Every Exit From AI Risk Now Leads Back Into AI Risk. Here's Why.

The market has quietly redefined what diversification means, and most allocators have not noticed the redefinition happened. For a decade, the answer to too much equity risk was simple: rotate into private equity, private credit, real estate, infrastructure. Different cash flows, different drivers, different risk. That answer no longer holds, and the reason is not a flaw in any single fund. It is a flaw in the map everyone is still using.

The facts are straightforward. Research from Deutsche Bank and Goldman Sachs puts AI-exposed mega-caps at approximately 40-45% of S&P 500 market cap as of April 2026, with Nvidia alone above 8% index weight. Separate work from Purpose Investments arrives at a similar figure once AI-adjacent revenue outside pure technology names is included. At the $327 billion New York City Retirement Systems, chief investment officer Monte Tarbox turned down a private equity allocation because the product being pitched was heavy on artificial intelligence holdings, and he is becoming cautious about how exposed the fund gets to the boom. That single decision is the whole story in miniature.

The mechanism: diversification collapsing into a single trade

Here is how this actually transmits. A pension fund's equity book is loaded with the same handful of AI-linked names that now dominate every passive index. Its private equity sleeve holds growth-stage companies whose valuations are anchored to AI adoption assumptions. Its private credit book is increasingly financing the data centers, power infrastructure and GPU leasing arrangements that sit underneath the entire capex cycle. Its real assets allocation owns the warehouses and land parcels being repurposed for compute. None of these were supposed to correlate. All of them now share one underlying driver: whether the AI capex cycle continues at its current pace.

This is not correlation in the statistical sense. It is common exposure dressed up as diversification. The mechanism that used to break a shock into pieces, spreading it across uncorrelated return streams, is instead concentrating it. When one leg of the AI financing chain wobbles, whether that is a hyperscaler pulling back on orders, a chipmaker guiding down, or a lender tightening terms on a data center loan, every sleeve of the portfolio that was supposed to absorb the shock moves in the same direction at the same time.

The second-order effect nobody is pricing

The part almost nobody is discussing is governance, not markets. Investment committees still report AI exposure by asset class, showing a public equities line, a private equity line, a credit line, each within its approved risk budget. What they are not measuring is the aggregate, look-through exposure to a single thematic bet that now cuts across every one of those buckets. A fund can be fully compliant with every internal diversification rule and still be running a concentrated, undisclosed AI position that no single risk report captures. The risk is not that allocators are reckless. It is that the accounting framework they use to define risk was built for a world where asset classes actually behaved independently, and that world is gone for this particular theme.

This is exactly the kind of dislocation a delta-neutral volatility book is built to sit inside of. The position is not a directional call on whether AI capex continues or stalls. It is a bet that the market is mispricing the correlation between these assets, not their level, and that the repricing of that correlation, whenever it comes, will move faster than most portfolios are structured to absorb.

The honest counter-case

The counter-case deserves real weight, not a token line. It is entirely possible this is not concentration risk at all but the correct pricing of a genuine platform shift. If AI-driven productivity gains show up broadly across earnings over the next several years, and if the capex being financed today generates durable cash flows rather than a rerun of dot-com-era fiber overbuild, then today's concentration is simply markets correctly identifying where value is being created. Index concentration around the dominant technology of an era is not new. Railroads, oil majors and industrial conglomerates all had their turn. The mistake would be assuming concentration itself is the tell, when history shows it can persist for a long time before, or instead of, correcting.

What Komey Tetteh is watching next is whether allocators start reporting look-through AI exposure as its own line item, the way they eventually did with subprime exposure after 2008. Until that measurement exists, funds are managing a risk they cannot fully see. If your own portfolio spans equities, private markets and credit, have you actually checked how many of those sleeves are quietly making the same bet.