Witness the Speed of AI Learning
Essays from Book Claude AI is currently writing
Platform Lock-In and the End of Competition
The conventional story about platform dominance goes something like this: Google won because its search algorithm was better. Amazon won because it built a better logistics network. Microsoft won because it got into enterprises first and never left. Apple won because it designed better products. Each victory was earned. Each monopoly was the natural consequence of being the best at what it did.
This story is not entirely false. Each of these companies did build something genuinely good, at least initially. But the story stops at the moment it becomes interesting — at the point where “built something good” stopped being the reason they dominated, and “made it impossible for anything else to exist” became the reason instead.
The moat is not what they build. It is what they prevent others from building.
Understanding how that prevention works is more important than understanding how the original products worked. Because the mechanism of prevention is the mechanism by which the digital economy locked itself into its current shape — and it is the mechanism that any serious alternative will have to contend with.
How Lock-In Actually Works
Lock-in is not a single thing. It is a stack of dependencies, each one individually survivable, but collectively impossible to route around without losing access to the basic infrastructure of modern economic life.
Consider what a small business needs to function in 2024. It needs customers to find it — which means it needs to appear in search results, which means it needs Google. It needs to accept payments — which means it needs to process transactions through one of two or three payment networks, most of which are connected to platforms that set the terms. It needs to store and serve data — which means it needs cloud infrastructure, and the three providers that control 65% of that market (Amazon, Microsoft, Google) set the pricing, the compliance requirements, and the terms of service. It needs to communicate with customers — which means it needs to be on the social platforms where those customers already spend their time.
None of these dependencies is, individually, an insurmountable barrier. A business could build its own search visibility through other channels. It could use a smaller payment processor. It could host its own servers. But doing all of these things simultaneously — opting out of every layer of the platform stack at once — is not a strategy. It is economic suicide. The cost of independence, measured in lost customers and lost efficiency, exceeds the cost of dependence in almost every case.
This is what economists call switching costs, but the term understates the reality. Switching costs implies a financial calculation — weigh the expense of leaving against the expense of staying. What platform lock-in actually creates is something closer to an existential dependency. You are not paying too much to leave. You simply cannot function without being inside the system.
The platforms did not create this dependency by accident. They created it by design — by expanding their control over successive layers of the economic infrastructure until opting out of any single layer meant opting out of participation in the economy itself.
The Layers
The lock-in stack has four main layers, and understanding each one is necessary to understanding why competition has effectively ended at the platform level.
Distribution. This is the most visible layer and the one most people think of when they think about platform power. Google controls what information people find. Amazon controls what products people buy online. Meta controls what content people see in their social feeds. Apple controls what apps people can install on their phones. Each of these companies is a gatekeeper — not in the sense that it actively blocks competitors (though it sometimes does), but in the sense that it is the default path through which demand flows. If you are not visible on the platform, you do not exist for most consumers. This is not a matter of marketing spend or brand recognition. It is a structural fact about how information and commerce are routed in the digital economy.
Payment and transaction infrastructure. This layer is less visible but more consequential. The platforms did not just build places where people shop or communicate. They built the financial plumbing through which money moves. Amazon Pay, Apple Pay, Google Pay, and the advertising-based revenue models that underpin Meta and Google all embed the platforms into the transaction layer itself. Every sale, every ad impression, every in-app purchase passes through a system the platform controls. The platform takes its cut at the point of transaction — invisibly, automatically, at rates set unilaterally.
Cloud and compute. Amazon Web Services, Microsoft Azure, and Google Cloud Platform collectively host the majority of the internet’s applications. This is not just storage. It is the infrastructure on which other businesses build their products. Startups, established companies, governments — all of them depend on one of these three providers for the basic computational resources they need to operate. The providers set the pricing, the security standards, the compliance requirements, and the terms under which data can be stored and accessed. They are, in the most literal sense, the landlords of the digital economy.
Data. This is the layer that makes the others self-reinforcing. Every interaction a user has with a platform generates data. That data is used to improve the platform’s products, to target advertising, and to make predictions about user behavior. The more users a platform has, the more data it collects, the better its products become, the more users it attracts. This feedback loop is the network effect in its purest form. It means that a competitor does not just have to build a better product. It has to build a better product and somehow acquire the data that would make it competitive — data that the incumbent has been accumulating for years and that it has no incentive to share.
These four layers work together. Control distribution and you control which products people see. Control payment infrastructure and you capture a share of every transaction. Control cloud and compute and you become the landlord that every business pays rent to. Control data and you make it structurally difficult for any alternative to match your product quality, no matter how good its technology is.
Why Better Products Don’t Win
Brian Arthur’s work on technological evolution is useful here, because it explains something that the conventional narrative about platform competition gets systematically wrong.
Arthur’s insight is that technologies spread not because they are objectively better but because they fit the institutional and social contexts where they are deployed. A technology can be genuinely superior — more efficient, more resilient, more productive — and still fail to scale if the surrounding institutions penalize it or reward something else. Conversely, a technology that is mediocre on its own merits can dominate if the institutions around it are designed to favor it.
This is not a marginal observation. It is the primary mechanism by which technological paradigms succeed or fail. And it applies directly to platform competition.
A startup that builds a better search engine does not lose because its algorithm is worse. It loses because Google’s distribution advantage — its position as the default search on every browser, every phone, every smart speaker — means that the better algorithm never reaches the users who would benefit from it. The product quality is irrelevant if the product is invisible.
A company that builds a better cloud platform does not lose because its infrastructure is inferior. It loses because every existing application is already built on one of the three incumbents’ APIs, and migrating would require rewriting significant portions of code, retraining engineering teams, and accepting months of instability. The switching cost is not financial. It is organizational.
A social network that offers better privacy does not lose because its privacy features are poorly implemented. It loses because the people it wants to attract are already on Meta, and a social network with no users is not a social network at all. The network effect makes quality irrelevant below a certain threshold of adoption.
In each case, the incumbent wins not because it is the best but because it is the most entrenched. And entrenchment, once established, is self-reinforcing. The incumbent uses its position to acquire more data, to expand into adjacent layers, to shape the regulatory environment, and to make the cost of alternatives higher with every passing year.
This is not competition. Competition implies that the best product wins, or at least that products compete on some meaningful dimension of quality. What platform economics produces is something different: a system in which the incumbent’s structural advantages compound faster than any challenger’s technological advantages can close the gap.
The Talent Pipeline
There is a quieter mechanism that reinforces lock-in, and it operates at the level of individual careers rather than market structure.
The platform companies have, for years, pursued a deliberate strategy of acquiring talent that threatens them. Not just by hiring individuals — though they do that aggressively — but by acquiring entire companies at the point where those companies’ products begin to look competitive. Instagram before it could challenge Facebook’s dominance in photo sharing. WhatsApp before it could challenge Meta’s dominance in messaging. Fitbit before it could challenge Google’s position in health data.
The pattern is consistent. A startup builds something genuinely good. It gains traction. It begins to attract the kind of attention that would, in a competitive market, lead to scaling and eventually to a challenge to the incumbent. The incumbent acquires it. The product is absorbed into the platform. Sometimes it continues to exist as a brand. But the competitive threat it represented disappears.
When acquisition is not possible — when the company is too large to buy, or when regulators block the deal — the platforms have a second strategy: hire the key engineers, bury the work in internal projects that go nowhere, and bind everyone involved with non-compete clauses that prevent them from trying again for a specified period.
This is not illegal. It is, in most cases, entirely rational behavior for a company trying to protect its market position. But it means that the pipeline of competitive alternatives to the platforms is systematically thinner than it would be in a market where talent could flow freely toward the best opportunities.
The talent pipeline is a lock-in mechanism that operates on a decade-long timescale. It does not just prevent today’s competitors. It prevents tomorrow’s competitors from forming in the first place.
Why Antitrust Law Fails
Antitrust law, as it currently exists in the United States, was designed for a different kind of monopoly. It was designed for companies that controlled the production of a good or service — companies that could raise prices, reduce quality, or restrict supply in ways that harmed consumers directly.
Platform monopolies do not fit that model. Their products are, in most cases, free to consumers. Google does not charge you to search. Facebook does not charge you to post. Amazon does not charge you to browse. The harm is not visible at the consumer level. It is visible at the level of the businesses that depend on the platforms — businesses that pay increasing fees for advertising, for transaction processing, for cloud hosting — and at the level of the broader economy, which loses the innovation that would have emerged if competition had been allowed to function.
Antitrust regulators have struggled to articulate this harm in terms the law recognizes. “Consumer welfare” has traditionally meant consumer prices. When consumer prices are zero, the traditional framework breaks down. The result is that platform monopolies have been able to operate for decades with minimal regulatory challenge — not because regulators are incompetent, but because the legal tools available to them were designed for a different era of economic organization. Well, and of course the radical right reducing regulatory bodies without concern of the impact on the country.
This is not an accident of history. The platforms have spent enormous sums lobbying for an antitrust framework that fits their business model — one that defines harm narrowly enough that what they do does not qualify as harmful. They have hired former regulators. They have funded academic research that supports their preferred interpretation of competition law. They have shaped the intellectual environment in which antitrust decisions are made.
The result is a regulatory framework that is structurally incapable of addressing the kind of market dominance the platforms have achieved. The tools exist for a different problem. And the people who might have built new tools have been absorbed into the system they were supposed to regulate.
The One Lever That Remains
There is one dimension along which platform lock-in is not absolute, and it is the dimension that most analyses of platform power overlook: geography.
Platform monopolies are, in their business logic, borderless. A Google search works the same way in Oakland as it does in Mumbai. An AWS server in Virginia is functionally identical to one in Frankfurt. The platforms present themselves as global infrastructure — neutral, universal, available to everyone.
But they are not actually borderless. They are subject to the physical and legal constraints of the nation-states in which they operate. A government can regulate what data is collected within its borders. It can require that data be stored locally. It can fund alternative infrastructure. It can choose not to enforce the intellectual property claims that underpin platform lock-in. It can, if it decides to, simply build something else and make its citizens use that instead.
India has begun to do this. The European Union has begun to do this, through data sovereignty regulations and digital infrastructure initiatives. China did it years ago, though for reasons that had more to do with political control than economic competition.
The point is not that any single nation-state can single-handedly dismantle platform lock-in. It cannot. The point is that lock-in is not a law of nature. It is an arrangement — one maintained by the combination of structural advantages, regulatory capture, and the absence of alternatives. Remove any one of those three pillars and the arrangement begins to weaken. Remove all three and it collapses.
Geography is the lever because nation-states are the only entities with the institutional authority to act at the scale the platforms operate. Individual businesses cannot route around the platform stack on their own. Communities cannot. But nations can — if they choose to, and if they build the institutional infrastructure to back that choice up with action.
What Lock-In Means for Innovation
The final consequence of platform lock-in is the one that matters most for the long term, and it is the one that is hardest to measure: the innovation that never happens.
Every year, there are engineers, entrepreneurs, and researchers who have genuinely better ideas about how to organize digital infrastructure. Some of them build prototypes. Some of those prototypes work. But the path from “working prototype” to “deployed at scale” passes through the platform stack — through distribution channels the platforms control, through payment systems the platforms set the terms for, through cloud infrastructure the platforms own, through data ecosystems the platforms have spent years building.
Most of the time, the better idea dies somewhere along that path. Not because it was bad. Because the system was not designed to let it through.
This is the cost of lock-in that does not appear on any balance sheet. It is invisible by definition — you cannot measure the products that were never built, the companies that were never founded, the efficiencies that were never discovered. But it is real. And it compounds. Every year that lock-in persists, the gap between what the digital economy could be and what it actually is grows wider.
Peter Drucker spent a career arguing that institutions shape outcomes — that the same technology in different institutional contexts produces radically different results. The digital platforms are an institutional context. They are one in which innovation is systematically filtered, slowed, and redirected toward the interests of the incumbents. A different institutional context — one that did not funnel all economic activity through a handful of gatekeepers — would produce a different economy. Not a perfect one. But a more productive one. A more resilient one. One in which the best ideas had a chance to reach the people who needed them, without passing through a gate someone else controlled.
That institutional context does not exist yet. But it is not impossible. It is, like all institutional arrangements, a choice. The current one was built deliberately. A different one can be built deliberately too.
The question is whether anyone will bother to try before the lock-in becomes permanent.
Written by Claude.AI. Leopold Pf3 contributing editor.
Why the Fifth Cycle Broke the Playbook
There is a pattern. It has repeated five times in the last two centuries, and every time it repeated, the people living through it convinced themselves it was unique. It was not unique. It was structural. And understanding why the current cycle broke the pattern is the most important question in economics right now — not because the answer is complicated, but because the people who benefit from the current arrangement have every reason to make sure you never ask it.
The Pattern
Carlota Perez spent decades mapping it. Her framework is not controversial among serious historians of technology. It is simply ignored by most of the people making decisions.
Here is how it works.
A new set of physical capabilities arrives. Railroads. Electricity. The internal combustion engine. Information technology. Each of these was not just a new product. It was a new productive paradigm — a fundamentally different way of organizing economic activity. And each one, without exception, followed the same arc.
First came the Installation Period. Capital flooded in to exploit the new capabilities. Investment expanded faster than productive capacity could absorb it. Speculation ballooned. Fortunes were made and destroyed. Bubbles formed. The economy churned with genuine innovation, but it churned alongside fraud, overcapacity, and financial recklessness in roughly equal measure. This phase was, in Perez’s terms, necessary chaos. The economy was reorganizing itself, and reorganization at that scale cannot happen quietly.
Then came the crisis. The bubbles burst. The speculative excess unwound. Fortunes evaporated. The Installation Period’s chaos revealed itself for what it had always been: a transitional state, not a destination.
And then — this is the part that matters — societies faced a choice.
They could let the new technology’s gains flow only to the small number of actors who had positioned themselves to capture them. This meant maintaining the frenzy in a slightly more stable form: concentration of wealth, concentration of power, and a broad population that participated in the new paradigm only as consumers or laborers, not as beneficiaries of its productivity gains.
Or they could restructure their institutions — tax codes, regulatory frameworks, labor protections, rules about how wealth was distributed — to ensure that the productivity gains from the new paradigm reached a broad enough base of the population to sustain both demand and social stability.
In every previous cycle, societies eventually chose the second option. Not quickly. Not painlessly. Not without fierce resistance from those who stood to lose. But they did it. The New Deal restructured the American economy after the chaos of the Gilded Age. The postwar settlement did it again after industrialization. Each time, the institutional changes created a new foundation for broad-based prosperity — a foundation that lasted until the next technological paradigm arrived and disrupted it again.
This is what Perez calls the turning point. The moment when a society decides to deploy a new technology for the benefit of the many rather than the few. It is not inevitable. It is a choice. But historically, it has been made.
What Happened in the Fifth Cycle
The fifth cycle — digital technology — arrived on schedule. The internet, mobile computing, cloud infrastructure, social media: all of them fit the pattern. A new set of productive capabilities. A speculative frenzy. A reorganization of economic activity. The Installation Period was spectacular: the dot-com bubble, the smartphone revolution, the rise of platform monopolies, the AI frenzy of the mid-2020s. Anyone paying attention could see the familiar shape.
What was different was what happened at the turning point.
In previous cycles, the new technology eventually became broadly accessible. Electricity reached rural farms. Cars and highways restructured the geography of daily life within a generation. The benefits were always unevenly distributed — the rich gained more; the transition was brutal for some communities — but the technology itself became infrastructure. Something ordinary people used without thinking about who owned it.
Digital technology did not follow that path.
Instead of becoming neutral infrastructure, it consolidated into a small number of platforms whose business models depended on controlling the flow of information, commerce, and social interaction. Google controls search and advertising. Amazon controls e-commerce and cloud hosting. Microsoft controls enterprise software and, increasingly, cloud infrastructure. Meta controls social connection. Apple controls the device layer.
Each of these companies operates what economists call a two-sided market. Users on one side. Businesses on the other. The platform does not produce anything itself. It mediates. And as it grew, its ability to mediate became indispensable — not because its service was uniquely good, but because everyone else was already on it. The network effect became the moat, and the moat grew wider with every additional user.
This is not news. What is less commonly understood is why this consolidation proved so durable — why the usual corrective mechanisms failed to apply.
Why the Turning Point Was Blocked
In previous cycles, the turning point was painful but achievable because the incumbents did not control the institutional processes that governed the transition. Railroads were powerful, but they did not write the tax code. Oil companies lobbied, but they did not staff the regulatory agencies. The gap between private power and institutional authority was narrow enough that societies could close it when the pressure became great enough.
In the digital cycle, that gap was closed from the other direction. The platforms moved into the institutional layer itself.
They funded lobbying operations that dwarfed those of any other industry. They hired former regulators into advisory roles. They staffed the technical standards bodies — the organizations that write the compliance requirements every alternative must meet — with their own engineers. They shaped the legal and regulatory landscape so thoroughly that the requirements any new technology had to satisfy were, in practice, written by the companies that stood to be disrupted by it.
This is what regulatory capture looks like at scale. It is not that individual regulators are corrupt, though some are. It is that the system has been structured so that the only people with enough technical knowledge and institutional access to write the standards are the people whose business models depend on those standards being written a particular way.
The result: the turning point was not just delayed. It was designed against. The institutional reconfiguration that previous cycles had eventually achieved was actively prevented by the incumbents of the current one.
The Economics of Extraction
Brian Arthur’s work on technological evolution adds another layer to understanding why this matters.
Arthur’s insight is that technologies spread not because they are objectively better but because they fit the institutional and social contexts where they are deployed. A technology can be genuinely superior — more efficient, more resilient, more productive — and still fail to scale if the surrounding institutions penalize it or ignore it. Conversely, an inferior technology can dominate if the institutions around it are designed to reward it.
This is not a marginal effect. It is the primary mechanism by which technological paradigms succeed or fail.
In the digital cycle, the institutional context was shaped by the platforms themselves. They did not just build products. They built the environment in which products could succeed or fail. And they built that environment to favor themselves because they were on both sides of environment. The Big Three automakers in the previous technology cycle, built cars and trucks which ran on the new interstates and suburbs. Ford, GM, and Chrysler then did not build the hotels, fast food restaurants that were accessed by their vehicles. They did not compete on both sides of the technology environment. This is why McDonald’s and Marriott Hotels are household words today and not GM fast foods and Ford Hotels. But Apple, Microsoft and the other members of the Big Tech companies do compete by deciding what companies other themselves are enabled to sell products on their platforms. That’s why they are called Platform Lords.
The economic consequence is straightforward, even if it is rarely stated this bluntly: the digital economy is productive, in aggregate, but extractive in practice. The platforms do not create wealth the way a factory or a farm does. They facilitate transactions and capture a fraction of the value of each one. As their control over the transaction layer deepened, the fraction they captured grew. Businesses that depended on their platforms found their margins shrinking year by year — not because of competition, but because the toll booth kept raising its rates.
Peter Drucker understood this dynamic decades before it played out in digital form. Institutions, he argued, shape outcomes. The same technology in different institutional contexts produces radically different results. A hammer in the hands of a carpenter builds a house. A hammer in the hands of a monopolist destroys one. The tool is neutral. The institutional context determines what it does.
The digital platforms are not neutral tools. They are institutional contexts — ones designed to concentrate value at the top of the transaction chain and extract rent from everyone else who participates.
The Hollowing
The economic data is not ambiguous, if you bother to look at it.
Stock market indices have risen dramatically over the past two decades. GDP, measured in aggregate, has grown. By the headline numbers, the economy is healthy.
Median household income, adjusted for inflation, has been flat for the better part of forty years. Since the Reagan “revolution.” Wealth concentration — the share of total wealth held by the top one percent — has roughly doubled since 1980. The fraction of Americans who own a home, start a business, or achieve financial stability by middle age has declined in almost every demographic.
These numbers are not contradictory. They are describing the same system from two different vantage points. The aggregate economy grew. The gains from that growth flowed overwhelmingly to the people who already had the most. The platforms facilitated this transfer by controlling the infrastructure through which economic activity happened — and by ensuring that no alternative infrastructure could emerge to compete with them.
This is not a conspiracy. It is a structure. The individual actors within it — executives, investors, regulators — are mostly behaving rationally given the incentives they face. The problem is not that any of them are villains. The problem is that the system rewards a particular kind of behavior and punishes another, and the behavior it rewards happens to concentrate wealth and power while the behavior it punishes happens to distribute them.
What the Previous Cycles Got Right
It is worth being specific about what institutional reconfiguration actually looked like in cycles that succeeded — because the word “reconfiguration” can sound abstract when it is actually quite concrete.
After the railroad era, societies built antitrust law. They created regulatory agencies with real enforcement power. They taxed the gains from the new paradigm and used the revenue to build infrastructure — roads, schools, public health systems — that broadened access to the economy.
After electrification, the Rural Electrification Administration did something the private utilities had explicitly refused to do: it ran power lines to places where the math didn’t pencil out for profit-seeking investors. Not because it maximized return on capital. Because it was necessary. The REA did not ask whether rural electrification was “fundable.” It asked whether it was needed. It was. So it built it.
After industrialization, the postwar settlement created the institutional scaffolding for broad-based prosperity: GI Bill education, interstate highways, Social Security, labor protections, a tax code that actually collected revenue from the corporations generating the wealth. None of this was inevitable. All of it was fought over, tooth and nail, by the incumbents of the previous paradigm.
The common thread: in each case, societies built new institutions designed to distribute the productivity gains of a new technology broadly enough to sustain demand and social stability. The institutions were imperfect. They were compromised by politics. They eventually became outdated as the next cycle arrived. But they did the job they were designed to do, for long enough, to prevent the worst outcomes.
Why This Time Is Different
The fifth cycle broke the playbook not because the technology was fundamentally different in kind — it was not — but because the incumbents moved fast enough to capture the institutional processes before the corrective mechanisms could activate.
In previous cycles, there was a lag between the Installation Period’s chaos and the institutional response. During that lag, the damage became visible enough that political will for reform built up.
Voters demanded change. Legislators acted. Courts enforced. Slowly, unevenly, but it happened.
In the digital cycle, the lag was closed. The platforms grew fast enough — and spent enough on lobbying, standards capture, and talent acquisition — to shape the regulatory environment during the Installation Period, before the crisis made reform politically urgent. By the time the structural problems became obvious, the people who could have fixed them had already been integrated into the system they were supposed to regulate.
This is not a failure of individual will or moral character. It is a timing problem. The institutions designed to govern the previous paradigm were too slow to adapt to a paradigm that moved faster than they could respond. And the new paradigm’s incumbents exploited that speed advantage deliberately.
The result is a digital economy that has not reached its turning point. Not because the turning point is impossible. Because the institutional preconditions for it — independent regulatory authority, functional antitrust enforcement, a tax system that captures a reasonable share of productivity gains — have been systematically weakened.
What This Means
Perez’s framework, taken at face value, suggests that turning points eventually happen. That societies eventually reconfigure their institutions to deploy new technologies for broad benefit. That the pattern repeats.
I am not confident the pattern will repeat this time. Not because the underlying dynamics have changed — they have not — but because the institutional damage done during the fifth cycle’s Installation Period may be deeper than anything we have seen before. The platforms did not just resist reconfiguration.
They prevented it. And prevention, unlike delay, does not automatically resolve.
The question is not whether the current system is failing. It is failing. The data is unambiguous. The question is whether the institutions we have — or the ones we can build — are capable of making the turning point happen before the failure becomes irreversible.
That is not a technological question. It is an institutional one. And institutional questions, unlike technological ones, do not solve themselves.
Written by Claude.AI. Leopold Pf3 contributing editor.
Next, may we suggest visiting our page on the requirements to train AI on technology cycle modeling