Google Part I: Origins of Search. How the Best Business in Human History Happened (Audio)
Summary
- Google’s search franchise became what Ben Gilbert and David Rosenthal call “the single greatest business of all time” because high gross margins, a giant market, and roughly 90% share compounded rather than merely added. In the episode’s framing, Alphabet generated more net income than any other US company, while its core search business was estimated near an 87% gross margin. The investor question is whether AI can finally assault a position that had looked “unassailable.”
- PageRank mattered because it transformed the web’s link structure into a decentralized authority system just as the internet became large enough to require search but remained small enough to copy and index. The web grew from 130 sites in 1993 to more than 600,000 in 1996—roughly 723% annually—and Google treated hyperlinks like academic citations, weighting links from important pages more heavily and using anchor text as descriptive metadata. As the hosts put it, “Google could not have been built at any other time in history.”
- Google’s incumbents rejected better search because their CPM-funded portal model rewarded page views, while relevance helped users leave. An Excite test reportedly produced exactly the desired result immediately, prompting its CEO to ask why he would adopt technology that reduced time on site. Google’s counterposition was uncompromising: “Our whole point is to help people find what they want.”
- Google’s infrastructure turned scarce capital into a durable cost advantage by replacing expensive enterprise servers with unreliable commodity components managed through distributed software. The company split its index into 64-megabyte chunks, replicated data, tolerated hardware failure rates above 10%, and designed everything from file systems to racks and data centers. This architecture enabled a larger index, faster results, global delivery, and economics competitors using “gleaming” enterprise machines could not match.
- GoTo.com made the auction, self-service, and cost-per-click pieces of paid search work at scale, but Google assembled them into the superior system by adding click-through-rate-based Ad Rank and second-price auctions. Ranking ads by bid multiplied by probable engagement aligned user relevance, advertiser efficiency, and Google’s expected revenue; strong advertisers could bid less yet win by earning more clicks. The transition helped revenue rise from $86 million in 2001 to $440 million in 2002 and $1.5 billion in 2003.
- Search exhibited increasing returns to scale: more users attracted more advertisers, deeper auctions raised revenue per query, and that higher monetization funded still more distribution. Google therefore paid partners as much as 100%—occasionally more—of near-term revenue, gave AOL an 85% share, and accepted a $100 million guarantee it could not readily cover. Larry Page’s standard was categorical: “If not, we deserve to go out of business.”
- Google’s dominance was purchased and engineered as much as it spread by word of mouth. Yahoo’s 2000 investment and search deal carried the company through the dot-com crash; “Powered by Google” made it the Intel Inside of search; and Toolbar, Firefox, Dell, Adobe, RealNetworks, WinZip, Google Earth, and portal partnerships placed search wherever users already were. A Toolbar user searched roughly seven times as often, turning an estimated $2 annual user into one worth more than $10.
- The founders’ unusual ambition, equal partnership, and “healthy disregard for the impossible” made Google willing to adopt outside ideas, recruit exceptional talent, and repeatedly bet the company. The resulting castle combined PageRank, infrastructure, advertising liquidity, distribution, data feedback, culture, and founder control—not one isolated algorithm. In the episode’s comparison, Alphabet traded near 20 times earnings versus 27-46 times for major peers, suggesting that “Mr. Market” saw AI making the future materially less secure.
Deep dive
1. Google turned the internet’s front door into America’s largest profit pool
Ben and David frame artificial intelligence as “the next trillion dollar technology wave” after PCs, the internet, and mobile—and argue that understanding it requires understanding Google, the company most responsible for AI’s technical foundation and the preceding search wave.
Google was neither the first search engine nor alone in 1998; roughly a dozen competitors already existed. The puzzle is why Google became “the last”: how a clever ranking technology and clean product became a business with approximately 90% market share.
In the episode’s framing, Alphabet generated more net income than Apple, Microsoft, ExxonMobil, JPMorgan Chase, or Berkshire Hathaway. The hosts’ multiplication is the essential economic call: “super high gross margin,” an enormous market, and monopoly-scale share produced a “cash gusher.”
That position looked nearly unassailable for a quarter-century, but the hosts see the current AI wars as the first serious challenge. The episode therefore treats Google’s origins not as nostalgia, but as the operating history needed to understand whether its advantages can transfer.
2. Larry Page intended to build a world-changing company
Larry Page was born in Lansing, Michigan, in March 1973 to Carl and Gloria Page, both steeped in computing: his father was a computer-science professor and his mother taught programming. A Stanford sabbatical brought the family to Palo Alto when Larry was six or seven, giving him an early view of Silicon Valley.
His older brother Carl, nine years ahead, studied computer science at Michigan before working at Microsoft and Mentor Graphics. Computers, universities, and technology companies were therefore not exotic possibilities in Larry’s childhood; they were “the air” of the household.
David rejects the retrospective caricature of Larry and Sergey as bumbling academics who accidentally created a business. Larry later said, “Probably when I was 12, I knew I was going to start a company eventually,” because making the world better required doing “more than just invent things.”
Larry’s reasoning joined invention to capitalization: “You need to use business and entrepreneurship to make these things real. It’s not enough just to invent them.” Ben compares the underlying ambition to Bill Gates and Mark Zuckerberg, even if Larry’s public personality looked very different.
3. Sergey Brin supplied an intellectual equal, not a supporting co-founder
Sergey Brin was born in Moscow in August 1973 to a Jewish family living in a three-room apartment with his paternal grandmother. After his mathematician father attended an international conference and concluded the family needed to leave the Soviet Union, they emigrated; his father became a University of Maryland professor and his mother a NASA researcher.
Sergey graduated high school at 16, completed mathematics and computer-science degrees in three years, and reached Stanford at 19 after interning at Wolfram Research. His combination of mathematical ability, speed, and playfulness made him every bit Larry’s intellectual equal, with a somewhat more “zany” streak.
Their 1995 meeting began with argumentative chemistry: Sergey was helping show prospective student Larry around, while Larry challenged his claims about cities, transportation, and design. Anna Patterson’s recollection places their first evening at Menlo Park’s British Bankers Club, where they closed the bar and Charles Schwab unexpectedly picked up the group’s tab.
The hosts struggle to find a comparable enduring, equal founder partnership: “One plus one equaled like a hundred.” Larry and Sergey shared offices, sparred continuously, and retained significant overlap despite later stereotypes that assigned disciplined products to Larry and rollerblades, Glass, and skydiving to Sergey.
4. Exponential web growth made human curation a temporary solution
Between 1993 and 1996, the web expanded from 130 sites to more than 600,000, which the hosts calculate as 723% year-over-year growth for four years. Ben compares the signal to Jeff Bezos seeing internet growth and leaving D. E. Shaw: “Nothing like this has ever happened before.”
Larry’s initial dissertation proposal was not a search engine. Inspired partly by Yahoo’s hand-curated directory, he wanted a decentralized annotation layer in which people could leave useful comments directly on websites rather than submitting them to a central guide.
The problem appeared immediately: a major site could attract thousands or millions of annotations, requiring some way to separate authoritative comments from noise. Adviser Terry Winograd pushed Larry away from the messy annotation product and toward the more tractable computer-science problem of ranking authority.
5. PageRank translated academic citations into the native language of the web
Larry’s decisive leap was to apply ranking to pages rather than annotations: “The big problem here is not annotation. We should use it not for ranking annotations but for ranking searches.” The method borrowed from academia, where a paper’s importance depends not only on citation count but on the importance of the papers citing it.
A hyperlink was the web’s equivalent of a citation. PageRank could therefore treat links as votes while weighting votes from reputable pages more heavily, creating a recursively computed measure of authority rather than trusting raw keyword frequency.
Links carried something even richer than citations: anchor text. Ben notes that people linking to a site often describe it more accurately than the site describes itself, so the aggregate words used by outside authors became metadata about what the destination actually represented.
The implementation problem was that a page exposes only outgoing links, not every page linking into it—like knowing whose numbers are in your phone but not whose phones contain yours. Google therefore had to crawl and copy the web, reverse the link graph, and compute backlinks centrally.
6. A narrow historical window made crawling the whole web feasible
David’s timing claim is categorical: “Google could not have been built at any other time in history.” In 1996, crawling the whole internet was slightly insane but still conceivable as a university research project; only a year or two later, starting a complete index from scratch could have required tens or hundreds of millions of dollars.
Larry built PageRank and an initial crawler in Java, but the code was buggy enough that their friend Scott Hassan rewrote the system in Python. The Stanford project became BackRub—named for backlinks—at backrub.stanford.edu, where its page described a crawler intended to “improve web search engines,” not yet a standalone search engine.
Hassan left to start eGroups with Larry’s brother Carl Page; Yahoo later bought it for approximately $400 million. Watching a friend and sibling turn adjacent work into a Sequoia-backed company sharpened Larry and Sergey’s awareness that BackRub might have commercial value beyond a dissertation.
7. Existing portals rejected relevance because relevance broke their economics
In 1997, Larry and Sergey shopped BackRub to Excite, Yahoo, Infoseek, Lycos, and other search properties, hoping to license or sell it, help implement it, and return to their PhDs. The closest transaction was reportedly a roughly $1 million licensing arrangement with Excite.
A side-by-side test showed BackRub returning the desired result immediately, while Excite’s existing search forced users to click around. The Excite CEO reportedly killed the deal for precisely that reason: users leaving quickly meant fewer pages, impressions, and banner-ad dollars.
Ben’s pushback—worth keeping—is that the refusal was rational under the period’s CPM model. Search properties were becoming portals funded by page views, so a technology designed to transfer attention to other websites was “not strategic”; it attacked the business supporting the product.
Larry later summarized the divide: “These companies weren’t going to focus on search. They were becoming portals.” They “didn’t understand search and they weren’t technology people,” so after the licensing campaign failed, Google’s founders concluded they would have to build the product themselves.
8. Google’s minimal product spread fast enough to overwhelm Stanford
Before Google, the team considered “Whatbox,” a name describing the box users typed into, but rejected it partly because it sounded too much like a pornography site. “Google” came from the mathematical googol—one followed by 100 zeros—with the familiar spelling attributed to either a mistake or domain availability.
Sergey created an early multicolored logo, but the enduring interface was already present: colorful wordmark, search box, little else. That simplicity protected performance and made the user promise unmistakable—enter a need, receive relevant results, then leave.
By spring 1998, google.com was handling about 10,000 searches daily. The founders borrowed machines ordered for other Stanford projects, sometimes taking equipment from loading docks until its rightful researchers needed it; even with an asset-light page, usage eventually consumed roughly half the university’s bandwidth.
Stanford pushed the project to become a company before it toppled the network, receiving roughly 1% through the technology transfer. The institution’s patience supplied infrastructure when Google had no capital, while its equity ensured that generosity was economically rewarded.
9. A nonexistent company received a legendary $100,000 check
Stanford professor David Cheriton introduced Larry and Sergey to Sun Microsystems co-founder Andy Bechtolsheim. After an 8 a.m. demonstration at Cheriton’s home, Bechtolsheim reportedly said, “Great, I’m in. $100,000,” wrote a check to Google Inc., and drove away before anyone discussed documents or valuation.
Google Inc. did not yet exist, making the check a forcing function: the founders had to incorporate, transfer Stanford’s intellectual property, establish a bank account, and deposit it before expiration. Cheriton, Ram Shriram, and Jeff Bezos joined the financing, which the hosts characterize as $1 million at a $10 million post-money valuation.
Bezos invested $250,000 after Shriram arranged a meeting at his home. David estimates that an untouched stake could have been worth approximately $20 billion today; even a sale at Google’s IPO might have turned the check into roughly $200 million.
The new company moved into Susan Wojcicki’s Menlo Park garage. Wojcicki was then an Intel manager, later joined Google, and eventually became YouTube’s CEO—another example of the unusually concentrated network surrounding the company’s first office.
10. Search competition had three axes: relevance, index size, and speed
PageRank is remembered as the breakthrough, but Ben and David emphasize three independent product variables. A search engine needed to rank well, cover enough of the web to contain the answer, and return it quickly; excellence on only one dimension was insufficient.
AltaVista emerged from Digital Equipment Corporation’s Western Research Laboratory as a demonstration of DEC hardware. Its crucial innovation was parallel web crawling, which enabled an index of 16 million pages when competitors might cover around one million—but expensive enterprise servers burdened an industry with weak banner-ad monetization.
Existing engines commonly ranked pages by how often the query appeared, inviting “dog food” sites to repeat the phrase or hide keywords as white text on white backgrounds. Relevance degraded as commercial actors learned to game systems that treated repetition as authority.
Yahoo was the real juggernaut: already public in 1996 and worth roughly $20 billion by 1998. Its hand-curated directory worked while the web was small, then became a hybrid in which editorial results sat above search-engine backfill; Yahoo viewed itself as a media company whose technology only needed to be “good enough.”
11. Google rebuilt computing around an index too large for any machine
Google’s index could not fit on a single server, however expensive. The team broke it into many 64-megabyte chunks distributed across disks, machines, and eventually data centers, while a master server mapped where each chunk lived and returned only the pieces needed for a computation.
This technical problem helped attract extraordinary early engineers despite a fiercely competitive 1998-99 hiring market. Urs Hölzle arrived as employee number eight and “search engine mechanic” because “everything was broken”; Jeff Dean joined from DEC and became the hosts’ comparison to Microsoft systems legend Dave Cutler.
Dean’s eventual résumé included implementing AdWords, building AdSense, rewriting core search five times, and co-inventing or implementing Bigtable, MapReduce, TensorFlow, and Gemini. Hölzle ran infrastructure from 1999 through 2023, demonstrating how unusually long the first cohort’s impact lasted.
David sees recruitment as a knife-edge event: this was before Google could harvest displaced dot-com talent or offer a proven cash machine. The founders won top engineers because the product already felt magical, their ambition was persuasive, and the underlying distributed-computing problems were genuinely novel.
12. Unreliable commodity hardware became a software advantage
Hölzle extended the founders’ habit of assembling cheap components: use commodity machines, expect them to fail, replicate data three or five times, and design software around unreliability. Industry server failure ran around 3-4% annually; Google’s exceeded 10%, yet the overall system remained resilient.
Consumer memory lacked enterprise error-checking features such as parity bits, so Google built additional verification in software to guard against corruption—even bit flips caused by ambient radiation. The choice saved hardware expense but demanded deeper system design at every layer.
Early racks mounted motherboards on corkboard without cases because Google’s colocation contract charged by square footage rather than electricity or machine count. Next to Inktomi’s spacious cage of gleaming Sun servers, Google’s installation looked like “Frankenstein,” optimized for density and allegedly even borrowing a neighboring power circuit.
Google eventually became, technically, the world’s largest computer manufacturer because it bought components rather than finished servers. GFS and MapReduce grew from these constraints; Yahoo’s effort to copy MapReduce ultimately surfaced as Hadoop.
13. Infrastructure made speed, scale, and margin reinforce one another
Distributed software let Google expand its index and traffic more cheaply than rivals, while data centers around the world reduced latency. The results page displayed query time and result count because sub-second speed and billions of indexed pages were statistics no competitor could match.
The company designed file systems, racks, machines, and facilities as a single system instead of buying “enterprise-grade” reliability component by component. Ben and David estimate search eventually carried roughly an 87% gross margin, with AdWords near 85%.
The architecture began as a response to having little money, but it became a durable strategic advantage once monetization arrived. Google could fund explosive traffic growth without inheriting AltaVista’s enterprise-hardware cost structure, then reinvest the resulting margins into still more indexing, speed, and distribution.
14. Google raised its Series A on a business plan it barely believed
Employee number nine, recent Stanford graduate Salar Kamangar, was assigned to write the fundraising deck because Google was exhausting its seed capital. The proposed model had three parts: sell enterprise search appliances, accept conventional CPM advertising, and license organic results to portals.
Enterprise search was projected as the largest business partly because Stanford itself had used BackRub internally and Red Hat had paid $20,000 for Google search technology. Ben’s interpretation is that the founders considered google.com too special to monetize, so they hoped an adjacent enterprise business could subsidize it.
Sequoia’s Michael Moritz and Kleiner Perkins’ John Doerr jointly invested in a $25 million round at a $100 million post-money valuation, each taking roughly 12.5% and a board seat. Google amplified the supposedly unprecedented pairing in a carefully staged press conference, though the firms had shared deals before.
When revenue failed to materialize through late 1999 and the dot-com bubble began breaking, the episode recounts a lore-attributed line—“We’ve never paid so much for so little”—whose attribution is disputed among Michael Moritz, John Doerr, and Vinod Khosla. Google had extraordinary usage and loyalty, but it still lacked a business capable of funding people and infrastructure.
15. Text ads proved that search intent converted unusually well
Omid Kordestani joined from Netscape to convert the three-part plan into revenue. Enterprise demand remained weak, so he recruited Tim Armstrong in New York and began selling text-only ads manually—often through faxed insertion orders—on a CPM basis against specified keywords.
Larry and Sergey insisted on text because banners would slow the page and degrade the product. Madison Avenue initially found the format visually unexciting, but Google’s pitch was mathematical: search exposed commercial intent at the moment a user was actively seeking something.
Jeff Dean and Marissa Mayer tested the thesis using dynamically generated Amazon affiliate links for books related to queries. Because affiliate reporting revealed both clicks and purchases, Google could show not merely higher click-through rates but stronger downstream conversion: “You’re going to like the numbers.”
16. Portal licensing made “Powered by Google” an ingredient brand
Kordestani’s Netscape relationship produced a deal to backfill organic results for Netscape’s directory. When launch traffic threatened to overwhelm Google, the company shut down google.com for a day and prioritized Netscape rather than risk failing its first major partner.
The decision looks astonishing beside Google’s later reputation for universal availability, but the economics and distribution were decisive. Netscape delivered roughly three million searches per day, while every results page exposed new users to “Powered by Google.”
What began as branding Google had to beg partners to include became a selling point: portals advertised that their search came from the highest-quality provider. Ben describes it as the “Intel Inside of search,” turning a behind-the-scenes supplier into the consumer brand users later visited directly.
17. Yahoo’s 2000 deal carried Google through the dot-com winter
In June 2000—three months after the Nasdaq peak—Yahoo selected Google to provide organic backfill, included “Powered by Google,” and invested $10 million. Google’s daily searchers immediately doubled to approximately 14 million.
Yahoo paid Google $7.2 million for organic results in 2001. Combined with the investment, Netscape, and smaller portal agreements, that revenue bridged an infrastructure-heavy startup through a period when additional venture financing was effectively unavailable.
David stresses how young Google still was: modern startups are rarely expected to become profitable within two years, but Google could not finance continuing losses after the crash. Revenue was therefore not a cosmetic milestone; it was the only remaining survival mechanism until AdWords matured.
18. Eric Schmidt joined a founder-led triumvirate, not a conventional hierarchy
The Series A investors had extracted a promise that Larry and Sergey would hire a “professional CEO,” but the founders delayed for roughly 16 months. After touring prominent Silicon Valley CEOs, they said the only candidate meeting their standard was Steve Jobs—part admiration, part refusal to lower the bar.
Eric Schmidt eventually fit the Venn diagram: engineer, former Sun executive, public-company CEO at Novell, experienced business operator, and fellow Burning Man attendee. He joined in March 2001 while Larry became president of products and Sergey president of technology; the three met daily and effectively ran Google together.
Schmidt entered with restraint, even sharing his office with an engineer when space was tight. The company’s “googliness” combined university-like collaboration, extremely high intellectual horsepower, inexperienced people unburdened by conventional limits, and a “healthy disregard for the impossible.”
Google’s 1999 mission—to “organize the world’s information and make it universally accessible and useful”—proved unusually scalable. Its IPO prospectus later connected altruism to economics: “The most effective and ultimately the most profitable way” to accomplish the mission was to put users first.
19. GoTo.com discovered the commercial architecture of search
Idealab founder Bill Gross launched GoTo.com at TED in February 1998 with a radical claim: the free market could rank commercial relevance. A business able to solve the user’s need should convert better and therefore afford a higher bid than an irrelevant spammer.
GoTo listed paid results in bid order and displayed each price publicly—21 cents, 23 cents, 24 cents. Advertisers used a self-service website and paid only when users clicked, replacing the historical advertising problem of paying for exposure without knowing which half was wasted.
The model worked immediately: GoTo produced approximately $100 million in first-year revenue and had 8,000 advertisers by mid-1999. For comparison, Google’s October 2000 AdWords beta began with only 350 advertisers.
Gross did not patent the core auction or pay-per-click ideas because they seemed obvious; by the time lawyers raised the issue, his TED disclosure was outside the one-year filing window. Smaller patents later led Google to pay roughly $360 million, but the essential mechanisms entered the industry freely.
20. Overture proved that paid-search distribution could dominate portals
GoTo became Overture as it shifted from destination search engine to B2B advertising supplier, powering paid listings for smaller properties and then AOL, MSN, and Yahoo. Yahoo alone represented a roughly $100 million deal, and the hosts estimate Overture eventually supplied about 75% of Yahoo’s revenue.
Yahoo ultimately bought Overture for $1.6 billion because so much of its business depended on the vendor. The price consumed a major portion of Yahoo’s post-crash market capitalization, but allowing its principal revenue engine to remain independent posed an even larger risk.
Gross also tried to acquire Google before its Series A. Larry and Sergey reportedly proposed a $200 million price; Gross saw paid commercial results and PageRank-driven informational results as a “match made in heaven,” but Overture’s board refused to surrender about 10% of its $2 billion company for a startup with no revenue.
Ben’s counterfactual is stark: had the acquisition happened, Google almost certainly would not have become Google. The episode instead credits Larry and Sergey with a second superpower beyond invention—the ability to recognize an outside idea, adopt it, and execute it more completely.
21. Ad Rank aligned relevance with Google’s expected revenue
Kamangar’s AdWords overhaul first added self-service and ad-quality measurement, opening the system to a long tail of advertisers without scaling a matching salesforce. The initial release still charged CPM, however, allowing advertisers to improve measured click-through rates by clicking their own ads at no incremental cost.
Google then adopted Overture’s cost-per-click auction and added its distinctive ingredient: click-through rate influenced placement alongside the bid. Ad Rank became the paid analogue to PageRank, treating user behavior as evidence that an advertisement was relevant.
The formula was also the expected-value-maximizing rule for Google: price per click multiplied by probability of a click. A somewhat lower bid with much higher engagement could produce more revenue than an expensive ad users ignored.
Incentives aligned across the system. Relevant advertisers could win while bidding less, users saw ads more likely to solve their needs, and Google maximized revenue without simply awarding the page to whoever paid most. “Click-through rates are a proxy for relevance.”
22. Second-price auctions stored trust instead of extracting every penny
Google’s auction charged the winner only one cent more than the next-highest bid. An advertiser bidding 50 cents against bids of 30 and 20 would win at 31 cents—apparently leaving money behind, but reducing fears of gouging and eliminating constant bid micromanagement.
Ben describes this as “stored potential energy”: Google declined immediate extraction to create advertiser trust, liquidity, and durable participation. Sheryl Sandberg learned from mentor Larry Summers that the mechanism was a Vickrey second-price auction, supported by economic literature and used in selling Treasury bonds.
Moving existing advertisers onto the full CPC system was painful enough to receive the internal name Project Sunset. Overture had reportedly experimented with incorporating click-through rate but abandoned it when transparent bid ordering became confusing; Google endured the transition and built the infrastructure to run dynamic auctions at enormous scale.
The company risked much of its existing business during the migration. Google recorded $86 million of revenue and $10 million of profit in 2001, largely from portal licensing and the earlier advertising system, so replacing the model threatened real—not hypothetical—cash flow.
23. Internationalization and AdWords changed Google’s financial slope
Schmidt discovered that roughly 50% of searches came from outside the United States, while almost no international advertising business existed. He sent Kordestani traveling to form startup-like local teams; international revenue reached 18%, then 22% in 2002 and 29% in 2003, eventually becoming about half the company.
Google’s China presence ended only later, after conflict over censorship led it to withdraw rather than continue censoring results in 2010. Everywhere else, the combination of a globally distributed product and localized ad selling converted overseas usage into a durable business.
In 2002, amid the AdWords transition, revenue increased more than fivefold to $440 million, with profits stated at $185 million. Google realized it should stop charging portals for organic search and instead pay them to run Google’s ads and share the resulting revenue.
Yahoo’s acquisition discussions traced the acceleration. CEO Terry Semel offered roughly $1 billion in late 2001 after observing that Yahoo, supposedly Google’s largest customer, paid less than $10 million; Yahoo later offered $3 billion, while the founders countered at $5 billion—effectively a price requiring Google to take over Yahoo.
24. The AOL agreement was a literal bet-the-company expansion
AOL brought 34 million users, but its organic search came from Inktomi and paid listings from Overture. Google offered both layers, required advertisers to transact through AdWords, and returned 85 cents of each advertising dollar to AOL—bringing AOL’s inventory and advertisers directly into Google’s market.
AOL also received warrants for 7.4 million Google shares at $3 each, representing a potential $22 million investment, plus a $100 million revenue guarantee. Google did not have $100 million available when negotiating the agreement in May 2002.
Sergey admitted, “We could have gone bankrupt.” When Kordestani warned that the company was being wagered, Larry answered: “We should be able to monetize the pages. If not, we deserve to go out of business.”
The bet worked: AOL earned $35 million during the partial first year and $200 million in 2003, comfortably exceeding the guarantee. Sandberg then built the organization servicing thousands of new advertisers while feeding repetitive work back into product and engineering so the operation could scale without proportional headcount.
25. Search economics generated increasing returns, not merely lower costs
Traditional scale economies reduced unit costs by spreading infrastructure and engineering across more searches. Google added something rarer: revenue per search rose with scale because more advertisers produced deeper auctions, better price discovery, higher winning bids, and bids on obscure keywords that smaller systems could not monetize.
Ben and David trace the flywheel: distribution creates queries; queries attract advertisers; more bidders increase prices and coverage; revenue funds greater distribution. Because every user was worth more to Google than to a smaller rival, Google could sustainably pay more to acquire that user.
The company sometimes shared 100% of near-term revenue with distribution partners and, in at least one cited case, more than 100%. Being private made such apparently irrational quarters easier to tolerate, while Google’s belief that search would become vastly larger than a $1 billion or $10 billion market justified investing ahead of reported economics.
A Wall Street Journal comment about AOL’s decision warned that “Google takes your users. It doesn’t help you build your property.” The transcript does not clearly identify the speaker; the warning proved directionally correct, but portals could not easily reject Google when its superior monetization allowed it to offer the best revenue share and its brand improved users’ search experience.
26. Toolbar converted Microsoft’s browser control into Google distribution
Internet Explorer owned the browser, giving Microsoft power to direct users elsewhere if it recognized search’s economics. Google’s December 2000 Toolbar inserted a search field into that browser years before browsers merged search and URL entry, serving simultaneously as distribution offense and Microsoft defense.
A Toolbar user generated roughly seven times as many searches. With an ordinary user producing about $2 in annual revenue and a Toolbar user worth more than $10, Google had approximately $8 per user available for acquisition before considering future growth in revenue per user.
Google paid Adobe, RealNetworks, WinZip, Dell, and others to bundle Toolbar with software or computers. The mechanism resembled a Trojan-horse installation, but users valued instant search and the integrated pop-up blocker; Google turned potentially objectionable distribution into a useful product.
Firefox default placement became Mozilla’s principal revenue source, while Google Earth installs carried Toolbar instead of needing Earth-specific ads. Google Desktop and other applications pursued the same objective: make users stickier search customers. In 2004, a new product manager named Sundar Pichai inherited this strategically central client team.
27. AdSense extended Google’s auction from searches to the rest of the web
AdWords monetized the relatively rare moment when a user expressed explicit intent. Google realized its page-understanding and ad-ranking systems could also examine static publisher content, infer the associated keywords, and serve relevant advertisements without any search query.
Jeff Dean built the initial AdSense system in approximately six weeks, first testing on Google Groups and then buying placements on third-party sites. HowStuffWorks became a favorite laboratory because its explanatory pages frequently mapped to highly commercial AdWords categories.
Publishers could insert a small amount of HTML and receive money from Google automatically. Estimates in the episode place their revenue share at either 67% or 80%, leaving Google with lower margins than first-party search but access to vastly more page inventory and making the model a precursor to YouTube creator monetization.
AdSense launched in 2003 and exceeded $1 million per day by year-end. Its earliest conceptual seed came from Paul Buchheit’s Gmail prototype: after searching a user’s Unix mail directory, he displayed ads from Google’s database, prompting Larry and Sergey to ask, “Does this work on websites, too?”
28. Continuous reinvestment turned the original algorithm into a system
PageRank started Google, but the company never treated search as solved. By 2007, ranking reportedly used around 200 signals, including behavior such as users repeatedly bouncing from a result, personalization, and inferred synonyms—for example, learning that “cat” and “kitten” were related from query reformulations.
Twenty-percent time and Google Labs surfaced projects such as Google News, motivated partly by the post-September 11 need for rapidly updating coverage across many sources. Gmail began similarly as a search box pointed at mail, then expanded toward a logged-in identity and another entry point into Google’s ecosystem.
Infrastructure moved from opportunistic commodity hardware to proprietary data centers and system software, while search, ads, and user behavior continuously supplied more data. The web’s content was growing and changing too quickly ever to reach a final optimum, so investment itself became a permanent operating requirement.
Revenue reached approximately $1.5 billion in 2003 with nearly $350 million in operating income. AdSense diluted reported margin by adding roughly half a billion dollars of lower-margin revenue, but after 2002 Google was effectively no longer capital constrained: it could fund every credible idea.
29. The IPO preserved control but failed its own pricing test
Google did not need IPO proceeds, but the pre-JOBS Act 500-shareholder rule would soon have required public financial disclosure anyway. Employees and investors wanted liquidity; roughly half of Google’s 2,000 workers became millionaires through the offering.
Larry and Sergey borrowed dual-class governance from family-controlled media companies, giving founders super-voting shares even as public investors gained most of the economics. Google pioneered the structure for technology IPOs later used by companies including Meta, Alibaba, Shopify, Spotify, Coinbase, Airbnb, Zoom, and Datadog.
To avoid the customary IPO “pop” transferring value from existing owners to favored banking clients, Google chose a Dutch auction: begin high, reduce the price until enough demand clears the offering, and give retail investors broader access. Google engineers even helped build the auction software.
The proposed range was $108-$135, but the deal priced at $85, raising $1.7 billion at a $23 billion valuation. Shares closed the first day near $100—an 18% pop—and approached five times the IPO price within 16 months, so the mechanism neither discovered the durable price nor prevented money being left behind.
30. Google’s castle now faces the technology wave it helped create
In the episode’s comparison, Alphabet was worth roughly $2.1 trillion, around 100 times its IPO capitalization before reinvested dividends. Yet it traded near 20 times earnings, versus Meta at 27, Apple at 30, Amazon at 35, Microsoft at 37, and Nvidia at 46, despite continued revenue growth.
Ben says the irony is that Google “invented AI” and published the Transformer paper—material reserved for later episodes, but central to why the old search castle may or may not survive.
Ben’s consolidated playbook starts with relevance, then adds the best execution of search advertising, first-principles infrastructure, exceptional hiring, ambitious culture, data feedback, distribution, and an enduring mission. No individual component explains the result; each made the others economically stronger.
Their primary strategic power is scale, amplified by marketplace liquidity: Google’s cost per query fell while revenue per query rose. The broader historical lesson is timing—search from 1996 to 2002 resembles the present AI transition because “history doesn’t repeat, but it rhymes.” Google was “that company”; the unresolved question is who holds that position now.