Fragmented

Chapter 1: The Bargain

In 1998, when Google was still a Stanford dorm room project, the internet had a problem that no one could quite solve. There was suddenly too much of everything. The World Wide Web, which had begun as a collection of linked academic papers, had exploded into billions of pages—most of them garbage, some of them useful, all of them floating in an undifferentiated sea. Alta Vista indexed 100 million pages and couldn't keep up. Excite and Lycos tried to organize the chaos with human editors and crude algorithms, and they were drowning. The question that consumed the brightest minds in Silicon Valley was brutally simple: How do you make money from infinite content and finite human attention?

Google's answer revolutionized more than search. It created the foundational bargain of the digital age.

The bargain worked like this: You would give Google your attention and your data. You would click on search results, follow links, spend minutes and hours on the platform. Google would track what you searched for, what you clicked on, where you went next. The company would learn your interests, your fears, your shopping habits, your health concerns, your political leanings. You would generate this data continuously, day after day, and you would do it for free. In exchange, Google would give you something of extraordinary value: a unified interface to human knowledge, indexed and ranked in seconds, tailored to what you actually wanted to find.

It was the best deal the internet had ever seen.

From a user's perspective, the mathematics was irresistible. You typed a query into a simple white box, pressed Enter, and received a ranked list of the world's most relevant pages. No ads cluttered the search results themselves—not yet, anyway. The experience was so elegantly efficient that it barely registered as a transaction. You weren't aware you were trading anything. You were simply searching the internet, the way you might browse an encyclopedia.

But underneath that simplicity, Google had engineered something unprecedented: a system that could monetize attention at scale. The data you generated while searching—your queries, your clicks, your behavior patterns—became immensely valuable to advertisers. If Google knew that you were searching for "knee pain" and "orthopedic surgeon," the company could sell that information to medical device companies and surgical clinics. If you searched for "best credit cards for travel," credit card companies would pay premium prices to show you their ads. Google extracted value from your data not by selling it directly to advertisers, but by selling them access to your attention. The company knew who you were and what you wanted, and it could show you ads you were likely to click on.

This created a feedback loop. The more data Google collected, the better it could target ads. The better the targeting, the more advertisers would pay. The more money Google made, the more it could invest in faster servers, better algorithms, and more sophisticated data collection. By 2004, Google went public at a valuation of $23 billion. By 2010, it was worth more than $150 billion. The bargain had created one of the fastest accumulations of wealth in human history.

And somehow, at every step along the way, the arrangement felt fair.

The reason is subtle but decisive: Google delivered genuine value at a scale no one else could match. The company's search results were objectively better than its competitors'. Yes, Google was collecting your data. But it was collecting everyone's data, and using that aggregate information to make search work better for everyone. If you searched for "Paris," Google had to decide whether you meant the city in France or Paris Hilton, based on what millions of other people had clicked on. This required massive amounts of data—but it benefited everyone equally. The data collection and the value delivery were functionally identical.

This is the crucial insight that would structure the internet for two decades: In a system where data improves the product for everyone, collecting data feels like a public good. You weren't just trading your privacy for a service; you were contributing to a commons. Every search query made Google's algorithm slightly better for the next billion people. Every click helped the system understand what users actually wanted. The deal was not zero-sum. Google was enriching itself, yes—but it was also genuinely improving the internet for everyone.

Facebook copied the model and refined it. Mark Zuckerberg understood that Google had solved the monetization problem for search, but search was only the beginning. Why limit data collection to what people explicitly searched for? Why not track everything they did online? Facebook would create a platform where you shared your thoughts, your relationships, your photographs, your locations, your interests. You would do this voluntarily because the platform made it easy and fun. And Facebook would monetize that data at unprecedented scale.

When Facebook launched the News Feed in 2006, it introduced a system nearly as revolutionary as Google's search algorithm. Instead of seeing what you wanted to see (your friends' posts in chronological order), you would see what Facebook's algorithm determined you wanted to see. The company ranked content based on engagement metrics—likes, comments, shares—which it could track because it controlled the entire platform. Again, the data collection and the value delivery were aligned: the more data Facebook had about your behavior, the better it could personalize your feed to keep you engaged.

The value was real. Your News Feed was, generally speaking, more interesting than a chronological list of your friends' posts. Facebook's algorithm learned that you preferred certain friends' updates over others, certain types of content over others, certain times of day to see different types of posts. The algorithm made the platform feel personal, tailored to your tastes. And again, this personalization required data about you—but it benefited you directly in the form of a more engaging experience.

The crucial feature of this arrangement was its universality. Everyone used the same algorithm. Everyone saw a ranked feed. Everyone's data went into the same system. This meant that the system could achieve network effects and scale that made it nearly impossible to compete with. If you wanted to share your life online, you had to go where everyone else was. If you were a small business, you had to advertise where your customers were. Fragmentation would have destroyed the value of the platform—the reason Facebook had 100 million users in 2008 was precisely that everyone was on the same service, sharing with the same people, visible to the same advertisers.

Unified platforms could only work if they were genuinely unified.

This is why the free-to-user, ad-supported model became hegemonic across the internet in the 2000s and 2010s. It wasn't because it was inevitable or optimal. It became dominant because it created the conditions for network effects and scale. Twitter could become the public square of global conversation because everyone was on one service, seeing one feed, subject to one algorithm. YouTube could become the default video platform not because it had the best interface, but because it had the most videos and the most viewers, which attracted more uploaders, which attracted more viewers. Spotify could dominate music streaming because it had licensing deals with every major label and could offer universal access to nearly all recorded music.

The magic of these platforms was that they were better for everyone when they were unified. A musician was better off on a platform that had 300 million users than on a platform that had 30 million. A user was better off on a platform that had universal music licensing than one that had only 80% of the catalog. An advertiser was better off targeting attention on a service where everyone was aggregated than buying ads across a dozen fragmented services. The incentives all pointed in the same direction: toward consolidation, toward monopoly, toward the unified platform.

This created a kind of technological inevitability. Venture capital funded dozens of competitors to Facebook, Google, and Twitter. Almost all of them failed, not because they were poorly executed but because they couldn't achieve the network effects of the incumbent. Why would a friend invite you to a new social network if none of your other friends were there? Why would you switch to a search engine that had indexed fewer pages? Even if a competitor had a superior product in some dimension—better privacy, better moderation, a better algorithm—it was fighting against the raw economic power of network effects. The unified platform always won.

By 2010, the shape of the internet had solidified. Google dominated search and was expanding into maps, email, and cloud storage. Facebook dominated social networking and was expanding into photo sharing and messaging. Amazon dominated e-commerce. Apple had created a closed ecosystem that gave iPhone users a unified app store and services experience. YouTube had become synonymous with online video. These platforms were not evenly distributed. They were consolidated among a handful of American corporations. But they operated on a consistent model: collect data from users at scale, monetize it through targeted advertising or premium services, reinvest the profits in better products and more data collection.

And the bargain held because the platforms were genuinely useful.

By 2015, a person in rural India with a smartphone could search for medical information in their local language, find tutorials for building a business, watch educational videos, and message friends across the world—for free. An artist in Lagos could upload photos to Instagram and reach millions of potential customers. A teenager in Istanbul could find communities of people with shared interests through Reddit or Tumblr. A journalist in Brazil could distribute stories to thousands of readers using Facebook's platform.

Yes, the platforms were extracting data and attention. Yes, a handful of billionaires and tech companies were accumulating extraordinary wealth. Yes, there were genuine problems: Facebook's algorithm promoted divisive content because divisive content generated engagement. YouTube's recommendation system could radicalize people. Google's search results could spread misinformation. But the fundamental bargain remained appealing: you got access to tools and networks that had economic value, and you paid for them with data instead of money.

For people without money, this was genuinely transformative. A teenager in a poor country didn't have to pay for her Gmail account or her Instagram account or her TikTok account. She could access the world's information through Google, connect with people like her through social networks, develop her creativity through free tools. The unified platforms enabled a kind of democratic access to digital culture that would have been impossible if everything had required a subscription.

This is why the bargain persisted even as critics pointed out its inequities. Yes, Google and Facebook were becoming powerful. Yes, they were accumulating data in ways that should have worried us more than it did. Yes, there was something troubling about building a global digital commons and then extracting monopoly rents from every interaction within it. But the alternative seemed worse: a fragmented internet where some people had access to premium services and others didn't, where network effects were distributed across dozens of incompatible platforms, where you had to pay for something that was now free.

The unified model also solved what economists call the "cold start problem"—the challenge of launching a platform when no one is using it yet. Twitter in 2006 had almost no value because almost no one was there. But the platform could offer something else: access to a unified, public feed that was free and open to anyone. That free and open structure, combined with investment from venture capital, allowed Twitter to grow until it became valuable. The free-to-user model was not just a monetization strategy; it was a growth strategy. It allowed platforms to bootstrap themselves to scale.

By 2020, the unified internet model had reached its maturity. Google processed more than 5 billion searches per day. Facebook had nearly 3 billion users. YouTube had more than 2 billion logged-in users per month. These were not merely large services; they were global infrastructure. For billions of people, the experience of "the internet" meant using one or a handful of these platforms. You could live your entire digital life on Google and Facebook—search, email, messaging, social connection, video, maps, calendar, document storage. The platforms had become so comprehensive and so central that it seemed impossible to imagine the internet functioning differently.

And from a certain perspective, the arrangement had become more fair, not less. By 2015, Google and Facebook were processing so much data, operating at such massive scale, that the data collection began to feel impersonal. Google wasn't some human watching your searches. It was an algorithm, operating at the scale of billions. The data wasn't being sold to the highest bidder; it was being aggregated into patterns so large that individual privacy concerns seemed almost quaint. You weren't being individually manipulated; you were part of a statistical distribution. The platforms had transcended individual exploitation and become something closer to a transparent system—you knew they were collecting data, and you could see what you got in return.

This was also the crucial moment when the unified platform model achieved ideological hegemony. By 2016, it seemed inevitable that the internet would be organized around a handful of consolidated services. The free-to-user, ad-supported model seemed to be the only sustainable way to build internet infrastructure. Attempts to build alternatives—based on subscriptions, or federation, or user ownership—seemed doomed by comparison. The unified platforms had won, not because they were evil or because they had destroyed their competitors through anticompetitive practices, but because they were simply better at the job. They could deliver more value at lower cost because they could aggregate demand and supply at a scale that no one else could match.

Even the privacy concerns, which had been simmering for years, felt manageable in the context of the value delivery. Yes, Facebook was tracking you across the web. But Facebook also gave you a way to connect with 2 billion other people. Yes, Google was collecting your search history. But Google's search results were objectively better than any alternative. The tradeoff seemed reasonable.

What almost no one anticipated in 2020—when the unified internet seemed to have achieved permanent dominance—was that the economic foundation of the entire system was about to change. The algorithmic feed had made centralized platforms valuable. But that same feed, operated at scale, depended on a particular kind of computational infrastructure: data centers, indexed databases, real-time ranking algorithms. These were expensive, but they had become predictable and scalable. A company could build the best search algorithm by 2010 and operate it for a decade with incremental improvements.

Artificial intelligence would break this assumption. Large language models, generative AI, deep learning systems trained on massive datasets—these were not more of the same. They required different computational resources, different types of data, different training regimes. They were vastly more expensive to operate. They were harder to commoditize. And most crucially, they could not maintain the unified platform model. You could have one search algorithm that worked for everyone. You could not have one large language model that worked equally well for medical diagnosis, legal analysis, creative writing, and software development. The economics no longer supported universal platforms.

But in 2020, this was still invisible. The unified internet seemed permanent. The bargain—your data and attention in exchange for free access to a global network—seemed like it would continue indefinitely. Billions of people had organized their digital lives around this arrangement. Entire businesses had been built on top of these platforms. The regulatory attention was just beginning, but even the regulators seemed to accept the basic structure: these platforms were too important to break up, they had become critical infrastructure, and the best we could do was impose rules on how they collected data and used it.

What was happening in research labs and startups, what the venture capital firms were quietly funding, what the major AI companies were beginning to build—this would systematically dismantle the conditions that made the unified internet possible. Not through malice or incompetence, but through basic economics. The technology was changing, the costs were shifting, and the unified platform model was becoming uneconomical.

The bargain was still in place. Billions of people were still trading data for access. But the infrastructure that made the bargain sustainable was decaying. And in the space where it decayed, new structures were already beginning to grow: walled gardens, proprietary APIs, subscription services, specialized systems that only worked for certain uses or for certain users. The fragmentation had not yet become visible. But the foundation for it was already laid.