat, almost stubbornly simple. "At the end of the day," he said, "if a user finds relevant information, they'll come back." And: "Our traffic keeps increasing. We're now the No. 1 Web site in China."
In December 2007, Baidu became the first Chinese company included in the Nasdaq-100. The shy man from the picnic now had a seat in the index of the American companies he had once worked for.
The Literal Cost
Baidu's dominance had a precondition that its founder rarely discussed in moral terms. Chinese censors oversaw the site; the government regularly blocked Google and imposed strict rules on foreign internet companies. The Times put the arrangement plainly in 2006: "In exchange for letting censors oversee its Web site, Baidu has sealed its dominance with support from the Chinese government." Documents leaked in April 2009 from Baidu's internal monitoring and censorship department showed long lists of blocked websites and forbidden topics. China Digital Times would describe the company as the most active and restrictive censor in search.
Li's way of talking about this was revealing in what it emphasized. In an interview with Bloomberg TV in August 2010, he described compliance not as a moral dilemma but as an expense. "We have to spend a lot of resources to make sure our content and services abide by Chinese law," he said, "and [Internet companies based outside China] don't." Gady Epstein of Forbes picked out the nuance: Li had raised "the literal financial cost, as opposed to what Google considers the moral cost." The man who had built a theory of search around measurable signals was treating censorship as a line item. Foreign Policy summarized the stance more bluntly the same week: Li, it said, "is said to have very cosmopolitan instincts in some regards, but has made the calculation that business in China requires being utterly apolitical — in other words, doing what the government requires you to do."
Apolitical is a generous word for some of what he said in public. Fortune in 2011 quoted Li's description of what made the Chinese internet strong: "walking the path of socialism with Chinese characteristics is the well-spring of strength that will allow the Chinese Internet to continue its healthy and rapid development." Fortune's writer called this rhetoric that might "veer precariously close to pandering," then added, with a straight face, that Li's job might be the hardest in the world.
The reason it might be became clear that August, when CCTV, the state broadcaster, aired a 26-minute documentary attacking Baidu. Millions of viewers watched a Baidu employee help a man posing as the owner of an admittedly fake healthcare company buy an advertisement for a phony weight-loss pill. Speculation ran for weeks: was the state uneasy with Baidu's 80 percent share of search? Was CCTV, which was building a search engine of its own, simply acting commercially? It was not the first round. In 2008, after an earlier skirmish between CCTV and Baidu, the company sponsored CCTV's 2009 Spring Festival Gala and increased its advertising spending by 41 percent, much of it flowing straight to the broadcaster. By the fall of 2011 a member of the Politburo had paid Baidu a highly public visit at a technology exhibition, the customary signal that a company was back in favor. Duncan Clark, chairman of the Beijing consultancy BDA China and one of the most experienced foreign observers of Chinese technology, read the whole episode as "more about control and economic interests than anything political."
The pattern is worth stating without decoration. Baidu sells attention. CCTV sells attention. The state owns one and licenses the other. When they collide, the private company pays, sometimes literally, with Spring Festival sponsorship money.
Google had already made its choice. In 2010 it moved its Chinese search service to Hong Kong, effectively leaving the mainland market. Publicly, Baidu played down its windfall. Privately, the numbers spoke: Baidu's share of China's search advertising market rose to about 70 percent, up six points in a single quarter. That same year a hijacker, posing as the "Iranian Cyber Army," talked a Register.com support worker into changing Baidu's domain email address and redirected Baidu.com for four hours. Baidu sued for negligence; Register.com apologized and settled. Even the company's worst day of 2010 was a dispute with an American vendor.
By 2011, Forbes estimated Li's fortune at $9.4 billion, which made him the richest man in mainland China. From 2013 to 2018 he sat on the National Committee of the Chinese People's Political Consultative Conference, the advisory body where the party gathers its approved non-party notables; he would also serve on the next one. He was a vice chairman of the All-China Federation of Industry and Commerce. On December 18, 2018, the Central Committee of the Communist Party and the State Council awarded him the Reform Pioneer medal. His official biography, in Chinese and English, describes him as a "non-party personage," a man outside the party and in no meaningful sense outside the system.
It was also in 2011 that the novelist and racing driver Han Han wrote Li an open letter. Han Han had just come back from a paper mill in Shandong where he and his publisher, Shen Haobo, had watched more than a million printed copies of the second issue of his magazine, Chorus, go into a pulping furnace along with industrial waste, three hundred-odd tons of paper. A few million yuan, he wrote, might be a small sum to Li; to a publisher it was a year's work wiped out. The letter was about Baidu Wenku, the company's document-sharing service, where Chinese writers' books circulated free. Its sharpest move was its framing. Here was a writer who had just lost a fortune to forces he did not name, addressing the richest man in the country, whose platform was taking the rest.
An Encyclopedia With Locked Doors
In 2005 the Chinese government blocked Wikipedia. Around April 20, 2006, Baidu released its own encyclopedia, Baidu Baike, created by Li and modeled closely on the thing that had just disappeared. Within twenty days it had more than 300,000 registered users and more than 100,000 articles, more than the Chinese-language Wikipedia had accumulated. Two years later, at the World Wide Web Consortium's WWW2008 conference, a Baidu executive named William Chang gave the most succinct justification anyone has offered. "There is, in fact, no reason for China to use Wikipedia," he said. "It's very natural for China to make its own products."
Baike is the clearest window into how Li built things, because it shows every tendency at once: the speed, the scale, the substitution, the integration, the commercial instinct, and the political ceiling.
Speed and scale first. By 2025 Baike claimed more than 30 million entries and some 8 million editors, the largest Chinese-language encyclopedia anywhere. Integration: when a Baidu search matches a Baike entry, the link to that entry usually appears at or near the top. The rival encyclopedia Hudong complained about precisely this; in 2010 its chief, Pan Haidong, posted "Three Questions to Robin Li" on Weibo, challenging the ranking Baidu gave its own services. Baidu sued him and won. A court ordered Pan to pay Baidu 120,000 yuan.
Substitution has costs. In 2007 Florence Devouard, then chair of the Wikimedia Foundation, said of Baike: "They do not respect the license at all... That might be the biggest copyright violation we have. We have others." Chinese Wikipedia editors compiled lists of allegedly copied entries; Wikimedia declined to sue. And then, almost two decades later, the story turned on itself. In 2025 a Beijing court found Douyin Baike, ByteDance's encyclopedia (formerly Hudong, which ByteDance had bought), guilty of illegally copying 600,000 Baike entries and faking user data to make it look like ordinary editors had done it. Damages: 8 million yuan. The encyclopedia once called the biggest copyright violator on the internet had become a victim of copying, and a court upheld its claim.
The commercial instinct showed up in small details. Editors earned "gift points" for edits, redeemable for snacks and stationery and, at the high end, refrigerators, washing machines and smartphones, a reward system that one researcher argued encouraged volume over quality. By 2016 nearly 100,000 physical Baike QR codes had been placed at more than 200 locations across China, linking temples and buildings to their entries. In 2015 Baidu launched "Baike Youming," a paid service letting entrepreneurs and artists control their own biographies. Businesses complained of being steered toward a 6,000-yuan-a-year enterprise package to speed up corrections. The encyclopedia does not accept edits between 11 p.m. and 8:30 a.m. Beijing time, ostensibly for the health of its volunteers, plausibly for the convenience of its auditors.
Then the political ceiling. All editors register under their real names; administrators review every edit before it appears. As of 2013, a scholar reported, entries about senior Chinese leaders were fully locked, as were the entries about Baidu itself and its CEO, Robin Li. Articles on the 1989 Tiananmen Square protests did not exist. In 2009, users invented the "Baidu 10 Mythical Creatures," fictional animals whose names sounded like obscenities, most famously the Grass Mud Horse, cao ni ma, a near-homophone for a crude insult. The Grass Mud Horse became a mascot of evasion across the Chinese internet. It was born inside Li's encyclopedia.
Here is the shape of it. A man whose great invention was the idea that a page's value comes from what others say about it built the largest encyclopedia in his language, then arranged for the page about himself to be locked against what others might say.
A decade after Baike's launch, the same integrations had turned on the company's main product. In January 2019, Fang Kecheng, a communications scholar, published a viral post declaring the Baidu search engine dead. "Baidu.com is no longer a place for you to search for content on China's internet," he wrote, "but rather an internal search for Baidu content... It won't lead you to good quality food for the spirit from China's internet, but content hoarded at home, gone bad." Part of the blame belonged to rivals who refused to let Baidu index their walled gardens; part of it, critics said, belonged to Baidu, which kept surfacing low-quality posts from its own blogging platform, Baijiahao. The poem behind the company's name is about searching a crowd and finding the one person who matters. By 2019, the crowd had been fenced off and the search engine was mostly finding itself.
Thirty Days
Wei Zexi was a computer science student at Xidian University in Xi'an. In his teens he was diagnosed with synovial sarcoma, a rare cancer of the soft tissue. He and his family searched Baidu for treatment. Near the top of the results was the Second Hospital of the Beijing Armed Police Corps, promoting an experimental immunotherapy. The family spent around 200,000 yuan, about $31,000, on the treatment. It failed. Before he died, Wei publicly accused the hospital of misleading his family and criticized Baidu for selling medical search listings to the highest bidder. He died on April 12, 2016. He was 21.
His death spread across the Chinese internet in a way few corporate scandals had. One report claimed medical advertising made up as much as 30 percent of Baidu's ad revenue, much of it from for-profit hospitals in the so-called Putian network, founded by entrepreneurs from one region of Fujian province. On May 2, the Cyberspace Administration of China sent investigators to Baidu. Within a week, a joint task force concluded that paid listings had compromised the objectivity of Baidu's results, that the company favored high bidders and failed to label sponsored links. Regulators ordered it to stop ranking results by payment, to add risk warnings, and to limit paid results to less than 30 percent of each page. Baidu's shares fell 12 percent over the month.
It is worth pausing on the chronology, because the chronology is the indictment. In 2005, the Times noted that a Baidu search for "cancer" returned paid hospital ads instead of information. In 2009, Li told Forbes he would overhaul the ad system, under a project code-named Phoenix Nest, to separate paid results clearly from organic ones. "Even if this means we have to suffer for a few quarters," he said then, "I'm willing to make that bet — or sacrifice — to make sure we're moving in this direction. It's not an easy decision for a CEO. But if we imagine the company ten years down the road, it's an obvious one." In 2011, CCTV showed a Baidu employee helping a fake company buy an ad for a fake pill. In 2014, Baidu began selling moderator rights to forums on Tieba, its message-board network. In January 2016, a Tieba forum devoted to hemophilia was sold to a hospital, which removed the volunteer moderators and deleted posts complaining about the change. A forum for Minecraft went to a knockoff mobile game for, according to one headline, 9.3 million yuan.
Bai du. A hundred times. The same lesson, relearned.
Li's response came as a letter to all employees, leaked within hours, and it is the most revealing document he has ever produced, because in it the man who usually spoke in stiff, measured English about traffic and execution sounds like someone who had been awake all night. He wrote of questions he had been asking himself "in the dead of night": "Why so many everyday Baidu users no longer love us? Where is the problem in the end?"
— Robin Li, internal letter to Baidu staff, May 2016I see senior engineers trapped in a conflict of interest between commercial interest and user experience, and often compromising the latter.
He did not blame regulators, or rivals, or the hospital. He blamed the dashboard. "The management and employees' obsession of KPI," he wrote, "has twisted our values... and distanced ourselves from users." Departments quarreled over performance indicators; engineers compromised to hit commercial targets. Then came the sentence that became the headline in every outlet from People's Daily to the BBC: "If we lose the support of users, we lose hold of our values, and Baidu will truly go bankrupt in just 30 days."
He compared the reforms to cutting one's own wrist. He proposed a "one-vote veto" for user experience, overseen by a separate division with authority to kill any product that violated it. He announced a 1 billion yuan fund to compensate users harmed by misleading promotions. He told the BBC's readers, through the letter, that Baidu must put "values before profits."
The line about thirty days is a strange one when you look at it. Baidu was not going bankrupt in thirty days; it held most of China's search market and its users had nowhere obvious to go. The line was not a forecast. It was a scream meant to be heard on the floor, an attempt by the founder to reach past a layer of managers who had learned to read his priorities off a spreadsheet. A company organized around KPIs had optimized for exactly what its CEO had measured, and the CEO was now discovering, sixteen years in, that the measure had swallowed the mission.
Four years earlier, in November 2012, he had tried a different metaphor in a different memo. The company, he wrote then, had lost its "wolf spirit" and turned into something more like a panda, comfortable and complacent, grazing on its search monopoly. The commentator Doug Young, a former Reuters journalist, compared Baidu to Yahoo, the pioneer that coasted. In 2014 Baidu's search share had slid to 56.3 percent against Qihoo 360's 29 percent. The memos kept coming because the problems kept coming, and the problems kept coming partly because the memos did not change the incentives.
Baidu's net income fell from 33.7 billion yuan in 2015 to 11.6 billion in 2016.
The Main Course
If the Wei Zexi scandal marked the end of something, it also accelerated a turn already underway. On May 16, 2014, Baidu had named Andrew Ng, the Stanford researcher who had helped found Google's deep-learning effort, as chief scientist, to run Baidu Research out of new labs in Silicon Valley and Beijing. The company had already been folding deep learning into Phoenix Nest, the ad system, which is a small irony worth noting: the first serious AI at Baidu was put to work on the advertising.
In January 2017, Li overhauled his management team to refocus the company on artificial intelligence, and his wife, Melissa Ma, who had worked alongside him in the early years, returned to Baidu as a special assistant. That April he told CNN that AI was the new electricity. The same month Baidu announced Apollo, an open self-driving platform offering vehicle, hardware, software, and cloud components to the auto industry; by September there was a $1.5 billion fund to back up to 100 autonomous-driving projects over three years, alongside partnerships with Bosch and Continental. In November 2017 the first Baidu World under the new theme was titled "Bring AI to Life."
The state followed, or led; with Chinese technology companies it is often hard to tell which. In 2018 Beijing designated Baidu one of its national "AI champions." Time put Li on the cover of its Asia edition on January 19, 2018, under the word "innovator," the first Chinese internet entrepreneur to appear there. In September, at the Bloomberg Global Business Forum in New York, seated beside Christine Lagarde, he gave a line that sounded rehearsed and was clearly meant to travel.
— Robin Li, Bloomberg Global Business Forum, September 2018If the internet was the appetizer, then AI is the main course. The internet changed a lot of our daily lives, but did not have much impact on the 2B industries. I think AI will change that.
Baidu's market value had touched $99 billion that May. In October, it became the first Chinese member of the Partnership on AI, the U.S.-based ethics consortium. That July Li had gone home, in a sense, giving a lecture on AI to nearly 500 students and faculty at Shanxi University in Taiyuan, which made him honorary president of its new big data academy.
Then came winter, which he named himself. In a New Year letter to staff on January 2, 2019, Li wrote that "winter is coming," and that economic restructuring was "as cold and real as winter to every company." He reached for an image from the Analects: "Only when the year grows cold do we see the qualities of the pine and the cypress." Revenue for 2018 had crossed 100 billion yuan. In April, BuzzFeed News reported that Android apps from DO Global, the overseas utility-app business Baidu had spun out the year before, had been secretly clicking ads in the background on users' phones; on April 26 Google banned DO Global and more than 100 of its apps from the Play Store. In May, Baidu reported its first quarterly loss since its IPO, 327 million yuan, about $47.5 million. Li wrote again. "For senior managers, saying 'I have tried' is not enough and we need to make sure we win in the battlegrounds that we must win." Net income for 2019 came in at 2.1 billion yuan, a sixteenth of what it had been four years earlier.
There is a pattern in the letters, and it has its own rhythm: a crisis, a memo, a metaphor drawn from nature (wolves, pandas, pines, winter), a demand for a change of spirit. The man who invented a ranking system based on external evidence keeps turning, under pressure, to the most internal tool a chief executive has: exhortation. It is not clear the letters ever worked. It is clear he kept writing them.
What he also kept doing was spending. Look at the composition of the revenue. In 2015, online marketing, the search ads and their relatives, brought in 64.0 billion yuan; everything else brought in 2.3 billion. By 2024, online marketing was 78.6 billion and everything else was 54.6 billion, a category that included iQIYI's video business, which Li chairs, and the cloud and AI work he had staked the company on. The ad engine had barely grown in a decade. The second business had grown more than twentyfold. It was paid for by the first.
In a speech at Baidu Create in early 2023, Li explained his theory of innovation with a thought experiment. Hand a blind person a scrambled Rubik's cube and ask them to solve it at one turn per second, he said, and it would take 137 billion years. Tell them after each turn whether they were getting closer or further away, and it would take two and a half minutes. "Innovation drives growth," he said, "feedback drives innovation." He predicted that smart transport would end the need for car-purchase and traffic restrictions in Chinese cities by 2027, and that traffic jams would be essentially gone by 2032.
The feedback arrived soon enough. On March 16, 2023, Baidu unveiled ERNIE Bot, its answer to ChatGPT. Li would later call it the first ChatGPT-like chatbot from any public company in the world. The planned public webcast was switched, at short notice, to a closed-door session with test customers; the public release came that August, Ernie 4.0 in October. By late 2024 Li was citing 300 million users. That September, Time put him on its inaugural list of the 100 most influential people in AI, the only Chinese entrepreneur included. At the World Artificial Intelligence Conference in Shanghai in July 2024, asked about the brutal discounting among Chinese model makers, he said what a veteran of the Chinese internet might be expected to say: "I think a price war is almost inevitable. Having been in the Chinese internet industry for so long, we are very familiar with price wars. It has come even earlier than I expected." He argued for closed-source models against open ones, and estimated that artificial general intelligence was more than ten years away, maybe twenty or more.
In January 2025 a small Chinese lab called DeepSeek released an open model that shook the industry. On March 16, 2025, exactly two years after ERNIE Bot's debut, Baidu released ERNIE 4.5 and a reasoning model, ERNIE X1, which it said matched DeepSeek's R1 at half the price. The champion of closed source was now pricing against an open-source upstart.
The cars did better. By April 2024 Apollo Go, Baidu's robotaxi service, had completed six million driverless rides in eleven cities, with more than 400 driverless vehicles in Wuhan alone. At the World Governments Summit in Dubai in February 2025, Li claimed the fleet had driven more than 130 million kilometers autonomously without a serious accident, that its insurance claims rate was one-fourteenth of an ordinary car's, that robotaxis were "at least 10 times safer" than human drivers. In late 2024 Baidu won Hong Kong's first pilot license for autonomous driving. In July 2025 it announced a partnership with Uber to deploy Apollo Go outside the U.S. and mainland China, the company whose founder had once named his engine after a Song poem now riding on the rails of the most American of platforms.
The machinery kept coming. In January 2026 Baidu filed to list its chip unit, Kunlunxin, in Hong Kong. In June 2026 the U.S. Department of Defense put Baidu on its list of "Chinese military companies." Revenue, meanwhile, had begun to slide quarter after quarter as advertising shrank faster than AI grew, and Reuters reported in 2026 that Li had vowed to return ERNIE to the frontier after months without a major upgrade.
His fortune had gone the same direction as the share price. A South China Morning Post profile in 2017 put his net worth at $20 billion. By May 2025, Forbes estimated $5.5 billion. He still holds about 18 percent of Baidu's equity and 59 percent of its votes, enough to keep making long bets whether or not the market agrees with them.
In May 2024 his head of communications, a vice president named Qu Jing, posted videos on Douyin defending punishing hours. In one she said she had asked an employee to take a fifty-day business trip during the pandemic, and observed that she was not the employee's mother. The backlash was immediate. She apologized and reportedly left the company. Baidu's Hong Kong shares fell 2.17 percent. A company founded on recruiting posters of Lu Xun and Mao, on engineers who stayed for a third of Google's pay, had produced a spokesperson who said aloud what the KPI letter had warned about in 2016.
Mr. AI
In October 2015, on the night Baidu's travel site Qunar announced its merger with its rival Ctrip, Li was not in a boardroom. He was at Fudan University in Shanghai, giving a talk. The host asked him to rate his life out of ten. He gave himself a ten. "I'm very optimistic about myself," he said. He rated his outfit that day a six, a pass, which in Baidu's engineering culture, the report noted, was about right.
The students had written him a song. It was called "Mr. Robin Li, I really want to meet you," and its lyrics, in a mix of Chinese and English, were tender and slightly absurd: Hi Robin, I've heard of you, the legendary idol is you; Hi Robin, where are you? Out past the Fifth Ring Road in Xi'erqi. Xi'erqi is the district in northwest Beijing where Baidu's search-box headquarters stands. Hi Robin, I can't forget you. Stay true to yourself; focus is the secret of all success.
Nobody had to tell these students which name to use. Six years after the wrangler outside the China World Hotel, the English name had become simply what you called him.
In February 2026, Baidu launched BaiduWiki, an international version of Baike in English, Japanese, French, Spanish and Russian, opening with more than a million articles translated from Chinese with the help of AI tools. The encyclopedia that began in 2006 as a replacement for a blocked Wikipedia was now being sent abroad in Wikipedia's own language. There is an entry for Li in it. Its summary reads, in the cadence of a personnel file: "Robin Li, male, Han ethnicity, non-party personage, born in November 1968 in Yangquan, Shanxi." Further down, it records the opera troupe. Under "Alias," it lists a name he did not grow up with and the students at Fudan never sang: Mr. AI.
No volunteer will edit that page, not in Chinese and probably not in English. No inbound link will move it. It sits at the top of the results when you search for him on his own engine, lit the way the company's founder once imagined the thing you are looking for would be lit, in the place you forgot to check. And across the top of it, above the Han ethnicity and the coal city and the boy who once trained for the stage, the machine has done without being asked what a man with a megaphone once had to tell a crowd to do.
It uses his English name.
Part IIThe Playbook
What follows are twelve principles drawn from Robin Li's career, as evidence, not hagiography. Several are cautionary: Li's record is unusually rich in lessons that were learned, forgotten and relearned, and an honest playbook has to include the cost of those cycles as well as the payoff of the bets that worked. Each principle is grounded in decisions documented in Part I, and each ends with a single tactic an operator, founder or investor can actually use.
The principles
- 1.Stay with the category the incumbents have abandoned.
- 2.Let inbound references rank you.
- 3.Sell the pipes before you own the storefront.
- 4.Monetize intent before the incumbent does.
- 5.Treat the IPO as a brand event, not a financing.
- 6.Price your regulatory environment as an operating cost, and know what it won't cover.
- 7.Build the substitute the moment the original is blocked.
- 8.Audit the metric before it eats the mission.
- 9.A founder's letter is a tool, not a fix.
- 10.Let the first act pay for the second.
- 11.Hold enough control to outlast the market's patience.
- 12.Engineer feedback loops, not forecasts.
Principle 1
Stay with the category the incumbents have abandoned
The most important thing John Wu remembered about the 1998 picnic was not Li's brilliance but his stubbornness. Yahoo, then the most powerful company on the internet, had decided search did not matter. Disney had decided the same about Infoseek. Li disagreed, and he disagreed in the specific way that creates fortunes: he was not contrarian about the future of the internet, only about which layer of it would capture value.
The opportunity in an abandoned category is not just reduced competition. It is that the people best positioned to dominate it have stopped hiring for it, stopped funding it, and stopped thinking about it. The talent and the ideas become available. Li was a staff engineer at a company that had lost interest in its own core competence; the sense of abandonment was something he had lived through, not just observed.
The lesson cuts both ways. Baidu itself became the incumbent that drifted. The "wolf spirit" memo of 2012 is Li diagnosing in his own company the disease he had exploited at Yahoo and Infoseek.
Tactic: List the capabilities your industry's leaders have recently deprioritized or outsourced, and ask which of them will matter most in five years.
Principle 2
Let inbound references rank you
RankDex was built on a single idea: a page's own claims are weak evidence, and what other credible pages say about it is strong evidence. When Scott Walchek asked Li to name the world's top three search engineers, Li named others, and one of them, William Chang, named Li when Walchek called. The seed round went from $1 million to $1.2 million on the strength of that phone call.
Whether it was humility or strategy, the effect is a working model of how reputation compounds. Self-promotion invites discounting. A referral from a node the listener already trusts gets priced in immediately, and it carries the referrer's credibility with it. Li's fundraising, recruiting and later political standing all ran on endorsement networks rather than self-description.
The darker side is the locked Baike page. A founder who understands that reputation is assigned by others will be tempted, once powerful, to control the assigners. The algorithm's principle is only virtuous if you leave the links free.
Tactic: Before your next important pitch, identify the two people your counterparty already trusts who can vouch for you, and make sure the call happens before the meeting ends.
Principle 3
Sell the pipes before you own the storefront
Baidu did not start as a destination. It started as invisible infrastructure, supplying search to Sina, Sohu, NetEase, Yahoo! China and, by Li's account, "any portal company you can think of in China." Li was candid that this was not a feasible business model. It was something better: a way to get paid while learning the market, building the index, and watching where queries actually went.
The portal phase gave Baidu scale data and credibility without the expense of acquiring consumers directly. When Li saw Overture's model and decided the money was in owning the query, not processing it, he had already built the engine and studied every partner's traffic. Baidu.com launched in September 2001 with no need to prove its technology.
The risk is obvious: your customers become your competitors the day you open the storefront. Li accepted that because the storefront's economics were so much better.
Tactic: If your end market is expensive to reach directly, find the incumbents who already own it and sell them a component, while instrumenting everything you learn about their users.
Principle 4
Monetize intent before the incumbent does
In 2001 Baidu let advertisers bid for placement and pay per click, ahead of Google's adoption of the same approach. The model fit China's market unusually well: millions of small businesses with no access to national advertising and no tolerance for flat fees that might not deliver. Performance-based pricing let them pay only for results. Baidu turned profitable in 2004.
The underlying insight is that a search query is a statement of intent, and intent is the most valuable signal in advertising. Li saw the model working in Pasadena and moved before the global leader had brought it to China.
The cautionary half is just as important. Intent-based ads are most lucrative in the categories where users are most vulnerable: health, money, desperation. The "cancer" search of 2005 and Wei Zexi's death in 2016 are the same business model ten years apart.
Tactic: Identify the moment your users express the most explicit intent, monetize it on a pay-for-results basis, and set category-specific limits in advance for searches where a bad answer can hurt someone.
Principle 5
Treat the IPO as a brand event, not a financing
Li pushed for a U.S. listing in 2004 because, the Times reported, he believed it would be a "huge branding event" for a company the world already called "China's Google." He was right. Baidu priced at $27 on August 5, 2005 and closed at $122, the biggest first-day pop on Nasdaq since 2000. The stock later fell to $44, and analysts called it overvalued, but the headline had already done its work.
The listing gave Baidu something more durable than capital: international legitimacy, an American currency for talent, a story that reached Chinese users through foreign media and back again. Two years later, Baidu became the first Chinese company in the Nasdaq-100.
An IPO that depends on spectacle is volatile. But for a challenger fighting a global brand on home turf, the market debut was a marketing campaign that paid for itself.
Tactic: When you choose where and when to list, weight the narrative value of the venue as heavily as the valuation, and plan the story you want told on day one.
Principle 6
Price your regulatory environment as an operating cost, and know what it won't cover
"We have to spend a lot of resources to make sure our content and services abide by Chinese law," Li told Bloomberg in 2010. Google had treated Chinese censorship as a moral cost and eventually left. Li treated it as an operating expense, a moat that foreign competitors would not or could not pay for. When Google moved to Hong Kong, Baidu's share of search advertising jumped to around 70 percent.
This is the most uncomfortable principle in the playbook and the most effective one in Li's career. Compliance capacity can be a competitive advantage in heavily regulated markets, whether in finance, healthcare or state-supervised media. The firm that builds it well can win markets its rivals refuse.
But the 2011 CCTV documentary, the Spring Festival Gala sponsorship, the 2016 investigation and the 2026 Pentagon listing show the limit. A regulator you accommodate is still a regulator, and an environment you price as a cost can reprice you overnight. In 2022, Sustainalytics downgraded Baidu to "non-compliant" with UN Global Compact principles, a cost that was measured differently.
Tactic: Build a line-item estimate of your compliance spend and the moat it buys, and alongside it write down, specifically, what you would refuse to do regardless of price.
Principle 7
Build the substitute the moment the original is blocked
Wikipedia was blocked in China in 2005. Baike launched in April 2006 and passed Chinese Wikipedia's article count in twenty days. The window was short, and Li took it. He took it, too, with every integration advantage he had: Baike entries ranked at the top of Baidu searches, and when Hudong's chief complained, Baidu sued him and won.
When a market constraint suddenly removes a dominant player, demand does not vanish. It waits for whoever shows up first with something close enough. Speed matters more than polish. Baike was messy, accused of copying and riddled with hoaxes, and it became the largest Chinese-language encyclopedia in the world anyway. By 2025 it was winning its own copying lawsuit against ByteDance.
The substitute inherits the constraint that created it. Baike is fast, enormous and locked at the top. Twenty years later, BaiduWiki is exporting it in English.
Tactic: Keep a standing list of global products that could be blocked or withdrawn from your market, with a rough plan for what you would ship within thirty days if each one disappeared.
Principle 8
Audit the metric before it eats the mission
The 2016 letter is a founder discovering that his organization had optimized precisely for what it was measured on. "The management and employees' obsession of KPI has twisted our values," Li wrote. Engineers were "trapped in a conflict of interest between commercial interest and user experience, and often compromising the latter." His fix was structural: a "one-vote veto" for user experience, administered by a separate division with power to cancel products.
The pattern appears across Part I. Tieba's moderator sales generated relatively little revenue and enormous damage. Paid medical rankings were profitable until they were fatal. Each was a reasonable local optimization by someone hitting a number. The 2009 Phoenix Nest promise did not stop it, because a CEO's stated priorities lose to a manager's actual bonus formula.
The non-obvious lesson is that harm usually arrives through the metric, not against it.
Tactic: For each of your top five KPIs, name the user harm that would most efficiently increase it, and assign someone with veto power to watch for exactly that.
Principle 9
A founder's letter is a tool, not a fix
Li governs, in moments of crisis, by letter. The record is consistent enough to chart.
✉
The Letters
Robin Li's internal memos, each written in response to a crisis.
| Date | Trigger | Core image | Type |
|---|---|---|---|
| Nov 2012 | Complacency in search monopoly | Lost "wolf spirit"; the company as panda | Culture |
| May 2016 | Wei Zexi's death; regulatory probe | "Bankrupt in just 30 days"; cutting one's own wrist | Crisis |
| Jan 2019 | Slowing economy, trade war | "Winter is coming"; the pine and the cypress | Macro |
| May 2019 | First quarterly loss since IPO | "Saying 'I have tried' is not enough" | Crisis |
The letters are well written, emotionally specific, and quickly leaked, which suggests they were written to be leaked: the audience includes regulators, investors and the public as much as staff. That is their real function. A founder's letter sets the narrative and signals seriousness at a moment when a company has to be seen reacting.
But the same themes recur because letters do not change incentives. The 2016 letter was preceded by a 2009 promise and a 2012 diagnosis. The 2024 Qu Jing episode suggests the culture the letters argued against kept reproducing itself.
Tactic: Pair every all-company crisis letter with one concrete, dated change to how people are paid or promoted, and publish a follow-up in ninety days reporting whether it happened.
Principle 10
Let the first act pay for the second
Baidu's advertising engine stalled for a decade. Its second business grew more than twentyfold over the same period, funded by the first.
¥
The Revenue Mix
Baidu revenue by source, in millions of yuan. Online marketing barely grew; everything else funded the turn.
| Year | Online marketing | Other sources | Net income |
|---|---|---|---|
| 2015 | 64,037 | 2,345 | 33,664 |
| 2017 | 73,146 | 11,663 | 18,301 |
| 2019 | 78,093 | 29,320 | 2,057 |
| 2021 | 80,695 | 43,798 | 10,226 |
| 2023 | 81,203 | 53,395 | 20,315 |
| 2024 | 78,563 | 54,562 | 23,760 |
Li hired Andrew Ng in 2014, while search was still highly profitable, and launched Apollo in 2017, a year after the Wei Zexi crisis had cut net income by two-thirds. He kept funding. "When technology still evolves at such a rapid rate, you just cannot stop investing," he said in Dubai in 2025. The second act was always going to be expensive and slow; the only question was whether the first act could keep paying.
The table also shows the cost. Net income fell to 2.1 billion yuan in 2019. Founders who make this turn have to be willing to look bad for years.
Tactic: Set aside a fixed share of your core business's profit for your next platform, protect it in writing from quarterly cuts, and agree in advance with your board on what evidence would end it.
Principle 11
Hold enough control to outlast the market's patience
Li owns about 18 percent of Baidu and controls about 59 percent of the votes. That structure is why Apollo, announced in 2017, could still be absorbing capital in 2024 when it reached six million driverless rides and why it could survive the decline of Li's net worth from roughly $20 billion to $5.5 billion. Robotaxis are a bet measured in decades. Public markets measure in quarters.
Dual-class control is widely criticized, and often fairly. It insulates founders from accountability as well as from short-termism, and the Baidu record contains instances of both. But for capital-intensive, slow-maturing technologies, control lets a founder keep investing through years when investors would prefer him to stop.
Tactic: If your strategy depends on a bet longer than five years, secure the governance structure to protect it before you need it, because once the share price falls you will no longer be able to negotiate for it.
Principle 12
Engineer feedback loops, not forecasts
Li's favorite parable is a blind person with a Rubik's cube: 137 billion years without feedback, two and a half minutes with it. "Innovation drives growth, feedback drives innovation," he said in 2023. In Dubai two years later he went further: "Innovation can't be planned. You don't know when and where the innovation comes. What you can do is just to foster an environment that is conducive to innovation."
The contrast with his forecasts is instructive. He predicted Chinese traffic jams would be essentially gone by 2032 and traffic restrictions obsolete by 2027. Those are statements of conviction, not plans. What actually produced Apollo Go's safety record was 130 million kilometers of feedback. What pushed ERNIE X1 to half DeepSeek's price was a price war he called "almost inevitable," which is to say a market giving brutal feedback.
The deepest irony of the 2016 crisis is that Baidu had built the world's most sophisticated feedback machine for advertisers and a weak one for users. The fix Li proposed was more feedback: making user evaluations the main determinant of rankings.
Tactic: For your most important product, shorten the time between a user's experience and the moment an engineer sees a signal about it, and treat that latency as a metric in its own right.
Free playbook
Get The Business Model Playbook
58 business models, one visual page each: how the money flows, the metrics that matter, and who runs it. Free when you join the Faster Than Normal email.
Free. No spam. Unsubscribe anytime.
Part IIIQuotes and Maxims
In their words
— Robin Li, interview with Forbes, Beijing, August 2009A lot of Chinese people have wondered if knowledge really means power in today's market economy. I think I've proven that it does.
— Robin Li, on the Phoenix Nest advertising overhaul, Forbes, 2009Even if this means we have to suffer for a few quarters, I'm willing to make that bet — or sacrifice — to make sure we're moving in this direction. It's not an easy decision for a CEO. But if we imagine the company ten years down the road, it's an obvious one.
— Robin Li, internal letter to Baidu staff, May 2016If we lose the support of users, we lose hold of our values, and Baidu will truly go bankrupt in just 30 days.
— Robin Li, World Intelligence Congress, 2023The biggest danger for humanity is not the uncertainty brought about by innovation. On the contrary, the various unpredictable risks brought about by following inertia, not inventing, not creating, and not making progress are the greatest threat to humanity.
— Robin Li, World Governments Summit, Dubai, February 2025Innovation can't be planned. You don't know when and where the innovation comes. What you can do is just to foster an environment that is conducive to innovation.
Maxims
- Value is assigned from outside. A page, a product or a founder is worth what credible others point to, not what it says about itself.
- Stay where the giants left. The category the incumbents abandon is often the one that decides the next decade.
- Own the query. Whoever holds the moment of intent holds the margin; everyone upstream is a supplier.
- Speed beats polish when a door slams shut. When a dominant product is blocked, demand waits for whoever arrives first with something close enough.
- Compliance is a moat with a price. It can keep rivals out, but the authority you pay can always raise the rent.
- The metric is the mission, whether you like it or not. Organizations become what they measure, so audit the dashboard before it audits you.
- A lesson learned a hundred times was never learned. The same ad problem surfaced in 2005, 2009, 2011 and 2016; only incentives, not memos, end the cycle.
- Let the old engine pay for the new one. A second act is funded out of the first act's cash, and the bill comes due in years of bad earnings.
- Control buys patience. Decade-long bets need governance that outlasts the market's mood.
- Feedback, not forecasts. You cannot plan innovation, but you can shorten the distance between an action and its consequence.
