Part IThe Story
In March 1996, a piece of software written for a Stanford doctoral project was given a single address and told to start walking. The address was Larry Page's own home page on the university's computer science servers. It was the only starting point. From that one seed, which was also the most modest page imaginable, a graduate student's place on the internet, the crawler was to follow every link it found, then every link on every page those links led to, and so on outward into a World Wide Web then estimated at around ten million documents. Nobody could say how many links connected those documents, and nobody had a reliable way to decide which of them mattered. Page was twenty-two or twenty-three, depending on the day. He did not, by the account of the writer John Battelle, entirely know what he was getting into.
Five months later he posted a progress report, the kind of plain-text status page that graduate students leave for advisers and for one another. Some Rough Statistics (from August 29, 1996). Total indexable HTML urls: 75.2306 Million. Total content downloaded: 207.022 gigabytes. The system, called BackRub for reasons that were less suggestive than they sounded, was "written in Java and Python and runs on several Sun Ultras and Intel Pentiums running Linux." The primary database lived on a Sun Ultra II with 28 gigabytes of disk. At the bottom came the credits, which in retrospect read like a seating chart for a dynasty: "Scott Hassan and Alan Steremberg have provided a great deal of very talented implementation help. Sergey Brin has also been very involved and deserves many thanks." The first public version, released on the Stanford website that same month, consumed nearly half of the university's entire network bandwidth.
The note does not mention that the core idea behind all this machinery, which would carry its author's surname into the vocabulary of mathematics, was a way of deciding that a thing is important because other important things point to it. In this scheme importance is conferred, never claimed, and it is calculated recursively, so that one's own rank depends on the rank of whoever is doing the pointing, whose rank depends in turn on someone else's. The name PageRank worked as a pun twice over: on the web page, and on the man.
On September 24, 2019, twenty-three years and six months after the crawler left that home page, the last of the patents associated with PageRank expired. The patents had never belonged to Google, or to Page. They were assigned to Stanford, which licensed them to the company in exchange for 1.8 million shares and sold those shares in 2005 for $336 million. Ten weeks after the expiration, on December 3, 2019, Page and Brin jointly announced that they were leaving their executive roles at Alphabet, the holding company Page had invented four years earlier as a kind of corporate lung to let Google breathe. Sundar Pichai would run it. The founders would remain employees, board members, and, through a super-voting share structure that gives them roughly 56 percent of the voting power on about 14 percent of the shares, the people who decide.
The paradox is easy to state and harder to resolve. The man whose name became shorthand for measuring importance by incoming links has spent the years since 2019 receiving remarkably few that he authored. The public record of Larry Page thins to the vocabulary of wealth indices and state filings: a net worth estimated in 2026 at $334.1 billion, second in the world; a Christmas Eve transfer of an entity he co-managed with Brin from California to Nevada. Everything he built points outward from him, and very little points back from his own hand. It started, after all, with a single page.
By the Numbers
The Arithmetic of Importance
$334.1BPage's estimated net worth in 2026, second-largest in the world
56%Stockholder voting power controlled by Page and Brin, on roughly 14% of shares
90%Google Search's share of the global search market in 2025
$403BAlphabet revenue in 2025
0.85The damping factor: the probability PageRank's imaginary surfer keeps clicking
$750,000Price at which Excite's CEO declined to buy Google in early 1999
$336MWhat Stanford realized in 2005 from the 1.8M shares it received for the PageRank patent
The Stuff Lying Around
In 1979 two families each received a machine for the future, though only one of the machines was made of metal.
In Lansing, Michigan, Carl Victor Page Sr. brought home an Exidy Sorcerer. Page Sr. held a PhD in computer science from the University of Michigan and taught the subject at Michigan State. The BBC's Will Smale would later describe him as a "pioneer in computer science and artificial intelligence." The Sorcerer was an early personal computer, the sort of object that came into a professor's household as a curiosity and stayed as furniture. His younger son, six years old, mastered it and began using it for schoolwork. Larry Page would recall that he was drawn to computers because he could "play with the stuff lying around," meaning the first-generation machines his parents had left in the house. His mother, Gloria, taught computer programming at Lyman Briggs College at the same university. In a house like that the computer was simply part of the décor.
Page described that house as "usually a mess, with computers, science, and technology magazines and Popular Science magazines all over the place." In his 2013 founders' letter he wrote, "I remember spending a huge amount of time pouring over books and magazines," and the misspelling, faithfully preserved with a [sic] by every encyclopedia that quotes it, is one of the few unpolished artifacts in a corporate archive otherwise scrubbed to a high shine. The writer Nicholas Carlson judged that the combination of that atmosphere and attentive parents "fostered creativity and invention." That is a generous way of describing a childhood in which the dominant pedagogical method was disassembly. His older brother, Carl Victor Page Jr., taught him to take things apart, and before long, the family story goes, he was taking "everything in his house apart to see how it worked." He became, by his own account, the first child at his elementary school to hand in an assignment produced on a word processor.
His parents divorced when he was eight. The record says, with the flatness biographies reserve for things that probably were not flat at all, that he kept a good relationship with his mother and with his father's long-term partner, Joyce Wildenthal, another Michigan State professor. Page grew up surrounded by academics, and the academy runs on one currency above all others. A scholar's worth is measured by how often other scholars cite them.
Four thousand miles east, and four months after Larry Page's birth, Sergey Mikhailovich Brin was born in Moscow on August 21, 1973, into a Russian-Jewish family that shared a three-room apartment with his paternal grandmother. His parents, Mikhail and Eugenia, were both graduates of Moscow State University, his father a mathematician. In 1977, returning from a mathematics conference in Warsaw, Mikhail announced that, because of Soviet antisemitism, the family would emigrate. They applied for exit visas in September 1978, and his father was "promptly fired." His mother had to leave her job too. For eight months they lived on temporary work and fear, the condition of the refuseniks. In May 1979 the visas came through. After Vienna and Paris and help from the Hebrew Immigrant Aid Society, the Brins arrived in the United States on October 25, 1979, the same year the Sorcerer arrived in Lansing.
The parallel is too neat to lean on, and I won't lean on it much. Still, it rhymes. Both boys were sons of mathematically minded academics, and both, at six, were handed a future: one as a machine on a table, the other as a country. Brin would finish his bachelor's degree at the University of Maryland at nineteen, with high honors in mathematics, following his father and grandfather into the discipline. Page would follow his father to the University of Michigan. Both men took up the family trade, and both left it before finishing the doctorate.
Children's librarians, who have to make sense of these things for eleven-year-olds, have filed them together. The juvenile biographies, Sergey Brin and Larry Page: The Founders of Google among them, do what the genre requires and make two people legible as one story. There is something to that. There is also something lost in it, because the Lansing house and the Moscow apartment produced very different temperaments. The early record catches Page, more than Brin, at the moment he decided what he was for.
— Larry PageFrom a very early age, I also realized I wanted to invent things. So I became interested in technology and business. Probably from when I was 12, I knew I was going to start a company eventually.
He was twelve, and in that phrasing the business came second. He wanted to invent, and he reasoned backward to the company because a company was what you needed to get the invention into the world.
Time Is the Primary Thing
For two summers his parents sent him north to Interlochen, the arts camp in the woods of northern Michigan, where he played flute but mostly saxophone and studied composition. This is the sort of detail that usually ends up in the quirky-facts sidebar. Page has insisted it belongs in the main text.
"In some sense," he said, "I feel like music training led to the high-speed legacy of Google for me." He has explained what he meant in terms precise enough to be almost technical.
— Larry PageIf you think about it from a music point of view, if you're a percussionist, you hit something, it's got to happen in milliseconds, fractions of a second.
"In music, you're very cognizant of time," he added. "Time is like the primary thing." A drummer who comes in a quarter-second late has not played a slightly worse note. He has played a wrong one. Page carried that unforgiving sense of timing into a field where the prevailing assumption of the 1990s was that users would wait, since waiting was what the internet was for. The portals of that decade, Yahoo, AOL, Excite, and the rest, were built on the opposite philosophy. They wanted to keep you, to maximize the time you spent inside their walled gardens, and the stock market loved them for it. In late 1998, while the dot-com bubble was still inflating, Salon observed that those overloaded portal sites were widely seen as "the future of the Web." The same article argued that a beta-stage search engine out of Stanford returned better results than its competitors.
Google built the opposite thing: a page designed to be left as fast as possible. Page and Brin had, after all, no web designer. They used Page's basic HTML skills to build a simple search box, because they couldn't produce anything visually elaborate. The famous minimalism started as a limitation and only later became a creed. It was also, to a percussionist, plainly correct. The page loaded and got out of the way. A beat landed on time.
The obsession compounded over decades in ways that rarely made headlines and always made money. On June 8, 2010, Google completed an infrastructure overhaul codenamed Caffeine that promised results 50 percent fresher, moving its back-end indexing off MapReduce and onto Bigtable. Three months later, on September 8, 2010, it launched Google Instant, which predicted a query as the user typed it and displayed results for the top guess before the user finished typing. The company estimated it would save two to five seconds per search. That was the number Google chose to advertise: two to five seconds, multiplied across a search volume that would reach more than 3.5 billion queries a day by 2012 and more than eight billion by 2023. Seconds become person-years at that scale, and a drummer's sensibility, scaled to the species, turns into a theory of civilization.
There is an irony in this, which Page cannot be blamed for but which belongs in the account anyway. A company that won by getting people off its page as quickly as possible would eventually earn three-quarters of its revenue by putting advertisements on that page. In 2025, 73 percent of Google's revenue came from advertising. By the 2020s the page that once loaded and vanished was prefaced, for many queries, by sponsored results and an AI-generated overview. It would no longer get out of the way, because it had learned to hold the beat.
What Is a Link But a Citation?
At the University of Michigan, where he earned a bachelor's degree in computer engineering with honors in 1995, Page built an inkjet printer out of Lego bricks, technically a line plotter. He had worked out that large posters could be printed cheaply using inkjet cartridges, so he reverse-engineered the cartridge and built the electronics and mechanics to drive it. He served as president of the Beta Epsilon chapter of the Eta Kappa Nu honor society and was a member of the university's 1993 "Maize & Blue" solar car team. He wrote a business plan for a company that would use software to build a music synthesizer.
He also proposed, in all seriousness, that the University of Michigan replace its bus system with a personal rapid-transit system: essentially a driverless monorail with a separate car for every passenger. The proposal went nowhere. It is the earliest recorded instance of a pattern that would recur for three decades. Page would see a transportation system and imagine it without drivers, and the institution would politely decline. Two decades later, he would own the institution.
At Stanford, enrolled in the computer science PhD program and casting about for a dissertation topic, he considered several. Telepresence was one. Self-driving cars were another. The idea that held him was mathematical. He wanted to treat the World Wide Web's link structure as a single enormous graph and study its properties. His supervisor, Terry Winograd, a Stanford professor who would later co-author the founding paper on the project, encouraged him to pursue it. Page recalled in 2008 that it was "the best advice I ever got." One is tempted to admire the adviser here more than the advisee. Winograd looked at a student holding three ideas, one of which would become Waymo, and pointed at the one that would become Google.
Battelle, the Wired co-founder, compressed Page's insight into a single rhetorical question. The "entire Web was loosely based on the premise of citation—after all, what is a link but a citation?" If Page could count and qualify every backlink, then, as Page put it, "the Web would become a more valuable place." The son of two academics had noticed that the web was secretly an academic literature. Pages cited other pages. Some citations counted for more than others, depending on who was citing. A link from a page that was itself heavily linked was worth more than a link from a page nobody pointed to. It was the logic of the tenure committee applied to everything.
Nothing about the idea came from nowhere, and the history of PageRank reads, fittingly, like a citation graph.
In 1895 the German mathematician Edmund Landau suggested using the same underlying eigenvalue problem to determine the winner of a chess tournament. The problem was simple to state: a victory over a strong player should count for more than a victory over a weak one. In the 1950s at the University of Pennsylvania, Eugene Garfield developed citation analysis, the practice of weighing scholarship by who cites it. In 1976 Gabriel Pinski and Francis Narin applied the eigenvalue approach to ranking scientific journals. In 1977 Thomas Saaty used it to weight alternative choices. In 1995 two cognitive scientists, Bradley Love and Steven Sloman, proposed it as a model of how humans represent concepts. Massimo Marchiori at the University of Padua built a system called Hyper Search. In 1998, the year PageRank was formally introduced, Jon Kleinberg published his HITS algorithm.
The closest neighbor was in New Jersey. In 1996 an engineer named Robin Li, working at IDD Information Services, designed a search engine called RankDex that scored sites by how many other sites linked to them. Li called his method "link analysis." He filed a patent in 1997 and received it in 1999, and in 2000 he used the technology to found Baidu, which would become China's dominant search engine. Li and Page were contemporaries on opposite coasts who reached for the same idea at nearly the same moment, and each built an empire in a market the other would never fully enter.
Page's own patent filing cites Li's. That fact deserves more weight than it usually gets. The founding paper, likewise, cites Garfield, Marchiori, and Kleinberg. Whatever else one says about the competitive ferocity of the company Page built, its origin document observes academic manners. It points to its predecessors, and in doing so it practices the very theory it describes.
The paper also contains a mistake. In "The Anatomy of a Large-Scale Hypertextual Web Search Engine," which became one of the most downloaded scientific documents in the history of the internet at the time, Page and Brin confused two versions of their own formula. They claimed that one variant formed a probability distribution over web pages when it was the other that did. Mathematicians have been politely footnoting the error ever since. The most consequential paper in the history of information retrieval got its central equation slightly wrong, and it made no difference to anyone. The engine worked. The convergence numbers in the companion technical report were reassuring: on a network of 322 million links, PageRank settled to within tolerance in 52 iterations, and on a network half that size, in about 45. The algorithm scaled roughly with the logarithm of the network's size. It would keep scaling, and the world would keep feeding it.
Two Dorm Rooms and a Garage
They met in the summer of 1995. Page was among a group of prospective students, and Brin, already a Stanford graduate student on a National Science Foundation fellowship, had volunteered to show them around the campus and San Francisco. The encyclopedic accounts all record the same detail, in nearly the same words. The two "seemed to disagree on most subjects." Then, after spending time together, they "became intellectual soul-mates and close friends." One would like the transcript of those disagreements. None survives.
Their interests were adjacent rather than identical. Brin worked on data mining systems. Page was extending, in one description, "the concept of inferring the importance of a research paper from its citations in other papers." Both were part of the Stanford Digital Library Project, a federally funded effort, largely through the National Science Foundation, to build "the enabling technologies for a single, integrated and universal digital library." They also received funding from a program called Massive Digital Data Systems, run by the CIA and the National Security Agency to improve intelligence databases on the disorganized web. It is a footnote most accounts skip, and it is strange to sit with: the search engine that came to organize the world's curiosity was seeded partly by a federal program aimed at making the web legible to intelligence analysts.
Before Brin joined, there was Scott Hassan. Page described his ideas to Hassan, and Hassan began writing the code. Hassan was BackRub's lead programmer, sometimes called its unofficial third founder, and he left before Google became a company. His compressed biography, as far as the sources allow, consists of a credit line on a 1996 status page, a phrase ("third founder") always placed in quotation marks, and an absence from the incorporation papers. He wrote much of the original engine and then walked out of the story before it got its name. Alan Steremberg, the other name in Page's 1996 thanks, is even more of a ghost. Page and Brin both cited them as critical to Google's development, which in a company built on the arithmetic of citation is the highest honor available and also, conspicuously, the only one.
The infrastructure was improvised. They turned Page's dorm room into a machine laboratory, scavenging parts from cheap computers to build a device that connected the young search engine to Stanford's broadband network. When Page's room filled up, Brin's became the office and programming center. As usage among Stanford students grew, they bolted on more servers from whatever parts they could find. The stuff lying around, again. In the Lansing house it had been the parents' computers; at Stanford it was surplus hardware. The method was the same: take what's available, take it apart, make it do something it wasn't built for.
Page later described the moment the research project became something else. "We realized that we had a querying tool," he said. "It gave you a good overall ranking of pages and ordering of follow-up pages." The domain google.com was registered on September 15, 1997. The name was a misspelling of googol, the number written as a one followed by a hundred zeros. It was meant to signify scale, and it misfired orthographically in a way that, in hindsight, nobody would trade for the correct spelling.
In July 1998 Page announced new features. "The most significant," he wrote, "is a summary for each result that highlights where your query matched." These were the query-specific snippets now so familiar they look like the natural state of search results. Google was the first web search engine to use them, and the Stanford research showing their value had been presented only that March. Hugh Williams, a former Google vice president, would call snippets "the unsung heroes of web search." It is a revealing detail. The idea associated with Page is PageRank, the elegant eigenvector. Much of what made Google feel better than its rivals was plainer and more ergonomic: a two-line summary telling you why this result was here, so that you could decide in milliseconds whether to click.
On September 4, 1998, Page and Brin incorporated Google in a garage in Menlo Park belonging to their friend Susan Wojcicki. She would later become a Google executive and eventually the chief executive of YouTube, which makes hers the most consequential sublet in Silicon Valley history. The garage, Business Insider would report, rented for $1,700 a month. Three years to the day later, on September 4, 2001, the U.S. Patent Office granted patent number 6,285,999, "Method for node ranking in a linked database," the original PageRank patent, assigned to Stanford.
By the end of 1998 the index held about 60 million pages. The homepage still said BETA.
Then came the episode every founder mythology needs, the moment when the founders tried to sell. Early in 1999, Page and Brin went to George Bell, chief executive of Excite, one of the portals Salon had described as the future, and offered him Google for $1 million. Bell said no. Vinod Khosla, one of Excite's venture capitalists, talked the founders down to $750,000. Bell said no again.
It is easy to mock Bell, and nearly everyone has. It is more instructive to see the decision from inside his frame. Excite was a portal, and a portal's business is keeping you. A search engine that sent users away to the best possible result, faster than its rivals, was a product built to make the portal's visitors leave. Bell wasn't being stupid. He was being consistent, and his business model told him the truth about the thing he was being offered, which was that it would destroy him. That March, Google moved into an office at 165 University Avenue in Palo Alto. Within a year it was the default search engine for Yahoo, replacing Inktomi. In 2003, the year google entered the Merriam-Webster Collegiate Dictionary and the Oxford English Dictionary as a verb, the company leased an office complex at 1600 Amphitheatre Parkway in Mountain View from Silicon Graphics. In 2006 it bought the property outright for $319 million. The building had belonged to a fallen giant of the previous computing era. Google moved into its rooms, then bought them, and the name on the lease changed while the address stayed the same.
"Pretty soon, we had 10,000 searches a day," Page said of mid-1998. "And we figured, maybe this is really real." The sentence reads as charmingly modest. It also contains a methodology. The founders didn't decide the thing was real because it was elegant, or because the paper was being downloaded, or because Winograd approved. They decided when other people started pointing at it, ten thousand times a day.
Inherently Biased Toward the Advertisers
The appendix of a scientific paper is where authors put the material that matters too much to cut and too little to feature, and so it is often where they say what they actually think.
— Sergey Brin and Larry Page, “The Anatomy of a Large-Scale Hypertextual Web Search Engine” (1998), Appendix AAdvertising funded search engines will be inherently biased towards the advertisers and away from the needs of the consumers.
They meant it. The record says that Brin and Page "had originally opposed any advertising," and that they changed their minds only "as years went by without profit." The phrase years went by compresses what was in practice about two years of a famous, beloved, unprofitable product before the reversal began. In October 2000, Google launched AdWords with 350 customers. At first advertisers could only display text ads on searches of their choosing, priced per thousand views, the way advertising had always been sold. Through 2001 it was the company's only revenue stream.
Even in surrender, the founders negotiated terms with themselves, and the terms are the most Page-like thing about the episode. Every ad had to be relevant to the keywords. Ads could appear on only 15 percent of all searches. Google was deliberately leaving money on the table and imposing restrictions on advertisers they weren't used to. In February 2002 the system shifted to a cost-per-click model, with prices set by a Vickrey auction, the second-price sealed-bid mechanism beloved of game theorists. It rewarded advertisers for being relevant: an ad that people clicked on was worth more to Google than one they ignored, so relevance became the price of admission. It was PageRank applied to commerce. An ad's value would be conferred by the behavior of the people who saw it, not claimed by the person who paid for it. In 2003 AdSense extended the model to third-party websites, selecting ads from the keywords of search, email, and the pages themselves.
Whether this squared the circle with Appendix A is a question Page has never, as far as the provided record shows, answered directly. The machine was meant to be incorruptible because it measured what others pointed to. Revenue requires that someone pay to be pointed to. The founders' solution was elegant: make paying insufficient, and require relevance too. Whether it held is a different matter. In August 2024 a federal court in the District of Columbia ruled that Google held an illegal monopoly over search and search advertising, maintained in part by paying large sums to phone makers and browser developers to make Google the default. The ruling is often described as the first antitrust ruling against a U.S. company in twenty-four years. The founders who once considered advertising a corrupting influence on rankings had built a company found to be paying for the most valuable placement of all, the one that comes before the search even begins.
There is a second chapter, quieter and in some ways more revealing. In March 2024 an automated bot committed internal Google documentation for its Content Warehouse API to a public GitHub repository, more than 2,500 pages describing over 14,000 attributes. Google confirmed the material was authentic on May 29, 2024, while cautioning against "inaccurate assumptions about Search based on out-of-context, outdated, or incomplete information." Among the attributes was a system called NavBoost that adjusts rankings based on clicks, with fields named "goodClicks," "badClicks," and "lastLongestClicks." In antitrust testimony, a Google vice president had said the system had used click data since about 2005. Google representatives had long downplayed the use of click signals. There were fields apparently drawing on Chrome browser data, and a "siteAuthority" attribute of the kind Google had publicly denied keeping.
The engine Page conceived to read what pages said about each other had, at some point, begun reading what people did, and Google had not said so plainly. The links had been joined by the clicks. In 2002, when Google made its first publicly announced change to its search algorithm, a number of internet commentators called it the death of PageRank. They were early, and also, in a way, right.
The Grown-Up, and the Return
Page was chief executive of Google from 1997, before the company formally existed, until August 2001, when Eric Schmidt succeeded him. The arrangement became one of Silicon Valley's canonical stories about founders and supervision. The young technical founder makes way for the experienced operator, the company matures, and the founder returns.
The sources provided for this profile are thin on Schmidt's prior life, and I won't import what they don't contain. What they do preserve is his voice, and the voice is revealing. In 2007 he told the Financial Times that "the goal is to enable Google users to be able to ask the question such as 'What shall I do tomorrow?' and 'What job shall I take?'" In 2010 he told The Wall Street Journal: "I actually think most people don't want Google to answer their questions, they want Google to tell them what they should be doing next." It is worth setting those sentences beside Page's 1998 description of what BackRub did: "It gave you a good overall ranking of pages and ordering of follow-up pages." In twelve years the ambition had moved from ordering pages to ordering lives. Page's engine sorted documents. Schmidt's articulated mission was to sort tomorrow.
During the Schmidt decade Google went public, in 2004, and the founders structured the company so that public investors could own it without controlling it. Page and Brin together would come to hold about 14 percent of the listed shares and 56 percent of the voting power. It is the opposite of PageRank's premise. In the algorithm, influence flows from the number and weight of those who point to you. In the cap table, influence was simply declared, by charter, in advance. The founders who built the most democratic ranking system in history protected themselves from its corporate equivalent.
The honors arrived in that decade: MIT Technology Review's list of top innovators under thirty-five in 2002; an honorary MBA from IE Business School in 2003; the Marconi Prize in 2004, which cited "the invention that has fundamentally changed the way information is retrieved today"; Forbes's fifth-most-powerful people in the world in November 2009, the two counted as one entry. In 2007 Page married Lucinda Southworth. Brin married Anne Wojcicki, Susan's sister, that May in the Bahamas. The founders shared, by then, a customized Boeing 767-200 and a Dassault/Dornier Alpha Jet, and paid $1.3 million a year to park them, along with two executives' Gulfstreams, at Moffett Federal Airfield, the NASA facility next door to the Googleplex. NASA installed scientific equipment on the aircraft so that experimental data could be collected in flight. Even the toys came with a research justification attached. The professors' sons had found a way to make a private jet count as fieldwork.
In April 2011 Page returned as chief executive. The record does not say much about the hand-off, and it would be fiction to supply it. It does show what Page inherited. Two months earlier, on February 23–24, 2011, Google had launched Panda, an algorithm update affecting about 12 percent of search queries and aimed at content farms, scrapers, and sites with heavy ad-to-content ratios. Matt Cutts, then head of Google's webspam team, later said that "with Panda, Google took a big enough revenue hit via some partners that Google actually needed to disclose Panda as a material impact on an earnings call." He added: "But I believe it was the right decision to launch Panda, both for the long-term trust of our users and for a better ecosystem for publishers." Here, briefly, was Appendix A revived. A public company knowingly cut its own revenue to improve the quality of its rankings. It is the kind of decision Page's original theory demanded and the kind his company would find harder and harder to make as it grew. Panda was eventually folded into the core algorithm in 2015, the same year Page folded Google into something larger.
Cleaner and More Accountable
On August 11, 2015, Page published a blog post announcing that Google would reorganize under a new public holding company called Alphabet. The language was unusually plain for a corporate restructuring of that size. The structure would allow "more management scale, as we can run things independently that aren't very related." Google itself would be "a bit slimmed down, with the companies that are pretty far afield of our main internet products contained in Alphabet instead." The goal was to make Google "cleaner and more accountable and better," and to improve "the transparency and oversight of what we're doing."
The words "pretty far afield" carry a great deal of freight. Under Alphabet sat X Development, the research lab; Calico, which studies aging; Verily, the life sciences unit; Nest; Fiber; and the investment arms CapitalG and GV. Later the list would include Waymo, the self-driving car company; Wing, the drone delivery operation; DeepMind; and Isomorphic Labs. These were the dissertation topics Page had set aside at Stanford, returning as subsidiaries. He had considered self-driving cars in 1995. He had proposed driverless monorails in Ann Arbor before that. Twenty years later, the man who couldn't get a university to replace its buses owned a company whose autonomous vehicles operated extensively in San Francisco. In January 2024 it filed with the California Public Utilities Commission to expand service to Los Angeles. In 2017 Alphabet sued Uber over stolen self-driving technology. The undergraduate's idea had become valuable enough to steal.
Eric Schmidt later supplied the genealogy. Speaking at a 2017 conference, he said the structure had been inspired by Warren Buffett's Berkshire Hathaway. A decade earlier he had encouraged Page and Brin to visit Buffett in Omaha and see how a holding company of subsidiaries with strong, trusted chief executives worked. One pictures the founders in Omaha, though the record gives no details of the visit. What they apparently took away was a principle Buffett had practiced for half a century: the person at the top shouldn't run things. The person at the top should pick the people who run things, then decline to interfere.
The mechanics were, characteristically, a matter of graph theory. Google Inc. first created Alphabet as its own subsidiary. A placeholder subsidiary was then created to own Alphabet, merged with Google, and Google's stock converted into Alphabet's, so that parent and child swapped places. Under Delaware corporate law, a holding company reorganization of this kind needs no shareholder vote, and this one had none. The restructuring was completed on October 2, 2015. Alphabet kept Google's stock price history and its ticker symbols. Investors woke up owning something new that traded exactly like something old. Sundar Pichai, then the company's product chief, became chief executive of Google. Page went up a floor, so to speak, to run Alphabet with Brin.
It was a structure designed to need Page less. That wasn't a flaw. It was the stated design goal: "We can run things independently." The founder who had once been replaced as chief executive by an outsider now did the replacing himself, one level at a time, by building a frame in which the people who ran things weren't him.
The Damping Factor
PageRank's mathematics rest on an imaginary person. The model calls this person the random surfer: someone who starts on a page and clicks links at random, forever, moving through the web. The surfer can't click forever, though. At any step, with a fixed probability, the surfer stops following links and jumps somewhere else entirely, to a random page with no link leading there. The probability of continuing to click is the damping factor, conventionally set at 0.85, an estimate the literature ties to how often an average person uses a browser's bookmark feature. With probability 0.15, the surfer simply leaves.
Without that small chance of departure, the mathematics break. The matrix needs it to converge, so that a single stable answer emerges instead of the calculation getting trapped in loops and dead ends. The surfer's occasional decision to stop following the chain is what makes the whole ranking computable.
On December 3, 2019, Page and Brin announced that they were stepping down. Pichai would be chief executive of Alphabet as well as Google. The founders would keep their board seats, their employment, and their majority vote. The PageRank patents, assigned to Stanford and licensed exclusively to Google, had expired ten weeks earlier. The two events have no causal connection, and the timing is a coincidence. It is the sort of coincidence that a profile of a man named Page is permitted to notice once and then leave alone.
What came after is mostly arithmetic. Alphabet's 2025 revenue was $403 billion, with $132 billion in net income. In June 2026 it employed 198,933 people. On September 15, 2025, it became the fourth company in history to top $3 trillion in market capitalization, partly on relief at an antitrust remedy that did not require it to sell Chrome. In January 2023 Pichai laid off about 12,000 people, 6 percent of the workforce, many of whom learned of it when their account access stopped working. In December 2024 the company unveiled Willow, a quantum chip that completed a benchmark calculation in five minutes that would take a classical computer longer than the age of the universe. Alphabet holds a 14 percent stake in Anthropic, an AI rival. In 2023 Brin came out of retirement to work on AI research at Alphabet. The record provided here says no such thing of Page.
The rest of the record is filings. In December 2025, facing a proposed California billionaire tax, Brin and Page, worth a combined $520 billion, terminated or moved some sixty limited liability companies holding their assets out of the state. The New York Times reported that one entity jointly managed by the two founders moved from California to Nevada on Christmas Eve. Their names appear on the paperwork, and that is all the paperwork shows.
The genre of the children's biography, in which Larry Page (Tech Icons) is a representative entry, ends where such books must, with a man who has achieved everything. It is a satisfying shape, and a misleading one. The adult version is stranger and unresolved. The engine he imagined has been reengineered past recognition: deep neural networks since 2016, RankBrain before that, hundreds of secret signals, clicks recorded and weighted, a generative model answering questions on the results page before any result appears. "PageRank continues to provide the basis for all of Google's web-search tools," one technical summary holds, while acknowledging it is now just one of many factors, and the Google Toolbar's public display of PageRank scores, once a zero-to-ten green bar obsessed over by every webmaster alive, was switched off on April 15, 2016. The name persists, the mathematics persist in biology and neuroscience and the ranking of NFL teams, and the man persists in the voting rights.
Every account of Google's origin, including this one, has to return to the same point, because that is where the record begins. Before the patents, the garage, the ads, the auction, and the holding company, there was one instruction given to one program in March 1996: start here. Here was a single URL on a Stanford server, the personal home page of a graduate student who would soon have no use for it. From that address the crawler followed the first link it found, and then the next one.
Part IIThe Playbook
What follows is an attempt to extract the operating logic beneath Larry Page's career: the recurring decisions, structural choices, and habits of mind that show up in the record from Lansing to Alphabet. Some of these principles Page articulated himself. Most he simply enacted, which is more instructive, since the gap between a founder's stated philosophy and his revealed one is usually where the useful lessons live. The analysis stays close to the evidence in Part I. Where the record is silent, so is the playbook.
Twelve Principles
- 1.Measure importance by what others point to, not what you claim.
- 2.Seed from the smallest thing you control, then follow the links.
- 3.Treat latency as a quality defect, not a performance metric.
- 4.Cite your precursors — credit is a structural asset.
- 5.Monetize under self-imposed scarcity.
- 6.Use mechanisms, not policies, to align incentives.
- 7.Let the market tell you when the project is real.
- 8.Keep the abandoned dissertation topics on file.
- 9.Secure governance before you need it.
- 10.Separate the engine from the experiments.
- 11.Pay for trust in visible revenue.
- 12.Build structures designed to need you less.
Principle 1
Measure importance by what others point to, not what you claim
The founding intellectual move of Page's career was to reject self-description as evidence. The search engines of the mid-1990s ranked pages largely by how often a search term appeared on them, which meant they ranked pages by what the pages said about themselves. That was trivially gameable and, worse, uninformative. Page's insight, inherited from the citation culture of the academy he grew up in, was that a document's importance lives outside the document, in the pattern of who refers to it and how much those referrers are themselves referred to.
The principle generalizes well beyond search. Inside an organization, the person everyone routes questions to is more central than the person with the larger title. In a market, the product that other builders integrate with is more strategic than the product with the louder launch. In hiring, a candidate recommended by people whom others recommend carries a different weight than one with a polished résumé. The recursion is the point: a signal is only as strong as the signals that confer it.
It also carries a warning that Page's own company eventually ran into. Any system that measures what others point to will be attacked by people manufacturing pointers: link farms, paid links, the whole industry of search engine optimization. Google spent decades building defenses, from the nofollow attribute in 2005 to penalties for paid links beginning in December 2007. Conferred-importance systems are robust only as long as conferring stays costly.
Tactic: For any decision about what matters, whether a feature, a hire, or a partner, ask who independently points to it and how credible those pointers are, before you ask what it says about itself.
Principle 2
Seed from the smallest thing you control, then follow the links
The BackRub crawler began in March 1996 with exactly one starting point: Page's own Stanford home page. From that single node it reached 75 million URLs within months. The lesson isn't that small beginnings are romantic. It is that a well-designed propagation mechanism makes the starting point almost irrelevant, so you should start from whatever you have direct, uncomplicated access to.
The same pattern recurs through the founding period. The search engine was first offered to Stanford users, its nearest network. The hardware came from whatever parts were lying around in two dorm rooms. The first office was a friend's garage. At each stage the founders seeded from the asset closest at hand and let the growth mechanism, more searches, more links, more users, carry them outward. By mid-1998 they were at 10,000 searches a day. By the end of that year, the index held 60 million pages.
Founders often stall trying to choose the perfect beachhead. If your mechanism is sound, the beachhead matters much less than your willingness to begin. If your mechanism is unsound, no beachhead will save you.
Tactic: Launch from the nearest network you already belong to, and spend your design energy on the propagation mechanism rather than on the choice of starting point.
Principle 3
Treat latency as a quality defect, not a performance metric
Page traced Google's obsession with speed to his musical training at Interlochen. "If you're a percussionist, you hit something, it's got to happen in milliseconds." In music, lateness isn't a degraded version of correctness. It is wrongness. That framing turns speed from an engineering optimization into a definition of quality.
Its strategic consequences were large. The portals of the late 1990s were designed to retain users. Google was designed to release them as fast as possible. The minimalist homepage, the snippets that let users judge results at a glance, Caffeine's 50 percent fresher index in 2010, and Google Instant's estimated savings of two to five seconds per query all follow from treating delay as an error rather than a cost. Excite's CEO couldn't value a product built on that premise, because his business depended on the opposite one.
The principle is non-obvious because most organizations measure speed but don't moralize it. A dashboard that shows load time is not the same as a culture in which a slow product is understood to be a broken product.
Tactic: Define a latency threshold beyond which your product counts as broken, not slow, and staff the fix with the urgency you would give an outage.
Principle 4
Cite your precursors — credit is a structural asset
Page's patent filing for PageRank cites Robin Li's earlier RankDex patent. The founding paper cites Eugene Garfield, Massimo Marchiori, and Jon Kleinberg. Page's August 1996 status note credits Scott Hassan, Alan Steremberg, and Brin by name. For a project whose core idea was that citation confers legitimacy, this was consistent, and it was strategic.
Citing precursors establishes the boundaries of your claim precisely. You aren't asserting that you invented link analysis or eigenvector ranking, whose lineage runs back to Edmund Landau's 1895 chess tournaments. You are asserting a specific application and implementation. Precise claims are more defensible than grand ones. Generous credit also builds the kind of reputation that recruits collaborators. The BackRub team grew because talented people saw that contributions were acknowledged.
The record also shows the limits of credit. Hassan, called the unofficial third founder, left before incorporation, and his recognition largely stops at a credit line. Acknowledgment is necessary. It isn't the same as equity.
Tactic: In every pitch, patent, and launch, name the prior work you build on; it narrows your claim to something defensible and signals to future collaborators that their contributions will be seen.
Principle 5
Monetize under self-imposed scarcity
Page and Brin opposed advertising outright in 1998, warning in their paper's appendix that ad-funded search would be "inherently biased towards the advertisers." When they reversed course in October 2000, they did so under constraints. Ads had to be relevant to the keywords, and they could appear on only 15 percent of searches. Google deliberately left revenue unclaimed.
That scarcity did real work. It protected the user experience that was the company's actual asset, and it raised the signal value of each ad, since an advertisement that appears rarely and only when relevant carries more weight with the viewer. It also bought the founders time to discover a better pricing mechanism (see Principle 6) before the ad load could degrade the product.
The arc of the story complicates the lesson. By 2025, advertising was 73 percent of Google's revenue, and the results page that once got out of the way was crowded with sponsored placements and AI overviews. Self-imposed scarcity tends to erode as revenue pressure builds. The principle is most powerful in the early years, when the constraint shapes the product's identity, and hardest to sustain later.
Tactic: When you introduce monetization, publicly cap its intrusiveness at a level that costs you something, and treat any loosening of the cap as a strategic decision requiring explicit justification.
Principle 6
Use mechanisms, not policies, to align incentives
In February 2002 AdWords moved from charging per thousand views to charging per click, with prices set by a Vickrey auction. The change was architectural. Under the old model, an advertiser could buy placement regardless of whether anyone cared. Under the new one, irrelevant ads attracted fewer clicks and generated less revenue for Google, so relevance became economically enforced rather than editorially policed.
This is PageRank's logic applied to commerce. The algorithm didn't ask pages to be honest about their importance. It measured importance from the behavior of others. The auction didn't ask advertisers to be relevant. It made irrelevance expensive. Policies require enforcement and invite gaming. Well-designed mechanisms make the desired behavior the self-interested one.
⚙
Policy vs. Mechanism in Page's Decisions
Where the founders chose structural incentives over rules, and where the record shows strain.
| Problem | Conventional approach | Page-era approach | Type |
|---|---|---|---|
| Ranking pages | Count keyword frequency on the page | Weight incoming links by the linkers' own rank | Mechanism |
| Ad relevance | Sell impressions per thousand views | Cost-per-click priced by Vickrey auction | Mechanism |
| Ad intrusiveness | Maximize inventory | Cap ads at 15% of searches | Policy |
| Link spam | Ignore it | Nofollow attribute (2005); paid-link penalties (2007) | Policy |
| Founder control | One share, one vote | Super-voting stock: 56% of votes on ~14% of shares | Structure |
| Default placement | Win on product alone | Payments for default status, ruled monopolistic in 2024 | Risk |
The table also shows where the approach strained. The policy-based measures needed constant revision, and the 2024 antitrust ruling suggests that buying distribution sat uneasily with the company's founding faith that the best result wins on its merits.
Tactic: Before writing a rule to prevent bad behavior, look for a pricing or ranking mechanism that makes the good behavior more profitable for the person you're trying to govern.
Principle 7
Let the market tell you when the project is real
"Pretty soon, we had 10,000 searches a day. And we figured, maybe this is really real." Page's account of the moment BackRub became a company is notable for what it leaves out. The decision didn't rest on the elegance of the mathematics, the paper's download count, or the approval of advisers. It rested on usage, on other people choosing, unprompted, to come back.
That is the PageRank principle turned on the founders' own work: importance is conferred by external behavior, not internal conviction. It guards against the common failure of falling in love with a technically beautiful system that nobody wants. It also explains why the founders were willing, in early 1999, to sell Google to Excite for $1 million and then $750,000. Before the external signal was overwhelming, they didn't over-value their own creation.
Tactic: Set a concrete, external usage threshold, chosen in advance, that will tell you a side project has become a company, and don't let enthusiasm substitute for crossing it.
Principle 8
Keep the abandoned dissertation topics on file
At Stanford, Page considered three dissertation directions: the web's link structure, telepresence, and self-driving cars. Earlier, at Michigan, he had proposed replacing the campus bus system with driverless personal rapid transit. He chose the link structure. The other ideas didn't disappear. They resurfaced, decades later, as Waymo and the other "pretty far afield" bets housed under Alphabet.
The pattern suggests a discipline: you can only pursue one primary idea at a time, but the ideas you defer are an option portfolio, not a graveyard. Success with the primary idea eventually supplies the capital, talent, and organizational structure to exercise the deferred options. Page's undergraduate monorail proposal went nowhere because he had no leverage. Twenty years later he had all of it.
◷
The Long Half-Life of a Deferred Idea
Autonomous transport in Page's record, from undergraduate proposal to operating company.
1991–1995
At the University of Michigan, Page proposes replacing campus buses with a driverless personal rapid-transit monorail.
1995
At Stanford, he considers self-driving cars as a dissertation topic; Terry Winograd steers him toward the web's link graph.
2015
Alphabet is formed to house businesses "pretty far afield" from Google's internet products.
2017
Alphabet sues Uber over stolen self-driving technology.
2024
Waymo, operating extensively in San Francisco, applies to expand service to Los Angeles.
Tactic: Maintain a written list of the serious ideas you chose not to pursue, with the reason you deferred each, and revisit it whenever your resources change by an order of magnitude.
Principle 9
Secure governance before you need it
When Google went public in 2004, the founders structured the company so that Page and Brin would come to control about 56 percent of the voting power while owning roughly 14 percent of the shares. In 2015 Alphabet was created through a Delaware holding company reorganization that required no shareholder vote. In 2019 the founders stepped down from executive roles while keeping their board seats and controlling votes.
Each of these moves depended on the one before. The super-voting structure made it possible to impose long-horizon decisions, such as Panda's revenue hit, the moonshot subsidiaries, and the Alphabet restructuring itself, without negotiating with short-term investors. It also made it possible for the founders to leave operating roles without losing the ability to choose who held them. Governance chosen early is cheap. Governance sought late, after dilution, is often impossible.
There is a tension worth naming. The founders' engine distributed influence according to what others conferred. Their charter concentrated it by declaration. Page applied democratic mechanics to the product and dynastic mechanics to the company, and the combination is what made the long game possible.
Tactic: Decide on your control structure before your first significant outside financing, because every later round makes the decision more expensive and the negotiation less yours.
Principle 10
Separate the engine from the experiments
Alphabet's stated purpose, in Page's August 2015 announcement, was to make Google "cleaner and more accountable" by moving out "the companies that are pretty far afield of our main internet products." The inspiration, Eric Schmidt later said, was Berkshire Hathaway: a holding company of subsidiaries run by strong chief executives trusted to operate independently.
The structural insight is that a profitable core business and a set of speculative bets have incompatible accounting needs. Mixed together, the core's profits conceal the bets' losses, and the bets' uncertainty clouds the core's performance. Each also distorts the other's culture: the core grows tolerant of waste, and the experiments become burdened with the core's processes. Separation produces "transparency and oversight," in Page's words, for both.
The model also sets up a clear test. If a bet can't survive scrutiny as a standalone entity under a holding company, it probably wasn't a bet. It was a subsidy.
Tactic: Once your core business is profitable, report your experimental initiatives as separate units with their own leaders and books, so neither the core nor the experiments can hide behind the other.
Principle 11
Pay for trust in visible revenue
When Google launched Panda in February 2011, the algorithm update cut into revenue from some partners enough that the company had to disclose it as a material impact on an earnings call. Matt Cutts later defended the decision as right "for the long-term trust of our users and for a better ecosystem for publishers." Page returned as chief executive two months later, inheriting a company that had just demonstrated it would sacrifice reported revenue to protect ranking quality.
This is the same principle as the 1998 appendix and the 15 percent ad cap, carried into the public-company era: the asset is the trustworthiness of the ranking, and revenue is a derivative of that asset. Companies routinely say they put users first. Few accept a disclosed hit to prove it.
The 2024 documentation leak complicates the record. It suggested Google had used click and Chrome data in ways its representatives had long downplayed. Trust is built partly through sacrifices like Panda and spent partly through opacity. The ledger has two sides.
Tactic: Identify one quality improvement that would visibly cost you revenue this quarter, and if it serves long-term user trust, ship it and explain the cost plainly.
Principle 12
Build structures designed to need you less
Page's career contains three handoffs: to Eric Schmidt in 2001, to Sundar Pichai at Google in 2015, and to Pichai at Alphabet in 2019. The last two were by Page's own design. Alphabet was explicitly built so that things could "run independently." By the time Page stepped back, the company's structure assumed his absence from daily operations.
The PageRank model offers an unexpected metaphor. Its imaginary surfer must, with probability 0.15, stop following links and jump elsewhere. Without that damping factor, the computation doesn't converge. Systems need their participants, even the central ones, to step away for the whole to settle into a stable state. A founder who can't leave builds a company that can't outlive him.
Page did not relinquish control. He kept the board seat and the vote. He relinquished operation, which is a different and more replicable act: keep the power to choose who runs things, and give up the running.
Tactic: Design your organization's next structural change around the assumption that you will not personally run any of its units, and judge it by whether that assumption holds.
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
— Larry PageIn music, you're very cognizant of time. Time is like the primary thing.
— Larry Page, on BackRub's early behaviorWe realized that we had a querying tool. It gave you a good overall ranking of pages and ordering of follow-up pages.
— Larry Page, on mid-1998Pretty soon, we had 10,000 searches a day. And we figured, maybe this is really real.
— Larry Page, BackRub status page (statistics from August 29, 1996)Scott Hassan and Alan Steremberg have provided a great deal of very talented implementation help. Sergey Brin has also been very involved and deserves many thanks.
— Larry Page, announcing Alphabet (August 11, 2015)More management scale, as we can run things independently that aren't very related.
Maxims
- Importance is conferred, not claimed. Judge anything by who independently points to it, and weight the pointers by their own standing.
- Start with what you own. The crawler began with one home page; a sound propagation mechanism makes the seed nearly irrelevant.
- Lateness is wrongness. Treat a slow product as a broken product, as a percussionist treats a late beat.
- Cite the giants. Naming your precursors narrows your claim to something defensible and earns you collaborators.
- Monetize like you're rationing. A self-imposed cap on intrusiveness protects the product that makes monetization possible.
- Prefer mechanisms to rules. An auction that makes irrelevance expensive works better than a policy that forbids it.
- Let usage declare it real. Decide in advance what external signal will turn a project into a company.
- Defer ideas; don't discard them. The topics you set aside at twenty-two can become subsidiaries at forty-two.
- Lock governance early. Control is cheapest to secure before anyone else has a claim on it.
- Build for your own absence. The best structure is one that keeps running after you stop operating it.
