· Solveion · Perspectives · 6 min read
Your index fund is an AI bet
A generation learned one rule: buy the broad index, hold, do not think about it. It worked. It also quietly turned the most popular savings product in the world into a concentrated position in one capital expenditure cycle.
Somewhere in the last decade, personal finance collapsed into a single piece of advice. Buy the broad index fund. Hold it. Ignore the noise. It is repeated on every social platform, in every forum, by people with no incentive to mislead anyone, and it has been excellent advice for fifteen years.
We are not going to argue it was wrong. We want to point at something that happened while everybody was following it, because we think the link between that advice and the AI build-out is much tighter than almost anyone holding either position realises.
This is analysis of a structural link, not investment advice, and we are not qualified to give any.
The diversification is not what it says
Start with what a broad US index fund actually contains today.
The ten largest companies now account for roughly 37 to 41 per cent of the S&P 500’s weight, depending on the day you measure. At the end of 2015 that figure was about 19 per cent. It has more than doubled in a decade. The Magnificent Seven alone are over 30 per cent of the index.
There is a second number that matters more than the first. Those top ten names carry around 41 per cent of the index weight while producing roughly 32 per cent of its earnings. That gap is not necessarily irrational — these are among the most profitable companies ever built, and a premium can be deserved — but it means the concentration is larger in price than in profit.
The practical consequence is that a person who buys a broad index fund believing they have bought five hundred companies has in fact placed about forty cents of every dollar into ten, most of which are exposed to the same technology cycle. That is a defensible position to hold. It is just not a diversified one, and it is being sold as diversification.
Why the flows matter more than they used to
The second piece is who is doing the buying, and how.
A large and growing share of the money entering equities each month is not making a decision. It arrives through payroll deduction, automatic enrolment and target-date funds, and it buys the index at whatever price the index happens to be. Nobody in that chain is forming a view that anything is cheap.
Academic work by Gabaix and Koijen on what they call the inelastic markets hypothesis found that a dollar flowing into the equity market raises aggregate market value by roughly five dollars. The mechanism is that the marginal holders — index funds, pension funds, insurers — operate under mandates that fix their allocations, so when demand shifts there is very little price-sensitive capital available to absorb it and prices must move a long way to clear.
The estimate is contested and hard to identify cleanly. But the direction is not really in dispute: when price-insensitive buyers dominate the margin, prices respond more to flows and less to fundamentals. And the same arithmetic runs in both directions.
The loop nobody designed
Now put the two halves together, because this is the part we find genuinely striking.
Automatic contributions buy the index. The index is disproportionately a handful of megacaps. Those megacaps are spending extraordinary sums on AI infrastructure — the five largest have guided to something like $775 to $800 billion of capital expenditure this year, roughly three times the 2024 figure, of which about three quarters is AI-specific. PIMCO estimates that capex will consume something close to 94 per cent of their operating cash flow, and Morgan Stanley and J.P. Morgan project the sector will need to issue around $1.5 trillion of new debt over three years to fund the rest.
Some of that spending is also circular. Vendor equity stakes, take-or-pay compute agreements and debt-funded GPU purchases among a small group of interlocking companies can make end demand look larger and more independent than it is. The comparison being drawn is to the vendor financing that inflated Lucent and Nortel between 1999 and 2001, and we think that comparison is being made carefully rather than hysterically.
So the chain runs: a payroll deduction, into an index fund, into ten companies, into data centre construction, into GPU orders, whose revenue is booked by another company in the same index, which raises the index, which the next payroll deduction buys. Nobody in that loop chose it. Most of the people funding it would not describe themselves as investors in AI infrastructure at all.
Where we think the honest uncertainty is
We want to be careful here, because the version of this argument that circulates online is much more confident than the evidence supports.
The passive-flows-will-break-the-market thesis has been made for over a decade by serious people and has not happened. Index funds are, in the standard rebuttal, price takers rather than price setters — the marginal price is still discovered by active participants trading around them. The earnings underneath the concentration are real, not accounting fictions. And the demographic reversal that the argument depends on may be a long way off: roughly 95 per cent of investors under forty are in passive vehicles against about 25 per cent of those over seventy, which means the cohort most likely to be selling in the next decade is the one least invested in the mechanism.
Most importantly, attempting to time any of this has destroyed far more household wealth than any crash has. The advice to buy and hold has outperformed nearly everyone who tried to be clever about it, including nearly everyone making this argument.
So we are not forecasting a crash, and we would be suspicious of anyone who is.
The narrower claim, which we think holds
What we do think is true, and underappreciated, is this: the correlation between “my retirement savings” and “the AI capital expenditure cycle” is now far higher than either label suggests, and it got that way without anybody deciding it should.
That matters for a business audience for a specific reason. We have written that today’s AI tooling is priced below what it costs to serve, and that a large valuation is best understood as an option on something that does not exist yet. Both of those depend on capital continuing to arrive on current terms. If capex discipline arrives — not a crash, just a normal tightening — the subsidised pricing goes first, well before anything dramatic happens to a share price.
Which means the question worth asking is not whether markets are overvalued, a question nobody can answer. It is a much more ordinary one: if the cost of the AI tools your operation now depends on tripled over eighteen months, what would you do? That is a planning question, it is answerable, and the loop above is the reason it is not hypothetical.