AI and Pareto

At the end of the nineteenth century, Vilfredo Pareto noticed a stubborn pattern in the data.

Studying land ownership and income in Italy, he found that wealth did not distribute evenly. A small minority held a disproportionate share, while the majority held comparatively little. He looked elsewhere and saw the same shape again and again. The exact percentages changed. The asymmetry didn't.

Pareto wasn't making a moral claim. He was documenting how things work: left to their own dynamics, human systems concentrate value, influence, and outcomes in the hands of a few.

Much later, we turned that observation into a slogan: the 80/20 rule. The numbers were never the point. The curve was.

And for a remarkably long time, the curve held.

The Resilience of the 80/20 Curve

For more than a century, the distribution Pareto described proved stubbornly resilient.

Across industrialization, digitization, and globalization, the pattern bent but rarely broke. A small fraction generated most of the value, but the imbalance stayed bounded. Some people mattered more than others. Some firms dominated entire industries. But the skew rarely exceeded what our intuition could tolerate. Teams still mattered. Institutions still required bodies. Work still demanded many hands.

Even in public equities—one of the cleanest laboratories for concentration—the curve rarely went vertical.

By 2018, the five FAANG stocks made up more than 11% of the S&P 500. By early 2020, the largest tech names approached nearly 18% of the index's total value. That's dramatic, but it's not total capture. Even at those peaks, the majority of market value still lived outside the winners. The curve steepened, but it held. The distribution remained recognizably Pareto-shaped: persistent, asymmetric, but bounded.

The same pattern appeared inside organizations. When Elon Musk gutted Twitter's headcount in 2022, the instinctive prediction was collapse. Yet the platform continued to operate, with a smaller fraction of contributors carrying most visible activity.

What stuck with me wasn't the drama. It was the familiarity of the outcome. I kept expecting the curve to finally break, and instead it settled back into the same shape.

Because the contour was enforced by limits. Not moral limits—mechanical ones. The finite bandwidth of an individual. The managerial capacity of a company. The capital constraints that slow expansion. These weren't inconveniences—they were the boundary conditions of modern work. They kept the distribution steep, but they kept it from going vertical.

For more than a century, systems resisted true extremes. You could optimize toward concentration, but the distribution rarely collapsed into shapes like 95/5 or 99/1—not because the world was fair, but because the physics of scale pushed back.

Until very recently.

The Inflection

For most of my life, I favored the term "machine learning" over "AI." The outcomes rarely felt satisfying enough to justify the word intelligence. Useful, yes. Sometimes impressive. But AI sounded like a claim the results didn't earn.

Then ChatGPT arrived, and the abstract became operational. Not a breakthrough in a lab, but a capability in the hands of everyone. And the market responded in the only language it truly speaks: capital.

Nvidia's valuation didn't just rise—it signaled a regime change. Suddenly, the "AI story" wasn't a niche narrative. It was the organizing thesis of the entire economy.

You can see it in how quickly concentration emerged—not just in consumer apps, but across the entire AI supply chain. Compute. Memory. Networking. Infrastructure and energy. The stack didn't just grow; it hardened into an ecosystem where value pulls inward, toward the few chokepoints that turn electrons into intelligence.

And you can feel it in productivity. The constraints that once kept output distributed—time, headcount, managerial capacity—are being replaced by something else: tool leverage, iteration speed, and the ability to turn intent into execution without an entire organization behind you.

The curve didn't just bend. It began to tip.

Stretching the Pareto

The curve is starting to behave like it's no longer bounded—not because Pareto stopped being true, but because AI is pulling disproportionate gravity.

The first signal is capital, and it's not subtle.

In most cycles, money follows traction. In this one, money is trying to manufacture advantage. It's pouring into AI as a full-stack land grab—the model layer, the application layer, the tooling layer, the infrastructure layer. An attempt to own the chokepoints where value will pool: compute, distribution, proprietary data, and the workflows where human labor gets converted into outcomes.

When capital compresses this hard into one sector, it doesn't just accelerate innovation—it reshapes the distribution of winners. It turns a modest lead into a compounding one. The top of the stack stops looking like a thousand competing companies. It starts looking like a small number of gateways through which the rest of the economy has to pass.

You're seeing multiple contenders in the same arenas funded in parallel—because investors aren't sure which company becomes the gateway, but they're increasingly sure that a gateway exists. Once you believe the loop compounds—usage → data → product → distribution → usage—"too much capital" stops being a coherent objection. Missing the winner isn't a missed return. It's missing the interface.

But the second signal is even stranger: productivity itself is stretching Pareto.

We're watching founders in their twenties build companies that look like empires on timelines that would have sounded absurd a few years ago. Companies like Cursor, Mercor, Lovable. Extremely small teams, insanely fast iteration, growth curves that feel less like startups and more like compression algorithms applied to industries.

This is not just "people working harder." It's a fundamental change in the economics of execution.

When model capability rises, the cost of producing a first pass collapses. When retrieval is cheap, learning speed spikes. When iteration loops tighten, output compounds. The people who can run that loop—specify clearly, delegate aggressively, evaluate ruthlessly—start operating on a different slope of the curve.

And once that happens, the distribution doesn't just get more unequal. It gets more extreme.

Capital is concentrating into AI. Productivity is concentrating into a smaller set of builders. And together, they're not just bending Pareto—they're pushing it toward the limit.