Google Part I: Origins of Search. How the Best Business in Human History Happened (Audio)

2025-06-30 - source - Read full transcript

Ben Gilbert (host)David Rosenthal (host)

Key insights

PageRank was not originally conceived as search technology at all.

Larry Page's Stanford dissertation started as a system for annotating web pages with user comments. The team realized ranking annotations required a way to judge site authority, borrowed the academic-citation model (papers ranked by who cites them), and only then realized the same ranking method applied directly to ranking web pages themselves for search.

search-technology-origins

Google's founders repeatedly refused to sell or license the technology because every buyer's business model conflicted with better search.

Excite's CEO killed a near-done licensing deal because faster, more relevant search meant fewer page views and less banner-ad revenue for Excite. Yahoo passed on PageRank for $1M for the same reason. The episode frames this portal-vs-search conflict of interest as the central reason Google eventually beat the incumbents rather than being absorbed by them.

business-model-evolution

Google was forced into a distributed, commodity-hardware infrastructure model by necessity, and that became a durable cost advantage.

The web's link graph was too large to fit on any single machine, so Google split its index into small chunks spread across cheap, unreliable commodity servers (with roughly 10%+ annual hardware failure rates versus 3-4% industry average) and relied on software-level redundancy instead of expensive enterprise gear. This let Google scale far more cheaply than rivals like AltaVista, which ran on expensive DEC hardware, and is cited as a direct driver of search's ~87% gross margin.

infrastructure-as-moat

Google's ad ranking system (Ad Rank) is mathematically identical to Google's own revenue-maximizing formula.

By ranking ads on a combination of bid price and predicted click-through rate rather than price alone, Google simultaneously rewards relevant advertisers with lower prices, gives users more relevant ads, and mathematically maximizes Google's own expected revenue per query - the incentives of advertiser, user, and platform all align in the same formula.

business-model-evolution

Google's early growth depended as much on aggressive, sometimes surprising distribution deals as on product quality.

Beyond organic word of mouth, Google paid massive revenue shares (up to 100% in some cases) to portal partners, bundled the Google Toolbar into installers for Adobe, RealNetworks, WinZip, and Google Earth to boost per-user search volume roughly sevenfold, struck default-search deals with Dell and Firefox, and used the resulting toolbar-driven ARPU increase to justify paying more per acquired user than any competitor could afford.

distribution-strategy

Search auctions produce 'super' scale economics: revenue per user rises with scale, not just costs falling.

Because ad slots are sold via auction, a larger pool of bidders on a keyword produces higher clearing prices (better price discovery) and a larger pool of advertisers means more searches get monetized at all. This means Google's per-search and per-user revenue actually increases as it scales, on top of the usual unit-cost economies of scale, creating a compounding advantage over smaller search engines.

scale-economics

The 2002 AOL deal was a genuine bet-the-company move that established Google as the dominant paid-search player almost overnight.

Google agreed to share 85% of ad revenue with AOL and guaranteed AOL at least $100 million even though Google didn't have that much cash and AdWords v2 had only just launched. Sergey Brin said outright they could have gone bankrupt. The deal paid off (AOL alone generated $35M in the second half of 2002 and $200M in 2003) and gave Google the advertiser liquidity and market position that Overture (AOL's previous partner) had held until then.

distribution-strategy

AdSense extended Google's ad business from search-query moments to essentially all web content.

Realizing users spend far more time consuming content than running searches, Jeff Dean built AdSense in six weeks to reverse-apply Google's existing keyword-matching and ad-quality algorithms to static publisher pages, effectively 'reverse serving' the ads a search query on that content would have triggered. It reached over $1M/day in revenue within months of its 2003 launch, at lower margin than AdWords but with an entirely new, much larger inventory of monetizable moments.

business-model-evolution

Google's timing window for building a crawlable, indexable search engine was narrow and largely accidental.

Larry and Sergey started BackRub in 1996 when the web was small enough that copying and backtracing the entire link graph was expensive but not yet prohibitive. A few years earlier and the web would have been too small to require technology-driven search (Yahoo's directory model was still adequate); a few years later and building a full index from scratch would have cost tens to hundreds of millions of dollars, making it commercially infeasible.

search-technology-origins

Google's founders adopted competitors' best ideas rather than insisting on originating everything themselves.

The core AdWords auction mechanics - self-serve bidding, real-time auctions, pay-per-click pricing - were pioneered by Bill Gross's GoTo/Overture, not invented by Google. Google's contribution was adding a click-through-rate quality signal (Ad Rank) and a second-price auction on top of a model it explicitly borrowed once GoTo proved it worked at scale.

business-model-evolution

Google's Dutch-auction IPO failed at its stated goal, but its dual-class share structure became an industry standard.

The 2004 IPO used a Dutch auction specifically to avoid the typical banker-driven underpricing 'pop,' but the stock still priced at $85 (below the $108-135 target range) and popped to $100 on day one; no major company has repeated the Dutch-auction mechanism since. By contrast, the founder-control dual-class structure Google borrowed from family newspaper companies became the template used by Meta, Alibaba, Snap, Airbnb, and nearly every major tech IPO afterward.

infrastructure-as-moat

Books referenced

The Search - John Battelle - Cited for the stat that the web grew from 130 sites to over 600,000 between 1993 and 1996, a ~723% annual growth rate.
In the Plex - Steven Levy - Primary source for the episode; quoted for the Excite/Vinod Khosla licensing story and Justin Rosenstein's 'that company' email about Google.
Seven Powers - Hamilton Helmer - Framework the hosts use in the 'playbook' segment to assess which of the seven competitive powers (scale economies, network economies, branding, etc.) Google possessed.
Googled: The End of the World As We Know It - Ken Auletta - Quoted for the Omid Kordestani/Larry Page exchange about betting the company on the AOL deal ('If not, we deserve to go out of business').

Media referenced

The Rehearsal (season 2) - show - Ben's carve-out; Nathan Fielder's extreme-commitment comedy series about reducing plane crashes through simulated pilot-communication rehearsals.
Your Friends & Neighbors - show - Ben's carve-out; Apple TV show starring Jon Hamm, compared to Succession.
Andor (season 2) - show - Ben's carve-out; Disney+ Star Wars series, called some of the best Star Wars canon.
GameCraft (season 3) - podcast - David's carve-out; gaming industry podcast hosted by Mitch and Blake.

Companies

Google - Subject of the episode; founded 1998 by Larry Page and Sergey Brin out of Stanford PhD research.
Stanford University - Where Larry Page and Sergey Brin met, built BackRub/PageRank as a research project, and hosted Google's early infrastructure until it outgrew the campus network.
Excite - First company Larry and Sergey tried to license BackRub to; the CEO killed the deal because better search would reduce page views and banner-ad revenue.
Yahoo - Dominant portal of the late 1990s; rejected buying PageRank for $1M, later became a huge Google organic-search customer, then bought Overture for $1.6B, then tried to buy Google for up to $3B.
Alta Vista - Spun out of DEC's research lab; pioneered parallelized web crawling and had the largest index, but had poor relevancy and no real business model.
Inktomi - Competing search engine that white-labeled organic search results to portals (the same model Google's early business plan proposed); lost AOL's business to Google in 2002.
DEC (Digital Equipment Corporation) - Built AltaVista as a showcase for its enterprise servers; also where Jeff Dean worked before joining Google.
GoTo.com / Overture - Bill Gross's company that invented pay-per-click, auction-based paid search; grew to $100M revenue in year one, was acquired by Yahoo for $1.6B, and tried to acquire Google for $200M before Google existed as a real business.
AOL - 2002 deal made Google its exclusive search/ads provider with an 85% revenue share and a $100M guarantee Google could not yet cover; the deal made Google the dominant paid-search player almost overnight.
Netscape - First major portal deal (1999) for Google's organic search backfill; traffic spike from the deal nearly overwhelmed Google's infrastructure.
Sequoia Capital / Kleiner Perkins - Co-led Google's $25M Series A at a $100M post-money valuation, jointly (an unusual arrangement); pushed Larry and Sergey to hire a professional CEO.
Amazon - Jeff Bezos was one of four Google seed investors ($250K); Google's first ad-monetization test used Amazon affiliate links to prove intent-based ads converted.
Microsoft - Feared as the company most able to weaponize browser control (Internet Explorer) against Google; motivated the Google Toolbar and other distribution defenses.
Red Hat - Google's first paying customer ($20,000), for an enterprise intranet search license, part of the original three-pronged Series A business plan.

Techniques and frameworks

PageRank - Ranks web pages by treating hyperlinks as academic-citation-style votes of authority, using anchor text as descriptive metadata.
Distributed file system / commodity hardware architecture - Google split its index into small chunks across many cheap, failure-prone machines (GFS precursor) because the index couldn't fit on one server and enterprise hardware was unaffordable.
Second-price (Vickrey) auction - Google's AdWords ad auction charges the winner only one cent above the second-highest bid, building long-term advertiser trust at the cost of some short-term revenue.
Ad Rank - Combines bid price with predicted click-through rate to rank ads, which also happens to be the mathematically optimal formula for maximizing Google's own ad revenue.
AdSense keyword-matching - Reverse-applies the same ad-targeting algorithms used for search queries to static publisher pages, matching ads to page content instead of search intent.
Dutch auction IPO - Reverse-price-discovery IPO mechanism Google used in 2004 to avoid banker-driven underpricing; the approach has not been repeated by a major company since.
Dual-class share structure - Google borrowed this from family-controlled newspaper companies to keep founder voting control after going public; became the template for nearly every major tech IPO since.

Summary

This episode opens Acquired's multi-part Google series by tracing the company from Larry Page and Sergey Brin's 1995 meeting at Stanford through its 2004 IPO, arguing that Google's dominance was never accidental. Larry and Sergey's PageRank algorithm grew out of a dissertation project about annotating web pages, which led them to borrow academic citation-ranking methods and apply them to hyperlinks - discovering, almost by accident, a far better way to rank search results than the keyword-stuffing approaches used by Alta Vista, Excite, Infoseek, and Lycos. Crucially, the hosts frame Google's early rejections (Excite wouldn't license BackRub because better search meant fewer page views; Yahoo passed on PageRank for $1M) as revealing a structural conflict of interest baked into the portal business model of the era, one Google was free of because it built its own company instead of selling the technology.

The episode spends significant time on Google's infrastructure as a second, underappreciated pillar of its success. Unable to afford enterprise-grade hardware like DEC's servers (which powered Alta Vista), Google built a distributed system that split its index into small chunks spread across cheap, unreliable commodity machines, tolerating a 10%+ annual hardware failure rate through software-level redundancy. This became a durable cost advantage that the hosts connect directly to search's roughly 87% gross margin. Early engineering hires Urs Hölzle and Jeff Dean are credited as essential to making this work, recruited during the height of the dot-com bubble despite Google having an uncertain business model at the time.

The business-model arc is the episode's second major thread: Google's Series A pitch (enterprise search licensing, banner ads, OEM search backfill to portals) largely failed, and the company survived the dot-com crash mainly through portal deals with Netscape and then Yahoo (which invested $10M and paid Google for backfill search). The real breakthrough came from watching Bill Gross's GoTo/Overture prove that pay-per-click, auction-based paid search worked at scale, then improving on it with Ad Rank - a formula combining bid price and click-through rate that happens to be mathematically identical to the formula that maximizes Google's own revenue. The 2002 AOL deal, in which Google agreed to share 85% of ad revenue and guaranteed AOL $100 million it didn't yet have, is framed as a genuine bet-the-company decision that, once it succeeded, made Google the dominant paid-search player almost overnight and unlocked a virtuous cycle: more searches drive higher-liquidity auctions, higher liquidity drives higher revenue per user, and higher revenue per user lets Google outspend anyone on distribution.

That distribution aggressiveness gets its own extended treatment: the Google Toolbar (bundled into Adobe, RealNetworks, WinZip, and Google Earth installers) roughly septupled a user's search volume, letting Google pay far more than competitors to acquire users, while deals with Dell and Firefox further embedded Google as the default search experience. AdSense, born out of Gmail prototyping work by Paul Buchheit, extended the ad business from the moment of a search query to essentially any content on the web. The episode closes with Google's unusual, VC-forced 2004 IPO: a Dutch auction meant to avoid IPO-day underpricing (which still mispriced the stock, and which no major company has repeated) paired with a dual-class share structure borrowed from family newspaper companies, which became the template for nearly every major tech IPO since.

Throughout, the hosts (Ben Gilbert and David Rosenthal) repeatedly draw parallels to the current AI moment, arguing the 1996-2002 window they're describing "rhymes" strongly with 2021-2025, and frame the episode explicitly as scene-setting for later installments on Google's AI history.

Notable Quotes

"It wasn't that we intended to build a search engine. We built a ranking system to deal with annotations." - Larry Page (quoted by David Rosenthal)

"We could have gone bankrupt. This is quite literally Google betting the company." - Sergey Brin (quoted by Ben Gilbert)

"You're betting the company if you do that... We should be able to monetize the pages. If not, we deserve to go out of business." - Larry Page (quoted by Ben Gilbert, from Ken Auletta's book)

"It's more ads equals better ads equals better business." - David Rosenthal

"History doesn't repeat itself, but it does rhyme." - David Rosenthal