Unum E Pluribus (One From Many)

Some critics see this as an easy parallel to the idea of anthropomorphic AI superintelligence and the core existential and ethical stakes of identity and humanity introduced, if one believes we are on the path to creating something of this nature. No disagreement there, if you believe. But in terms of our own work, the (we believe unsustainably) heavy index concentration and feverish hype around AI and its potential draw another, more prosaic (yet imminently concerning) one.

The index was built to represent the many, as a diversified cross-section of the American economy (industrials, financials, healthcare, energy, consumer staples, etc.) with each constituent a discrete business with its own balance sheet, cycle, and reason for being in or out of favor. A steelmaker, a regional bank, and a pharmaceutical company should not rise and fall together; that independence is the point. Out of many, one coherent representation of an economy.

What we have now is nine companies (Nvidia, Apple, Alphabet, Microsoft, Amazon, Broadcom, Meta, among others) that have grown to represent almost 40% of the S&P 500 by weight, but perhaps most importantly, they are not in very different industries doing very different things; they are all technology and technology-adjacent businesses whose fortunes are increasingly tied to the single thesis that AI will be transformative and its potential justifies current valuations. The passive investor who believes they own the market owns, to a significant degree, a concentrated bet on one idea and a cannibalistic ouroboros of similar companies reliant on one another’s relatively closed-loop buying and selling of each other’s products and services (we examined the circularity earlier in Circular Reasoning and compared this level of concentration with prior market peaks in our Q2 commentary, Amor Fati). Out of one, many ticker symbols expressing the same underlying conviction.

This is the dynamic Pluribus anatomizes. The hive is not inherently malevolent. It is optimizing, efficient, self-reinforcing, and by its own accounting, harmonious. And it is quietly eroding due to this closed-loop nature. To maintain its integrity without being forced to become something other than itself, which is its overarching, frustratingly simple objective, it cannot tolerate a genuinely independent variable and survives by subsuming any challenger. Carol is not a threat because she has any especially dangerous skills or talent. She is simply different; in a system built around conformity, such difference is inherently destabilizing.

The index, at sufficient concentration, produces a structurally similar condition. Price discovery weakens, while capital flows to what’s already large. What gets called ‘consensus’ is just the echo of the last loud voice in a room already in agreement, while what gets called ‘innovation’ is often just last cycle's idea, repackaged, re-rated, and recycled. The feedback loop is not sinister in itself, but simply doing what it was designed to do. And the tower of Babel inexorably rises.

The risk here is not that the thesis is wrong. It may prove right and still be costly to own at the wrong size, because being right about a technology is a different matter from being right about its price, its timing, or which companies capture the gains. Earlier commentaries have addressed the reversion that eventually arrives; the harder challenge facing an institution is what the portfolio has to live through before it does. What counts as too much depends less on our view of AI than on what the institution has to fund and how large a shortfall it can absorb; holding the index at its current weights can be a deliberate choice, so long as it is made knowingly rather than reflexively.

The contrarian investor is, in this environment, the holdout, not indifferent to the logic of the current environment but refusing to be subsumed by it and sensing its inherent limitations. That refusal is structurally destabilizing, and we would argue to our purposes, structurally necessary.

Before the takeover, Carol was a popular and award-winning “romantasy” writer. While she believes she has at least one serious, literary novel in her, she herself denigrates her more famous yet lucrative writing as hacky and formulaic. However, the hive-mind loves it.

The hive has access to everything humanity ever produced, every novel, music score, painting, line of code, etc. It has, in the most literal sense, consumed the entire archive of human creativity. But what does it want more of? Carol's low-brow romantasy she spent years writing for an audience she quietly despised, simply because she's capable of producing something outside the group-thought. The hive is the tyranny of the majority carried to its limit, a danger James Madison warned of in Federalist 10, where he feared a majority driven by a shared passion or interest overriding the rights of everyone else, and which Alexis de Tocqueville later popularized. Madison and his fellow founders understood it as among a democratic process’s most dangerous internal vulnerabilities. It does not elevate taste or critically distinguish innovative thought by the elegance or subtlety of its underlying ideas; it mediates and averages. It brings comfort, familiarity, and the path of least resistance, palatable to everyone, yet challenging no one and devoid of even the hope of novelty, let alone insight. It's the echo chamber and "slopification" rolled into one. Investing has its own version in consensus investing, where the majority view sets the price and the price then appears to confirm the view. A cap-weighted index makes the process mechanical, weighting each dollar toward whichever companies the majority has already bid up the most. A majority can prevail without being right, and so can a consensus, which is why the dissenting investor matters.

So, having flattened all friction, all dissent, all the productive irritation that generates something genuinely new, the collective finds itself dependent on the one remaining source of novelty it cannot manufacture, which happens to be a single unassimilated, cranky, and resistant consciousness. Our real-world implications are too numerous to unpack here, but I'll leave it to your still uniquely human imagination.

Any system built around a single objective tends to push its costs onto whoever sits outside it. The hive lives minimally and will not willingly hurt another living being, even a bug, yet the passive casualties and damage incurred in realizing its objective are treated as externalities, borne as necessary, regrettable in the abstract and invisible in the aggregate. The initial collectivization is not exactly bloodless.

The same logic is visible in the data-center buildout itself. Global data centers consumed approximately 448 terawatt-hours of electricity in 2025, and that figure is projected to reach 945 terawatt-hours by 2030. The same UN University report projects AI water consumption will equal the needs of 1.3 billion people by the same year; a single large data center can consume up to five million gallons of water per day for cooling. In Querétaro, Mexico, expanding data center capacity is already drawing on water supplies amid prolonged drought. Ireland, where data centers consume nearly a quarter of all metered electricity, was forced to halt new connections around Dublin altogether and has since moved to require new facilities to bring their own generation or storage. These are not hypothetical externalities but costs being paid now by people and communities who did not consent to the transaction and will not share in the returns, in exchange for a technology whose primary commercial applications (if, admittedly, undoubtedly useful and additive) range at present from unproven to underwhelming in the face of overblown hype. The hive, at least, promises harmony in return.

Costs like these rarely stay external for long because the people who bear them can vote, organize, and litigate, and governments, ideally, eventually respond. That points to the second kind of risk a concentrated portfolio carries. The first is endogenous, from inside the loop, as when the companies themselves disappoint; the second is exogenous, from outside the loop entirely, from decisions no company or investor controls, such as who builds and pays for the power, who gets priority when the grid is strained, and who is liable for what the platforms deliver.

Those decisions are already being made. In April, the White House invoked the Defense Production Act to expand domestic production of transformers and other grid equipment. Virginia began taxing data center electricity use on July 1, and New Jersey, Illinois, Ohio, Minnesota, and Washington have ended, suspended, or scaled back data center tax incentives. In Congress, the House passed the bipartisan Ratepayer Protection Act, 417 to 3, which would direct states to consider making large data centers bear the cost of the grid upgrades they trigger, but the motion to advance it in the Senate fell short of the 60 votes needed, 57 to 43, after critics called its voluntary approach too weak. Child online safety bills have passed the House and cleared a Senate committee in different forms, with a similar disagreement over how strong they should be.

This is where the election cycle comes in. On November 3, voters decide all 435 House seats and roughly a third of the Senate, and in 2028 they will choose the next president. Those results will not settle whether AI succeeds. They will shape who sets the rules on power, taxes, and liability that the AI buildout depends on, and nothing guarantees those rules will land in a way that helps or hurts the thesis. Three potential, highly contingent paths are worth laying out.

The point is not to guess which path arrives. Each one assigns a different price to power, compliance, and liability, and a portfolio concentrated in the same few companies absorbs any move against them in full, with little else held that would gain from it. A diversified portfolio holds businesses that react differently to the same outcome, so a result that hurts one part of it can help another. (It is a theme we explored in The Ties That Bind, which argues that real diversification means building optionality across scenarios, not merely holding assets spread across categories.)

That protection has a price. In any stretch when the strongest segment keeps leading, diversification looks like a mistake; its value is hard to demonstrate while the house is still standing. The question is how much of the portfolio any single outcome is allowed to decide and what an institution is willing to give up today to maintain the ability to participate tomorrow. The answer takes more patience than a quarterly comparison allows.

The many endowments and foundations that anchor our work exist because someone believed in the particular: a specific community, an overlooked disease, a school in a place others weren’t paying attention, or a composer who needed three more years for their opus. The corpus that funds these commitments was not assembled to track an index. It was assembled to outlast the people who built it and to keep faith with what they intended. Its purposes are to preserve purchasing power, support spending, maintain liquidity, avoid permanent impairment of capital, diversify sources of return, and survive more than one kind of market. A benchmark is a useful measuring stick for how a portfolio did against the crowd, and a fair basis for accountability, but it was never meant to be the mandate, and none of those purposes calls for concentrating in whatever segment of the equity market is currently strongest.

Concentration risk is not abstract to a program officer explaining why the arts grant cycle is suspended; it’s not theoretical to the graduate fellow whose funding evaporated when the portfolio underperformed in the same quarter the market decided the AI thesis needed repricing. Nor does the cause need to be the market itself, since an election, a tax change, or a regulatory ruling can reprice the same thesis just as quickly. The many, in these institutions, are not shareholders; however, they are the reason the endowment exists at all.

What AI promises and what passive concentration in AI enables is a version of the hive's bargain: efficiency, scale, and the elimination of friction, though at least in the latter case we believe the bargain will last only a short while. But when passive concentration is invested in unthinkingly or by default, it removes the price discovery, independent judgment, and productive friction that make markets self-correcting, and in doing so, risks mortgaging the long-term returns it promises to deliver. And particularly in its application, again when used unthinkingly, an LLM can silently remove the texture that makes the work meaningful or far-sighted. For example, the grant reviewer who reads the application twice because something in it doesn't fit the rubric, or the program that gets funded because a chairperson made a judgment call no algorithm would have had the stakes to make. Maybe it’s a casual conversation between an artist and scientist at an event that produced an idea neither could have arrived at alone. These are not inefficiencies to be replaced.