So the empires keep coming back. Post 2 ended on that, and it’s the bit that nags. Standard Oil got broken into thirty-odd pieces and Rockefeller got richer; two of those pieces, Exxon and Mobil, later merged right back together.¹ The Baby Bells got split apart and reassembled into a couple of giants. The East India Company got nationalised, not dispersed. Every time someone clears the board, the pieces seem to drift back into the same shape. Which raises the obvious, slightly bleak question: is that just what gravity does? Is one-person concentration the resting state of the world, and everything else a temporary intervention that wears off?

I want to be honest that this is an old fight, older than any of the names in this series. Strip away the trillion-dollar pay packages and the Starlink terminals and you’re left with a question political scientists have been having since the 1950s without ever quite settling it: does power, left alone, gather into the hands of a few? Or does it spread itself out? Almost everything in these posts has been one long argument about that, without me ever saying so. Time to say so.

The fight nobody finished

The cleanest version of the disagreement is two American books, five years apart.

In 1956, C. Wright Mills wrote The Power Elite and said: yes, obviously, one elite rules. Not a conspiracy, nothing so crude. Just a small, interlocked set of people sitting atop the big corporate, military, and political hierarchies, who went to the same schools, sit on each other’s boards, and move between the three worlds as if they were one.² The issues that look like democratic contest, in his telling, get decided well above the level where voting happens.

Then in 1961, Robert Dahl wrote Who Governs?, studied who actually made the decisions in one American city, and said: no. Power is pluralistic. There isn’t one elite, there are many, each strong in its own patch and weak everywhere else, and they check each other. The business lobby loses to the unions on one thing and beats them on the next. Nobody is in charge of everything, because the moment anyone tries, a competing bloc forms to stop them.³ This is the optimistic answer, and it isn’t a naive one. It’s the intellectual backbone of “don’t worry, it self-corrects.”

For a while Dahl basically won the argument in the academy. Then G. William Domhoff spent the next several decades publishing Who Rules America?, which is essentially “Mills was right, and here’s the data” updated every few years.⁴ And here’s the twist I find genuinely disarming: Dahl himself, late in life, drifted toward the gloomier view. The founder of pluralism, watching money pool at the top through the 2000s, started writing about how economic inequality was corroding the political equality his whole theory rested on.³ When the optimist’s own author migrates to the pessimist’s side, you should at least take the pessimist seriously.

So that’s the shape of it. Now the economics, which is where the argument actually gets numbers. There are roughly four camps, and I want to build the optimistic one to full strength before I knock anything down, because the lazy version of this post strawmans it.

Camp one: the gradient runs downhill

The first camp says concentration is the default because the math of money quietly favours it.

The famous version is Piketty’s r > g. When the return on capital runs ahead of the growth rate of the economy, the wealth that already exists compounds faster than the wealth people earn, so old money pulls away from everyone living on a wage, and absent some shock the gap just widens on its own.⁵ I find this one genuinely hard to wave away, and not because I’ve done the maths. It’s that it doesn’t need anyone to be greedy or clever. It’s arithmetic. The money does it while everyone’s asleep.

And underneath the arithmetic is something dumber and, to be honest, even more convincing: the rich get richer because being rich is itself the qualification for getting richer. Network people call it preferential attachment, the node that already has the most connections is the one most likely to get the next one, and once you’ve seen it you see it everywhere, in citations, in followers, in capital.⁶ Organisations do the same thing to themselves, which is the old “iron law of oligarchy”, any group, even one founded to be radically democratic, drifts toward being run by a small clique because someone has to run it and incumbency is sticky.⁷ Notice none of these need a villain. That’s exactly what makes them feel like gravity instead of a crime.

The modern accelerant is winner-take-all. Sherwin Rosen wrote about the “economics of superstars,” and Frank and Cook turned it into a book: once the best performer can serve the entire market at near-zero marginal cost, the market tips to one. A slightly better search engine doesn’t get a slightly bigger share, it gets nearly all of it.⁸ Brian Arthur added the lock-in piece, increasing returns and path dependence, the idea that an early lead becomes self-reinforcing and the better product can still lose.⁹ Put it together and you have a clean structural story for why digital markets keep producing one giant per category.

This camp is strong. It’s also, I think, the one most likely to be overstated, which brings me to the people who fight it.

Camp two: the gale that topples the kings

The optimist’s name is Joseph Schumpeter, and the word is creative destruction. Yes, monopolies form, the optimist says, but the very profits that make them fat are what attract the innovator who kills them. The incumbent isn’t toppled by a regulator. It’s toppled by the next thing. The dominant firm of one era is the cautionary tale of the next, and the data on corporate lifespans is brutal in a way that should comfort anyone worried about permanence: the average tenure of an S&P 500 company has fallen from around 61 years in the late 1950s to roughly two dozen now, and it’s still heading down.¹⁰ Empires die. They die faster than they used to.

And the live version of this argument is sitting right in the middle of 2026. The optimist’s best card today is AI diffusion. The frontier looked like the ultimate moat a couple of years ago, a thing only three labs with billion-dollar compute budgets could touch. Then open-weight models started commoditising it, capability leaking outward, the gap between the frontier and what you can run yourself narrowing month over month. The honest historical analogy is the printing press: a tool that, once it diffused, broke the monopoly the literate few held over knowledge itself. If that’s the pattern AI follows, it’s a gale, not a moat. There’s also a serious scholarly mugging of the lock-in story worth knowing about. Liebowitz and Margolis went back to the canonical examples everyone cites, QWERTY keyboards and VHS beating Betamax, and argued the “we got locked into the worse product” tale is mostly myth. The markets, they say, didn’t actually trap us.¹¹

I want this camp to land at full weight, because it might be right. But I can’t pretend the trend lines support it cleanly, and this is the part the optimists tend to skip. Business dynamism in the US has been falling since around 1980: fewer new firms, less churn, more entrenchment. Markups, the gap between what firms charge and what it costs them to produce, rose from about 1.21 to 1.61 between 1980 and 2016, which is the signature of firms with more pricing power, not less.¹² Frontier firms hoard patents. The creative destruction this camp depends on may itself be weakening, even as the theory predicting it stays elegant. (To be fair, the causation is genuinely contested. The trend is solid; what it means is argued.) So the optimist’s gale is real, but it’s blowing weaker than it used to. Hold that thought, because it’s the hinge of the whole series.

Camp three: it only breaks under catastrophe

The third camp is the one that ruins your week. Walter Scheidel, a Stanford historian, went looking through several thousand years of records for cases where serious inequality actually fell, and came back with a grim finding: it almost never falls peacefully. His “Four Horsemen” of leveling are mass-mobilisation warfare, violent revolution, state collapse, and lethal pandemic. Those flatten the distribution. Reformist policy, tax tweaks, the things we’d actually choose, mostly move it at the margins.¹³ In Scheidel’s telling, the mid-century moment when the rich got relatively poorer wasn’t a triumph of good governance. It was the two World Wars and the Depression smashing capital, and it un-did itself the moment the catastrophe receded.

Mancur Olson supplies the mechanism for why, absent a shock, things only get more stuck. Stable societies accumulate what he called distributional coalitions, interest groups that get very good at carving out advantages for themselves and very bad at letting anyone dislodge them. Over time the economy ossifies under their weight, until some external shock clears the board.¹⁴ But Olson also hands us the single best pro-concentration argument in the whole literature, and an honest post has to state it: the “stationary bandit.” A roving bandit steals everything and moves on. A stationary bandit, a settled ruler who knows he’ll be taxing the same people next year, has an incentive to let them prosper so there’s more to tax. Concentrated, durable power, in this reading, isn’t only theft. It can be the thing that ends the chaos and lets anyone build anything at all.¹⁴ That’s not a strawman I’m setting up to knock down. It’s the genuine case for the strongman, and it’s why “concentration” and “bad” aren’t synonyms.

Now, before this hardens into fatalism, the counter-evidence in the same breath, because Scheidel’s “never peacefully” is contested at the edges. The postwar welfare states of Western Europe dispersed wealth fairly durably without a fresh catastrophe to do it. And there’s a live example most leveling theorists didn’t have when they wrote: India in 2024, where a governing party widely expected to romp home was knocked below a parliamentary majority by ordinary voters. No war, no collapse, just a ballot.¹⁵ Peaceful dispersal is rare. It is not impossible. The difference between “rare” and “impossible” is the entire space in which politics happens.

Camp four: it’s a choice, not a law

Which brings me to the camp I actually lean toward, and the one that turns the question from arithmetic into argument.

Daron Acemoglu and James Robinson, who won the 2024 Nobel in economics for roughly this body of work, say the whole “is it a law of nature” framing is wrong.¹⁶ Inequality and concentration aren’t the output of a mechanical process. They’re the output of institutions, and institutions are choices. Extractive institutions, the kind that let a narrow elite write the rules and rake off the gains, concentrate power. Inclusive ones spread it. Same arithmetic, same technology, wildly different outcomes depending on the rules a society chooses to live under. And the cleanest evidence for “it’s a choice” is the thing that was supposed to prove it was automatic and then collapsed.

For about fifty years the textbook answer to “does inequality fix itself?” was the Kuznets curve. Simon Kuznets, 1955, sketched an inverted U: as a poor economy industrialises, inequality rises, but as it matures, inequality falls again, automatically, as a natural consequence of development.¹⁷ Reassuring. You didn’t have to do anything; the curve bent back down on its own. Then the data kept coming, and after about 1980 the downward leg didn’t just flatten, it reversed. Inequality in the rich world turned back up, the inverted U became something closer to a U, and the self-correcting story died on contact with the numbers.¹⁷ Acemoglu and Robinson’s reading is the knife: the mid-century fall that fooled everyone into believing in the curve wasn’t automatic at all. It came from political coalitions, unions, progressive taxation, mass enfranchisement, things people built, and when those got dismantled after 1980, concentration came right back. The curve was never a law. It was a policy regime that people mistook for physics, and then stopped maintaining.

There’s an older name for that policy regime: John Kenneth Galbraith called it countervailing power. Concentration on one side of a market, he argued in 1952, tends to summon organised power on the other, big retailers against big manufacturers, unions against big employers, regulators against big finance, so the system can rebalance itself without anyone having to smash it.¹⁸ It’s the optimistic structural theory, the answer to Mills. But it has a fatal condition attached, and the condition is the whole ballgame: countervailing power fails when one side captures the state itself. When the bloc that’s supposed to do the checking is owned by the bloc it’s meant to check, the dispersal mechanism stops dispersing. Which is exactly why the absence of countervailing power today, weak unions, captured regulators, money pooling in politics, reads less like an accident and more like a diagnosis of why dispersal has gotten so hard.

I’ll add one honest complication to my own preferred camp, because the structural pessimists aren’t wrong about everything. Even Piketty’s r > g, the cleanest version of the gradient, took a serious hit: Matthew Rognlie showed that most of the rising capital share Piketty documented is actually housing, not capital-in-general doing its inexorable compounding thing.¹⁹ So the “structural default” is real as a tendency and shaky as a law. Good. That’s the point. If it were a law, there’d be nothing to argue about, and no choice to make.

The honest verdict

Let me just say it plainly, because the whole series has been circling this and dodging would be cowardly.

Concentration is the strongest default. Capital compounds faster than wages, organisations tilt toward the few, and advantage snowballs in money, attention, and networks all at once. Left genuinely alone, the gradient runs toward one person. That part of the pessimist case I think is basically true.

But “default” is not “fate,” and the thing the evidence actually screams once you line it all up is stranger and more useful than doom: it’s the dispersal that’s the anomaly. Concentration is easy, cheap, and self-sustaining. Dispersal is rare, expensive, and almost always imposed from outside the concentration, by a court, a legislature, a revolution, a plague, or a deliberately built coalition, and it doesn’t stay imposed unless something keeps actively holding the pieces apart. The empires in Post 2 didn’t reassemble because anyone willed it. They reassembled because nobody was spending the energy to keep them apart, and apart is the unnatural state.

That reframes everything. The question was never “why does power concentrate,” that’s just gravity. The real question, the one Acemoglu and Robinson get the Nobel for and the one this whole series turns on, is: where does the correction send the power when it finally comes? Out, into many hands, the way the US dispersed Standard Oil? Or up, into one even bigger hand, the way a state crackdown can break a tycoon only to hand his power to the party? Correction isn’t the same thing as dispersal. A board can be cleared and re-stacked into the exact same pile.

So, is any of this avoidable? Honest answer: the concentration isn’t. The shape it leaves behind is. I’d love to hand you something cleaner than that, but I’d be making it up.

Which leaves the one question I can’t answer from theory, only watch unfold. The optimist’s gale and the pessimist’s moat are, in 2026, the same object. AI is either the next great diffusion that commoditises the frontier and topples whoever’s sitting on it, the printing press all over again, or it’s the ultimate lock-in, the moat so deep it crowns the few who got there first and never lets anyone else across. Same technology. Opposite endings. Nobody knows which yet, including the people building it.

So where is this actually going? That’s the next part: not what the theory says should happen, but how you’d actually bet on it.

References

Footnotes

1. Standard Oil (Wikipedia); the 1911 Supreme Court dissolution split the company into 34 successor firms, and Exxon and Mobil (two of those successors) re-merged in 1999. ↩

2. The Power Elite (C. Wright Mills, Oxford University Press, 1956); the thesis of a single interlocked corporate-military-political elite. ↩

3. Robert A. Dahl (Britannica, on Who Governs?, 1961); the pluralist counter-thesis. Dahl’s later-life turn toward concern about economic inequality eroding political equality is noted in the same body of work. ↩ ↩²

4. Who Rules America? (G. William Domhoff, 1967 and updated since); the empirical case that Mills was substantially right. ↩

5. Capital in the Twenty-First Century (Thomas Piketty, 2013); the r > g argument that capital returns outpace growth, concentrating wealth absent shocks. ↩

6. Preferential attachment (the Pareto / Merton “Matthew effect” / Barabási lineage of cumulative-advantage models). ↩

7. Iron law of oligarchy (Robert Michels, 1911); “who says organisation, says oligarchy.” ↩

8. The Winner-Take-All Society (Robert Frank & Philip Cook, 1995; building on Sherwin Rosen’s “economics of superstars”). ↩

9. W. Brian Arthur (on increasing returns, lock-in, and path dependence). ↩

10. Corporate-longevity figures: the ~61-year average tenure of the late 1950s traces to McKinsey / Richard Foster’s work; the decline to the low-to-mid 20s by 2016, and the further projected drop, are Innosight’s (2021). ↩

11. Liebowitz & Margolis, “The Fable of the Keys” (Journal of Law and Economics, 1990), disputing the QWERTY/VHS lock-in stories; see QWERTY (Wikipedia) for the dispute. The parallel Standard Oil revisionism is Dominick Armentano, Antitrust and Monopoly (1982). ↩

12. The Rise of Market Power and the Macroeconomic Implications (De Loecker, Eeckhout & Unger, Quarterly Journal of Economics, 2020); average markups rose from ~1.21 to ~1.61, 1980 to 2016. ↩

13. The Great Leveler (Walter Scheidel, Princeton University Press, 2017); large inequality has historically declined only through mass-mobilisation war, violent revolution, state collapse, or pandemic. ↩

14. The Rise and Decline of Nations (Mancur Olson, 1982); distributional coalitions and institutional sclerosis. The “stationary bandit” pro-concentration argument is Olson’s later work on why a settled ruler has an incentive to let subjects prosper. ↩ ↩²

15. 2024 Indian general election (Wikipedia); the BJP fell short of a parliamentary majority, a peaceful electoral check. Paired with the Supreme Court striking down the Electoral Bond Scheme (Al Jazeera, Feb 2024) as a live non-catastrophic dispersal mechanism. ↩

16. Why Nations Fail (Daron Acemoglu & James Robinson, 2012; the authors won the 2024 Nobel Memorial Prize in Economic Sciences); inequality as a product of extractive vs inclusive institutions, i.e. a political choice. See also The Narrow Corridor (2019), which models the US/China/India spread. ↩

17. The end of the Kuznets curve (Monash, on Kuznets’ 1955 inverted-U and its post-1980 reversal); Acemoglu & Robinson argue the mid-century leveling came from political coalitions, not an automatic mechanism. ↩ ↩²

18. American Capitalism: The Concept of Countervailing Power (John Kenneth Galbraith, 1952); the dispersal theory that concentration summons opposing blocs, which fails under state capture. ↩

19. Deciphering the fall and rise in the net capital share (Matthew Rognlie, Brookings, 2015); the rise in the capital share is mostly housing, denting Piketty’s mechanical r > g reading. ↩