By Thierry Hasse, Chief Investment Officer
Elevage Partners | August 17, 2026
If you own an S&P 500 index fund, you probably think of it as the safe, diversified core of a portfolio. That is what the label has always meant. It is worth knowing that the label has quietly stopped being true, and that in its place you may be holding one of the most concentrated bets in the market’s history.
S&P 500 Is a Highly Concentrated Position
In 2006, the five largest names were Exxon Mobil, Microsoft, Citigroup, General Electric, and Walmart — energy, software, banking, industry, and retail — so a shock to any one of them rarely moved the other four. Today’s top 10 opens with Nvidia, Apple, Alphabet, Microsoft, and Amazon, and six of the 10 seats belong to companies whose earnings are tied directly to AI chips, cloud infrastructure, or data center spending (Source: FinHacker, S&P Dow Jones Indices data, August 2026). Concentration on its own is not new to markets. Concentration built on one correlated thesis is.
A passive S&P 500 investor is no longer diversified across the American economy in the way that label implies. Increasingly, they are underwriting the success of a single technological outcome, with little ability to reduce that exposure without leaving the index itself. This is not pure hype: the top 10 now generate roughly a third of the index’s earnings, up from about 17% a decade ago (Source: GHP Investment Advisors). But they still trade at a materially higher multiple than the other 490 companies combined, so the price has run well ahead of even that improvement.
None of this is an argument against owning these companies. Several are among the best businesses in the world, and we own a number of them ourselves. What matters is how and why you come to own them.
The Capex Arms Race
The money committed to that thesis has moved from large to almost unimaginable. The four biggest spenders — Amazon, Microsoft, Alphabet, and Meta — together budgeted roughly $725 billion for 2026, a 77% jump in a single year, and analysts expect the group to top $1 trillion in 2027 (Source: company Q2 2026 earnings calls, compiled by ValueAddVC). Internal cash no longer covers it. Morgan Stanley estimates this spending is approaching 100% of these companies’ operating cash flow this year, against a historical norm closer to 40% (Source: Morgan Stanley Research, cited by Forbes, July 2026).
The rest is borrowed, and that is where it stops being someone else’s problem. AI-related debt issuance is on pace to reach roughly $570 billion in 2026, more than double the prior year (Source: Morgan Stanley Research). When a handful of borrowers compete for that much capital at once, every other issuer, corporate or municipal, prices its own debt against the same backdrop. The cost of insuring Oracle’s debt has already more than tripled since last September (Source: Axios, July 2026). Morgan Stanley separately sees a $1.5 trillion gap between what the buildout will need through 2028 and what these companies can fund on their own. We believe this bears watching for what it could mean for borrowing costs well beyond technology.
The Inbred Nature of the AI Trade
The last symptom is how self-referential the money has become. Nvidia has announced more than $540 billion in financing and investment arrangements this year with the very companies that buy its chips, including stakes in OpenAI, CoreWeave, and Nebius (Source: Bloomberg and Wall Street Journal, July 2026). Capital the chipmaker puts into its customers has a way of returning as chip revenue. The demand side is just as narrow: OpenAI alone accounted for close to 70% of Microsoft’s AI revenue in the year ended in June (Source: Bloomberg analysis of Microsoft filings, August 2026), and OpenAI has in turn signed roughly $1.15 trillion in infrastructure commitments through 2035, obligations that assume revenue it does not yet earn (Source: company disclosures, 2026).
In our assessment, this looks less like a broad, self-sustaining technology cycle and more like the telecom buildout of a generation ago, when Global Crossing, WorldCom, and Nortel laid fiber for internet traffic that arrived a decade later than their financing assumed. Hyperscaler spending as a share of the economy is already running higher than telecom ever reached (Source: Apollo Chief Economist, January 2026). The lesson of that era was not that vendor financing is unsound. It was that demand came far later than the money assumed, and by the time it did, several of the most aggressive borrowers had already failed. We do not know whether this cycle ends the same way.
We do know the payoff has not yet broadened. It has not shown up as earnings growth across the other 490 companies in the index, nor, for all that has been spent, as a clearly better life for the people funding it, directly or through their portfolios. Barely one in ten Americans say they feel more excited than concerned about AI’s growing role in daily life (Source: Pew Research Center, June 2025). None of that tells us how the bet resolves. It tells us how much is riding on it.
Next week, we turn from the risk to the response: how we build portfolios that can share in this opportunity without staking a client’s future on how it ends.