AI Stocks Now Run on Borrowed Money

For the first stretch of the AI boom, the spending had a reassuring quality. The companies building data centres were the most profitable businesses on earth, and they were paying for it out of pocket.

That has changed, and it is the most important shift in the market right now. Capital expenditure at the largest AI stocks is on track to reach around 75% of cash flows, and growth beyond that is increasingly funded by debt rather than earnings.

If you hold a broad index fund, this is not a story about someone else's portfolio. A small group of AI stocks now accounts for about half the value of the S&P 500. Here is what the funding shift changes and what it does not.

Key Takeaways

• A universe of 28 direct AI stocks represents roughly 50% of S&P 500 market capitalisation
• The largest seven companies alone account for about a third of the index, the highest concentration in modern market history
• Capex is heading toward 75% of cash flows, a ratio last seen in the late 1990s technology cycle
• Those 28 companies hold only about 5% of S&P 500 net debt, so the balance sheets remain strong for now
• Debt-funded spending removes the automatic brake that cash-funded spending provides

How much of the index is AI stocks

Index funds are sold on the idea that you own the whole market. That description has been drifting away from reality.

The largest seven companies now make up roughly a third of the S&P 500 by market value, the highest concentration in modern market history. Widen the lens to 28 companies with direct AI exposure and you reach about half the index.

Half. In a fund holding 500 names, one theme decides roughly half the outcome. The remaining 472 companies share the rest, which is a very different product from the one most people believe they bought, as covered in why your index fund is not as diversified as you think.

What that means for a real position

Put numbers on it. Someone holding $100,000 in a total US market index fund has, in round terms, around $50,000 riding on the AI theme and roughly $33,000 on seven individual companies.

Now imagine that same person also holds company stock at a technology employer, plus a technology-heavy fund they picked deliberately. The true exposure climbs well past what any of those decisions looked like individually. That stacking is the practical risk, and it is invisible unless you add it up on purpose.

The funding shift that actually matters

Concentration has been discussed for a while. The newer development is where the money is coming from.

Until recently, AI infrastructure was bought with operating cash flow. That arrangement contains its own safety valve: if earnings soften, spending naturally slows, because there is less cash to spend. The market gets a gradual signal.

Debt financing removes that valve. Borrowed money arrives as a fixed commitment with a repayment schedule attached. It does not shrink when revenue disappoints, and it must be serviced regardless of whether the data centre it paid for is earning yet.

The reassuring detail is that these balance sheets start from a position of real strength. Those 28 AI-linked companies carry only about 5% of S&P 500 net debt while representing 50% of its value. They are not overextended today. The change is directional: the spending is moving onto borrowed money, and that is a different risk profile from the one that existed a year ago.

Why 75% of cash flows is the number to watch

Capital expenditure approaching 75% of cash flows is the statistic that makes experienced investors uneasy, and the reason is precedent.

Goldman Sachs research notes technology companies in the late 1990s reached comparable ratios. The infrastructure they built was genuinely useful, and much of it underpins the internet you are reading this on. That did not prevent a severe repricing, because the spending assumed a revenue ramp that arrived years later than the capital did.

The question is not whether AI is real. It is whether the revenue arrives on the same schedule as the debt service. A data centre financed over several years needs the earnings to show up inside that window, not eventually.

Which connects this directly to the enterprise picture, where most AI pilots still produce no measurable profit impact. The infrastructure is being built against demand that is real but has not yet converted into earnings at anything like the scale of the spending.

What an ordinary investor should take from this

Not that you should sell. Timing a theme this large has a poor track record, and being early is financially identical to being wrong.

Three practical responses.

First, measure what you actually own. Look at the top ten holdings of every fund you hold and add up the overlap. Most people discover the same handful of names repeated across three or four funds they chose for variety.

Second, rebalance on a schedule rather than an opinion. If AI names have grown from a third of your equity allocation to half, selling back to target is not a market call. It is arithmetic, and it happens to enforce selling into strength. Our explainer on what diversification actually requires covers why correlated holdings do not count as spread.

Third, check your time horizon honestly. Money needed within three years does not belong in a concentrated equity position regardless of how convincing the theme is. Money that will sit untouched for twenty years can absorb a drawdown that would be ruinous on a shorter schedule.

Frequently Asked Questions

Is the AI trade a bubble?

The honest answer is that nobody knows in advance, and the label rarely helps. What is measurable is that valuations are historically expensive at roughly 21 times earnings, concentration is at a modern high, and spending is shifting onto debt. Those conditions raise the cost of disappointment without guaranteeing one.

Should I sell my index fund because of AI concentration?

Concentration is a reason to know your exposure and rebalance to a target you chose deliberately, not a reason to exit the market. Investors who sold out of concentrated markets early have historically given up more to missed gains than they saved in avoided losses.

How do I find out how much AI exposure I have?

Open each fund's top ten holdings, which every provider publishes, and add up how often the same companies appear across your funds. Include any employer stock. The total is usually higher than the sum of the individual decisions suggested.

What would signal that the AI buildout is slowing?

Watch capital expenditure guidance rather than headline earnings. Companies revise planned spending before results deteriorate, so a guided reduction in data centre investment is an earlier signal than a profit miss. Rising borrowing costs against unchanged spending plans is the other one to track.

Know what you are holding

None of this requires a prediction. The AI buildout may well earn its cost, and the infrastructure will be useful either way.

What has genuinely changed is that the spending no longer slows down automatically when earnings do, because a growing share of it is committed in advance to lenders. That makes the next disappointment more expensive than the last one would have been, and it makes knowing your real exposure worth an hour of your time.

Disclaimer: This article is for informational purposes only and does not constitute financial advice. Always consult with a qualified financial advisor before making investment decisions. Past performance is not indicative of future results.