Rethinking the competitive advantages of AI-exposed companies

For decades, investors have relied on a familiar set of labels to evaluate competitive advantages: network effects, switching costs, data moats, brand, talent. These frameworks still describe real economic forces, but they were built for a world where human knowledge work was expensive and organizational effort took time. With the advent of AI, that assumption is now breaking down.

AI is not going to eliminate every competitive advantage. It is, however, forcing investors to look more carefully at what makes a moat durable, and in many cases the answer has changed. Every moat rests on some form of scarcity, which is a resource or capability that is constrained and hard to replicate. That scarcity does not disappear as technology advances; it relocates. Where it lands helps determine which companies can sustain their edge and which will have to adapt.

A pattern that repeats

This is not the first time technology has shifted where scarcity lives. In the late 1990s, finding information was the bottleneck. Google indexed the web and made search effectively free. Scarcity moved to attention and distribution, which is why the next generation of dominant businesses were ad networks and platforms.

In the 2010s, owning and managing servers was the constraint. Cloud computing made infrastructure elastic and metered. Scarcity migrated to power, cooling, chips, and physical capacity – a shift that still underpins much of today’s investment case for utilities and infrastructure.

Now we are in a third cycle. AI is absorbing the bottleneck of human cognitive labor. Routine analytical work that once took a person hours can now be completed by an AI agent in minutes. The former scarcity is moving toward three areas: judgment that requires human discretion, trust in the form of verification and accepted standards, and the physical inputs required to scale AI systems.

Cheaper to do, not cheaper to check

An academic paper¹ published earlier this year by researchers Christian Catalini, Xiang Hui, and Jane Wu offers a useful economic framework for understanding this shift. Their core argument is that the cost to automate a task is falling rapidly, but the cost to verify the result is not falling nearly as fast.

Tasks that can be clearly defined, checked by a machine, and repeated cheaply is approaching zero marginal cost. That covers much of what companies have historically paid junior analysts, entry-level engineers, and consultants to do. Verification is different: Someone must still confirm the output is correct, accept liability for it, and resolve disputes when the stakes are real. Those costs stay sticky, and in some cases rise, since more AI-generated content increases demand for content that can be trusted.

That distinction between generating an output and verifying it has meaningful investment implications. If a company’s core advantage rested mainly on work AI can now do cheaply, that advantage may be eroding. If it rested on trust, accepted standards, or physical scarcity, it may be getting stronger.

Three outcomes: Collapsed, moved, strengthened

Building on these ideas, one way to think about how AI is impacting competitive advantages is to sort companies into three categories based on what happened to their core scarcity.

Scarcity collapsed. Chegg, the education technology company, built what appeared to be a classic data flywheel: More students generated more questions, which produced more answers, which attracted more students. But the real scarcity was not the database; it was the human work of taking a textbook problem, explaining it clearly, and making it searchable. Once AI made that kind of reasoning abundant, the accumulated content was no longer the advantage. In some respects, it may have even become training material for the models competing with it.

This illustrates a risk worth watching: If a company’s data is essentially stored human cognitive work and a model can reproduce the same utility at lower cost, the advantage can unravel quickly.

Scarcity moved. eBay’s moat was traditionally described as liquidity-driven network effects: More buyers attract more sellers, and vice versa. That is still partly true, but it bundles together very different layers of value. For commodity goods, AI agents could potentially search across multiple marketplaces, compare prices and reviews, and route a buyer to the best deal regardless of where they started. That layer faces pressure over time.

But eBay also has categories built on items that are physical, unique, and irreplaceable, such as collectibles, vintage goods, and rare parts. These items either exist or they do not – they cannot be generated by a model. And in a world of increasing synthetic content, the ability to verify a physical item’s authenticity could become more valuable. The moat hasn’t disappeared; Rather, it is shifting from generic discovery toward unique supply and authentication.

Scarcity strengthened. Axon, the public safety technology company, may be the clearest example in this category. AI can make transcription, tagging, report writing and evidence review cheaper. But none of that undermines Axon’s core asset: an authenticated chain of custody that tells courts, prosecutors, and municipalities that a piece of evidence is real, preserved, auditable, and admissible.

In a world filling up with synthetic video, synthetic audio, and AI-generated claims, that kind of trust becomes more valuable, not less. Whereas Chegg sold explanations into a world where explanation became abundant, Axon sells authenticated evidence into a world where authenticity is becoming scarcer.

Look under the hood

The practical implication is clear: Now is the time to check in on the state of companies’ competitive advantages.

AI does not need to impact next year’s earnings to affect today’s valuation; it only needs to shorten how long the market expects a company’s advantage to last. That repricing is already underway across application software, information services, marketplaces, and platform businesses. But it is often too blunt, treating every moat in a category as equally vulnerable when the underlying scarcity differs significantly from one company to the next.

The old moat vocabulary is not dead. Network effects are real. Switching costs are real. Data advantages are real. But they are a starting point, not a definitive answer. The work now is to look underneath them; identify whether the scarcity that supported the advantage has collapsed, moved, or strengthened; and judge durability accordingly. That is where our research is focused.

¹ Catalini, Christian and Hui, Xiang and Wu, Jane, Some Simple Economics of AGI (February 24, 2026). MIT Sloan Research Paper, available at SSRN: https://ssrn.com/abstract=6298838 or http://dx.doi.org/10.2139/ssrn.6298838