For the past several months, I have been following the federal government’s antitrust cases against Google and Apple with growing interest. At their core, these lawsuits argue that some of the most successful companies of the Information Age used their control over search, smartphone operating systems, and app distribution to disadvantage competitors and strengthen their own market positions.
The cases have renewed a broader conversation in Washington. A coalition of startup companies, technology incubators, and software firms recently launched the Little Tech Association, arguing that the federal government has spent too many years listening to the largest technology companies while paying too little attention to the entrepreneurs building on their platforms. Among the organization’s first priorities is renewed support for the American Innovation and Choice Online Act (AICOA), legislation intended to limit the ability of dominant online platforms to favor their own products over those of competitors.
But as I have followed these cases and the efforts of small tech to check the power of big tech, I keep coming back to a different question. Are we spending so much time litigating the monopolistic practices of the Information Age that we are overlooking how market power may emerge in the Data Economy?
Artificial intelligence is creating a new generation of technology infrastructure, and with it, a new set of incentives. If history offers any lesson, it is that every infrastructure layer eventually accumulates economic power. The challenge is recognizing how that power develops before it becomes entrenched. Congress continues to wrestle with a question that has defined much of the past decade: What responsibilities accompany extraordinary market power in the digital economy?
The debate is an important one. But I worry that debate is too focused on the closing chapter of the Information Age while a different form of market concentration is quietly emerging alongside artificial intelligence. History rarely repeats itself exactly. It does, however, have a habit of following similar patterns and that’s what I’d like to dig into today.
Every major technological era develops its own source of economic leverage. The railroad industry concentrated power around transportation. Electrification rewarded those who built and operated utility networks. The Information Age shifted that leverage toward digital distribution.
Companies such as Microsoft, Google, Apple, Amazon, and Meta built remarkable businesses by becoming indispensable gateways. Microsoft’s operating system became the standard platform for personal computing. Google organized access to information. Apple controlled access to hundreds of millions of smartphone users through the App Store. Each created enormous value for consumers while simultaneously occupying increasingly influential positions between innovators and customers.
As distribution platforms mature, familiar tensions emerge. Apple’s App Store policies became the subject of antitrust investigations and high-profile lawsuits over commissions, access, and competitive practices. Google has faced years of scrutiny over allegations that it favored its own services within search results. Developers even coined a term that I believe will become exceedingly relevant in the AI era: Sherlocking.
The expression dates back to Apple’s introduction of Sherlock 3 in the late 1990s, when features offered by a popular third-party application appeared within Apple’s own software. Since then, developers have used the term whenever a platform owner introduces native functionality that competes directly with successful independent applications. Apple has consistently argued that integrating broadly useful features improves the experience for users, while critics contend that doing so can eliminate entire categories of third-party software. The significance of Sherlocking was never the individual feature. It was the realization that success on someone else’s platform could eventually invite competition from the platform itself.
The broader economic lesson is less controversial. Companies that control distribution inevitably possess opportunities that others do not. The ability to influence how products reach customers creates incentives to expand into adjacent markets, favor internal offerings, or capture a greater share of the value created by others using the platform.
That pattern is hardly unique to technology. It has appeared repeatedly throughout economic history whenever a small number of organizations controlled essential infrastructure. And not all infrastructure companies have followed the Sherlocking path.
When Amazon Web Services emerged as the dominant cloud computing platform, thousands of startups entrusted their businesses to it. Entrepreneurs built companies on AWS because they believed Amazon’s objective was to provide infrastructure rather than compete in every market its customers entered. That distinction mattered.
AWS supplied computing power, storage, databases, and networking. Startups remained responsible for discovering markets, developing products, and serving customers. The relationship created one of the most productive periods of entrepreneurial activity in modern technology because each participant focused on a different layer of the value chain.
Neutral infrastructure did more than improve efficiency. It established confidence. Founders could invest years building products because they trusted that the platform’s success depended on enabling innovation rather than appropriating it.
No infrastructure provider is perfectly neutral, and Amazon itself has faced competitive criticism in other parts of its business. Even so, the cloud computing ecosystem demonstrated how powerful that separation between infrastructure and applications could become.
Artificial Intelligence occupies a different position. Foundation AI models are clearly infrastructure. But they introduce a relationship that differs in important ways from previous generations of computing infrastructure.
Cloud platforms rented processing power. AI systems increasingly participate in the work itself. Developers ask them to design software architectures, write code, analyze customer requirements, summarize research, refine pricing strategies, generate marketing campaigns, and explore new product ideas. Increasingly, the infrastructure is not merely executing instructions. It is participating in the creation of the application.
That distinction is important. And to be clear, I’m not arguing that foundation model providers are misusing customer information today. Major providers have published commitments regarding enterprise privacy, API usage, and model training. There is little evidence that they are systematically exploiting customer interactions to identify products to replicate. But unquestionably a small number of AI companies are providing intelligence infrastructure upon which thousands, and perhaps millions, of future businesses are likely to depend.
History suggests that incentives usually outlast intentions. Infrastructure companies rarely begin with ambitions to own every layer of the value chain. Over time, however, the economic incentives to expand become increasingly difficult to ignore. For the first time in the history of computing, the companies providing the foundational infrastructure also occupy a position from which they can observe, at extraordinary scale, how developers describe problems, explore markets, and build solutions. Whether that visibility ultimately changes competitive dynamics remains uncertain. The structure itself, however, is unlike anything created during the Information Age.
It raises questions extending beyond traditional antitrust doctrine. Will entrepreneurs view foundation model providers the way previous generations viewed AWS; as trusted infrastructure that succeeds by enabling others? Or will competitive pressures encourage foundation model providers to move steadily upward through the value chain, from infrastructure to development tools, distribution, and eventually applications themselves?
There is very little that prevents the largest AI infrastructure companies from offering analogous services to any application built with AI from the beginning. They have the knowledge of how the tech is built, along with the business logic, market situation and other know-how that used to be held as trade secrets inside enterprises. Entire enterprises in specific markets could become new features for a new oligarchy of foundation model companies.
Markets do not answer such questions overnight. They evolve through incentives, investment decisions, and competitive behavior accumulated over many years. Those decisions matter because entrepreneurs and investors respond to incentives just as platform companies do. If founders conclude that building directly on foundation models exposes them to fundamentally different competitive risks than previous technology platforms, investment strategies will evolve accordingly. The next generation of startups may be built differently, financed differently, or pursue entirely different forms of defensible advantage.
Each technological revolution creates a new layer of infrastructure. Over time, that infrastructure accumulates economic influence because everyone else depends upon it. The challenge for policymakers has never been preventing infrastructure from becoming successful. It has been preserving the conditions that allow the next generation of innovators to build upon infrastructure with confidence.
As artificial intelligence becomes the infrastructure of the Data Economy, that question deserves attention. Whether AICOA ultimately succeeds or fails is almost secondary. The legislation asks how to regulate the infrastructure of the Information Age. The harder question, and one that will define the next decade, is how to preserve competition when the infrastructure itself increasingly participates in creating the applications built upon it.