The eighth decade of McKinsey’s journey, from 1996 to 2006, saw digital business models arrive, collapse, and emerge again on firmer ground. The previous decade built a new digital infrastructure: the web browser, enterprise software, and the first mass-market mobile phones. This decade tested what all of that technological progress was worth.
Capital arrived before the answers did. Deregulated telecom markets, falling computing costs, and a rush of retail investors into equity markets made it possible to fund a company on a story about a market that did not yet exist. Then the market changed its mind. After reaching a peak in 2000, the tech-heavy Nasdaq Composite Index had dropped nearly 80 percent by the fall of 2002. And the dot-com crash wasn’t just a two-year phenomenon; markets would remain largely flat for nearly a decade—illustrating that innovation can move fast, but extracting durable value from it is much harder.
Yet 1996–2006 was also a crucible of real progress that created the underpinnings of today’s tech-driven economy. By 2006, a small group of companies had built digital models that generated real cash: online advertising, marketplaces earning transaction fees, and music sold by the track, for example. And though many early e-commerce companies didn’t survive, the model did: Online shopping was here to stay. For today’s technology leaders, the sequence should feel familiar. AI capability is advancing quickly, and capital is abundant. What remains scarce is the understanding needed to connect the two, which is why realizing the full value of AI is today’s biggest challenge.
The internet becomes a place to do business
Amazon went public in May 1997 as a bookseller. Around two years later, it was selling toys, electronics, and CDs and had opened its site to third-party sellers who supplied inventory Amazon never had to buy. The marketplace model took off, with eBay going public in September 1998 and turning a profit almost immediately by charging fees to match buyers with sellers.
Music showed that web business models sometimes clashed with traditional industries. Music-sharing service Napster launched in June 1999, and its verified user base peaked at over 26 million subscribers worldwide in February 2001. The Recording Industry Association of America sued Napster in December 1999, and the service shut down less than two years later. The music industry won the case but still lost the decade because litigation did nothing to stop the demand for digital music that Napster had exposed. Apple capitalized on that demand by building a legal sales channel. After opening the iTunes Music Store in April 2003 with 200,000 tracks at 99 cents each, it sold more than a million songs in the first week. But questions over who owned the rights to music online, how much people would be willing to pay for it, and how artists should be compensated only intensified.
100 years of technology innovation
As McKinsey celebrates its centenary in 2026, this series explores the defining technologies of each decade and the lessons they offer for today’s technology leaders.
For today’s technology leaders, timing is the lesson. Demand for digital music was evident before a legal and repeatable business model existed to meet it. AI is now producing comparable signals, with employees adopting agentic tools faster than their organizations can build business use cases. The companies successfully extracting value from agentic AI can spot bottom-up usage patterns and then build guardrails to turn adoption into value at scale.
Search finds a business model
The web was a Wild West in the early 2000s, and finding content was the major business problem to solve. Google was founded in September 1998, but what turned it into a business was AdWords, launched in 2000, which auctioned advertising against search terms. Google went public on August 19, 2004, at a valuation of roughly $23 billion, four years after the dot-com crash and already profitable. Google’s early success in monetizing search paved the way for it to become a massive all-purpose technology company with a market cap of over $4 trillion today.
Google’s search algorithm earned it the traffic, but the ad model converted those searches into revenue to fund the company’s expansion. Many organizations now have real AI capabilities but lack a comparable mechanism to translate usage into revenue, which is why building an operating model for agentic AI must include an economic framework.
If you build it, they will come
A network has to be built before business can be built on top of it. And this decade showed that the wait between the two is expensive. In October 2001, NTT DoCoMo switched on a commercial 3G network in Tokyo and Yokohama. Consumer uptake was initially tepid. Mobile handsets were bulky, and coverage was patchy. But DoCoMo kept investing in its network buildout, and by 2008, the same network carried more than 40 million users.
Capacity built ahead of proven demand can look like poor judgment but in hindsight can appear prescient. Many business leaders now face this unknown. They are funding compute, data platforms, and agent orchestration for workloads that do not yet exist. (And on a far wider scale, data centers to support AI infrastructure could require nearly $7 trillion in investment by 2030.) As AI costs spiral, CIOs need to make hard decisions about which investments will deliver real value. They will need to retool their infrastructure for and with agentic AI, making near-term architectural decisions with long-term economic results in mind.
Capital surged ahead of business value
The decade’s failures were rarely failures of technology. Webvan is the clearest case because the idea was sound. Backed by $375 million raised in its 1999 IPO, the online grocer built automated warehouses and opened in ten markets before it had shown that any single one of them could make money. It filed for bankruptcy in July 2001 due to expensive logistics that couldn’t translate into a profitable business. Groceries weren’t the only failure. In Europe, the online fashion retailer Boo.com spent $135 million in 18 months building an overly complicated site so slow that most customers never completed purchases. The company soon went into receivership. Online grocery delivery and the online sale of high-end fashion were legitimately strong business concepts; the failures of these companies were in execution.
Today, agentic economics demands the same arithmetic vigilance that Webvan and Boo.com skipped. Technology leaders must think beyond deploying agents for technology’s sake alone. They must think about cost to serve, unit margins, and the path from pilot to profit.
Learning from the dot-com bust
The decade from 1996 to 2006 was a painful reminder that just because a technology (like the internet then and AI now) is a truly disruptive innovation doesn’t mean businesses can easily be built on top of it. Napster proved demand for online music but captured little revenue from it. Webvan proved people would buy groceries online, but it could not do so at a palatable cost to consumers. Today, online music is a huge industry, but not in the buying-one-song-at-a-time way that iTunes pioneered. Instead, subscription-based streaming rules. And grocery delivery is now a multibillion-dollar market, though not in the way Webvan envisioned it. Nearly three-quarters of sales happen on grocery retailers’ sites, with only one-quarter coming through today’s delivery intermediaries such as Instacart and DoorDash.1
For leaders investing in AI, it’s worth remembering how the internet economy evolved. Billions were invested in early web models, but only a few companies found ways to turn the internet’s potential into profitable, long-lasting business use cases. Today, after a few years of frenetic AI adoption, parallels are emerging. Companies are moving from indiscriminate spending toward measuring where real value can be created at scale. As the early web proves, technology alone is not a business model. Leading companies are rewiring their organizations around AI rather than just adding it on top of existing processes. Like the web, AI is more than a capability. It’s an entirely new way to structure processes, teams, and operating models.
Chandrasekhar Panda is a partner in McKinsey’s Riyadh office, Henning Soller is a partner in the Frankfurt office, Klemens Hjartar is a senior partner in the Copenhagen office, and Sven Blumberg is a senior partner in the Düsseldorf office.
1. Arielle Feger, “Walmart, Amazon remain on top as the digital grocery market matures,” eMarketer, January 28, 2025.