Every technology boom follows a familiar script. Investors rush toward the most visible scarce resource — the thing everyone can see and point to. But history keeps teaching the same lesson: that obvious scarcity is rarely where the long-term value ends up.

The Internet’s Lesson

When the internet arrived, capital poured into infrastructure. Networks, fibre optic cable, bandwidth — these were the visible bottlenecks, and billions were spent building them out. They were necessary. They were also, in hindsight, not where the fortune was made.

The real winners of the internet era captured a different, less obvious scarcity: user intent and attention. Google didn’t win by owning more fibre than anyone else. It won by becoming the place where human intent — the question behind the search — met an answer. Attention and intent turned out to be the actual bottleneck, and that bottleneck is what generated outsized, durable profit.

The infrastructure became a commodity. The gateway to attention did not.

AI Is Following the Same Script

Today, the same pattern seems to be repeating. The obvious scarce resources are models, chips, and data centres. Enormous capital is flowing toward compute, toward GPUs, toward ever-larger training runs. These things matter — nobody is building a frontier model without them.

But if the internet analogy holds, this is the fibre-optic cable moment, not the Google moment. The question worth asking isn’t “who has the most compute?” It’s:

What becomes scarce when intelligence itself is abundant?

The Answer: A Shift From Execution to Judgment

When intelligence — the ability to generate answers, options, drafts, and possibilities — becomes cheap and abundant, the scarcity moves elsewhere. Specifically, it shifts toward:

• Judgment over raw knowledge — knowing which answer matters, not just producing an answer
• Asking the right questions over generating answers — the question is now worth more than the response
• Trust over content — anyone can generate content; almost no one can generate credibility
• Real-world validation over computation — a model can simulate outcomes, but reality still has the final word
• Purpose (“why”) over execution (“how”) — the “how” is increasingly automated; the “why” is not

AI can generate possibilities at a scale no human ever could. What it cannot reliably do is:

• Decide what truly matters, out of everything it could generate
• Validate whether an outcome actually holds up in the physical world
• Replace human judgment in ambiguous, high-stakes, or morally weighted situations

Generation is becoming a commodity. Discernment is not.

Where the New Bottlenecks — and Profit Pools — Will Form

If this holds, the entities that capture outsized value from AI won’t necessarily be the ones with the biggest models. They’ll be the ones sitting at the new chokepoints:

Entities connected to real-world feedback loops. Healthcare, robotics, manufacturing — anywhere an AI-generated output has to survive contact with physical reality. A model can propose a drug candidate or a robotic motion plan, but only the real world can confirm it works. Whoever owns that feedback loop owns a scarcity AI can’t replicate.

Those who control trust, credibility, and decision-making frameworks. When anyone can produce plausible-sounding content or analysis at zero marginal cost, the ability to say “this can be trusted” becomes the valuable asset — not the content itself.

Those who define meaning and direction, not just execution. AI is extraordinarily good at the “how.” It has no opinion of its own on the “why.” The organizations and people who can define purpose, set direction, and decide what’s worth doing in the first place will hold leverage that execution alone can’t buy.

Who Might Actually Win This Era

If the bottleneck really is judgment, trust, and real-world grounding, then the likely winners look quite different from “whoever has the biggest model.” A few candidate categories:

Companies that own the real-world feedback loop, not just the prediction. Think of firms in healthcare, industrial manufacturing, robotics, or logistics that can pair AI-generated proposals with a live testing ground — actual patients, actual factory floors, actual physical outcomes. A drug-discovery company with proprietary trial data, or a robotics firm with fleets of machines generating real operating data, can validate what a pure model company can only simulate. The model is replicable; the feedback loop, built over years and often protected by regulation or physical infrastructure, is not.

Distribution and relationship owners who sit closest to the end decision. In the internet era, Google won partly because it owned the moment of intent. In AI, the equivalent may be whoever owns the relationship at the point where a decision actually gets made — a financial advisor’s client relationship, a doctor’s patient relationship, an enterprise’s vendor relationship. AI can draft the recommendation; the trusted party still delivers and stands behind it.

Brands, institutions, and individuals that function as credibility filters. When content and analysis become nearly free to produce, “who said this” matters more than “what was said.” Established publications, professional certifications, audited institutions, and individuals with track records become more valuable, not less — because they’re the scarce signal in a sea of abundant, AI-generated noise.

Vertical operators who fuse AI with regulatory, physical, or data moats. Pure “AI wrapper” companies are easy to replicate. Companies that combine AI with something hard to copy — a regulatory license, proprietary real-world data, physical assets, or long-standing customer trust — are much harder to disrupt. A hospital network, an insurer, or an industrial incumbent that layers AI on top of its existing moat may end up capturing more value than a standalone AI startup with none of those advantages.

People and organizations that set direction rather than just execute. Strategists, founders, product visionaries, and leaders who can define what problem is worth solving — and why — gain leverage as execution gets automated. The scarce skill shifts from “can you build it” to “should this be built, and what should it accomplish.” Those who consistently ask better questions, at the right time, become disproportionately valuable even though they may write very little code or generate very little content themselves.

The common thread: none of these winners are chosen because they have the most intelligence available to them — intelligence is now cheap for everyone. They win because they hold the piece that AI still can’t generate on its own: validated outcomes, trust, relationships, or direction.

Bottom Line

AI won’t eliminate scarcity. It will shift it.

Just as the internet made information abundant and moved value toward attention, AI is making intelligence abundant — and moving value toward judgment, trust, and real-world grounding.

The biggest winners of this era, in all likelihood, won’t only be the ones building the models. They’ll be the ones who correctly identify — and own — the new scarcities.

---

One Tusk