The previous piece ended on a question it deliberately did not answer.1 Return compression is a claim about something disappearing from someone's account. Money that leaves one account arrives in another, or stops being spent at all. Either way, there is a destination.
It is also the question the wider compression argument has been deferring.2 Margin compression, valuation compression, and return compression all describe something being squeezed out. None of them say where it lands.
That question turns out to be more interesting than the compression itself, because the answer is not one place. It is four; they are wildly unequal in size, and the order in which they are usually discussed is close to the reverse of the order that matters.
The Arithmetic Was Always Uncomfortable
Begin with an observation that predates every model in this argument by thirty-five years.
Before costs, the return on the average actively managed dollar equals the return on the average passively managed dollar. After costs, it is lower. This is not an empirical finding that better data might overturn. It is arithmetic. Active managers collectively hold the market, so collectively they earn the market, minus whatever they spend trying not to.3
Which means the thing the industry has spent decades calling alpha was never, in aggregate, a pool of return sitting on top of the market. It was a distribution around it. Every dollar of outperformance had a matching dollar of underperformance held by somebody who did not read the filing, did not rebalance, or sold in March.
So the first answer is deflationary, and it is also the largest: a great deal of the displaced return is not displaced anywhere. A transfer simply stops occurring.
Compression does not drain a pool. It narrows a distribution.
It Goes Back to the Person Who Was Paying It
Read the dispersion argument from the other side.
A mediocre investor going from 3% to 6.5% is not receiving a gift from the market. They are keeping money that used to leave their account and arrive in somebody else's. The counterparty who harvested their panic selling, their neglected rebalancing, and their unread filings is the party whose return actually fell.
This is the largest channel and the least discussed, for a structural reason. It accrues to people who never knew they were paying it. Nobody receives an annual statement itemizing what their inattention cost them. The loss was silent, so the recovery is silent too, and silent recoveries generate no commentary and no constituency.
It is also the channel that makes this transition genuinely good news for most participants, which is worth stating plainly in an argument that otherwise reads as a warning. Most people are on the donating side of that trade. They have been for their entire investing lives.
It Goes to the Toll Booth
The second channel is less comfortable, and it has a number attached.
Averaging over 1980 to 2006, investors spent roughly 0.67% of the aggregate value of the market every year searching for superior returns, about $102 billion in 2006 alone. Capitalized, the cost society pays for price discovery runs to at least 10% of total market value.4
That is the size of the toll booth. It is not a transfer between investors. It is a payment out of the investor pool entirely, to the people who supply the searching: managers, analysts, data vendors, platforms, exchanges.
The optimistic reading is that AI collapses the cost of search. An agent does for a subscription what a research department did for a payroll line, that 0.67% falls, the saving stays inside the investor pool, and the market stops capitalizing a tenth of its own value to fund the discovery of its own prices.
Half of that is visibly happening. The asset-weighted expense ratio across US funds reached 0.32% in 2025, less than half what investors paid two decades ago and worth about $6.8 billion to them in that year alone.8
Now follow the other half, which is what happens on the way out of the cost base.
Institutional fees have compressed about 3% a year since 2010. Over the same fifteen years assets under management tripled, reaching $147 trillion, while revenues grew at 5.1% a year against costs at 5.4%. Operating margins finished that period close to 30% — roughly where they stood in 2010.9 Fifteen years of asset growth bought the industry no margin at all. Then, in August, BCG told asset managers where the margin is.
Important
A traditional manager running a cost base of 15 to 20 basis points that "reshapes their organization to deploy AI at scale could reduce expenses by 3 to 6 basis points, perhaps a 25% to 30% cut."
Set the two numbers beside each other.10 The investor's fee line is 32 basis points, and still falling, though by Morningstar's own account at a slowing pace. The manager's cost base is 15 to 20 basis points and can fall by a quarter to a third. That gap is not a saving in transit. It is the first margin expansion available to this industry in fifteen years, and it arrives at precisely the moment the thing being automated is the searching.
The precedent said it would go this way. High-frequency trading attacked a known inefficiency and the rent did not return to investors; it dissipated into the arms race. More than a fifth of trading volume in the BIS sample occurred in latency-arbitrage races,5 contests over fractions of a tick fought with enormous capital expenditure. The inefficiency shrank. The spending did not. It changed denomination, from spread to infrastructure.
An arms race does not return the rent. It converts it into a cost of entry.
Apply that one level up the intellectual stack and the shape is familiar: subscriptions, proprietary data licenses, inference spend, latency to the model rather than to the exchange. The fee line falls, the cost base falls faster, and the difference is booked as margin. Unlike the first channel, this is not a transfer that stops. It is a transfer that changes recipient.
The expense ratio is the part of the toll booth anyone measures, and it is the part that is falling. What French was counting — trading costs, and the whole apparatus of searching — nobody has re-run since 2006. So I can show you the fee line falling and the margin about to move. I cannot show you the total, because in the twenty years since it was last measured, nobody has rebuilt the series.
It Goes to Whoever Owns What Stays Scarce
The third channel is where the return that was genuinely real ends up.
When information, execution, and portfolio construction commoditize, the compensation attached to them travels with them. What does not commoditize is capital, duration, liquidity provision, control of scarce assets, regulatory privilege, and the willingness to hold a position at the moment every optimizer recommends trimming it.
Notice what those have in common. Not one of them is a skill. Every one is an endowment: something a participant either has or does not, and mostly something bought rather than learned.
That distinction decides the distributional outcome. Cheap analysis is genuinely democratizing, because the agent costs roughly the same for a retail investor as for an endowment fund. Cheap analysis does not democratize a balance sheet. The same technology flattens the premium on the thing anyone can now acquire while leaving untouched the premium on the thing almost nobody can.
The return does not merely move sideways. It moves upward.
The fundraising market is already pricing it that way. Established firms took 90.9% of US venture capital raised in the first quarter of this year. Through May, $62.4 billion went into 288 funds, against $66.1 billion across 537 funds for the whole of 2025 — nearly the same money, into roughly half as many vehicles. The median fund now takes about fifteen months to close, the longest in more than a decade.11 A manager arriving with a record and no access is finding out what a record is worth now.
This Is an Incidence Question
There is a version of this argument already written, asked about productivity rather than about alpha.6
When a productivity gain arrives, it goes to capital, to labor, or to customers, and which one is determined by pricing power rather than by contribution. The dividend is real. Its destination is contested. And the party that created it is frequently not the party that keeps it.
Market efficiency is the same event wearing different labels.
The efficiency gain is real. Its incidence is decided by who holds the scarce input. And the professional whose edge was diligence, who read the filing others skipped, who rebalanced on schedule, who kept their head in a drawdown, created the gain and holds none of the inputs that capture it.
The gain is real, and it is not theirs.
The Exposed Band
This produces a result that should look familiar to anyone who has followed the labor side of these arguments.7
The bottom of the distribution gains, because the transfer out of their accounts stops. The top gains, because the scarce factors concentrate there and the competition for them thinned out. The band in the middle, competent and diligent and professionally serious, compensated for operating machinery that has now been automated, loses at both ends simultaneously.
They lose the premium, because the premium was payment for running the machinery. And they cannot retreat into the surviving advantages, because those require capital rather than effort. Diligence does not convert into duration. Judgment does not convert into a balance sheet.
Important
This is not a story about intelligence being devalued. It is a story about one particular kind of intelligence becoming abundant, and compensation following the scarcity rather than the intelligence.
The Weights, Not the Analysis
Active return is the gap between portfolio weight and benchmark weight, multiplied by what the security did. Commoditized analysis compresses disagreement about the second term. It does nothing whatever to the first, and the first is where the money has been.
This is why manager return dispersion has been widening rather than narrowing. The three-year rolling excess return for bottom-decile managers ending in the second quarter of 2026 was −8.3%, against an average of −3.9% over the past two decades.12 Look at what those managers were holding and the number stops being about analysis at all.
Over the three years to that date the bottom decile ran −10.15% against the benchmark while the top decile ran +5.43%. The losers were not diversified into irrelevance. Their ten largest holdings were 55.0% of assets, against 45.6% for the winners. They were more concentrated. They were concentrated somewhere else: 23.2% in technology against 35.6% at the top, and 30.5% in financials and healthcare against 20.8%. And the gap has been opening for a decade — bottom-decile managers sat 1.2 percentage points below the universe in technology across 2009 to 2013, and 5.9 points below it across 2020 to 2026.12
That is not a failure to know what to own. It is a failure to own enough of it.
The first thing AI commoditizes is knowing what to buy. The last thing it commoditizes is how much to own.
Two managers can reach an identical and correct conclusion about a company and still hold it at 10% and at 2%, because what separates them is mandate, risk limit, capital structure and nerve — not research. And the weight gap has never mattered more: the top ten companies in the S&P 500 are now more than a third of the index.13 At that concentration, a two-point difference in a single position outweighs a decade of careful selection everywhere else, and the penalty compounds against whoever is underweight. The winner rises, its benchmark weight rises, and the manager is further underweight than before without having traded.
The inattentive investor who stops donating alpha is moving toward the index — the likeliest thing an agent ever tells them is to own it. The active manager is obliged to move away from it, because reproducing the consensus portfolio cannot justify the fee. Commoditized analysis pushes those two populations in opposite directions, and only one of them is in this sample.
So this is a claim about the rent attached to information, not about return dispersion. The two can move in opposite directions, and right now they are.
The Answer
The displaced return goes to four places, in descending order of size and ascending order of comfort.
Most of it was never a return at all. It was a transfer from the inattentive to the attentive, and when the inattention is automated, the transfer stops rather than relocating. The next largest share goes to whoever sells the capability, and the industry has already priced what that is worth to itself. A real but smaller share migrates toward the endowments — capital and duration and liquidity and control — and concentrates as it travels. A residual is simply saved: a cost the market used to pay for price discovery that it no longer has to pay.
The return did not go missing. It went backward to the people who were quietly paying it, sideways to the people selling the tools, and upward to the people who own what no model can manufacture.
The only party it did not go to is the one that used to earn it.