AI didn't lose its stranglehold on U.S. stocks. It just moved its grip.

Read any market commentary from the past year and you will meet the same relieved paragraph. The big AI names are finally dragging, not carrying. Equal-weight funds are beating the cap-weighted S&P 500 for the first time in years, small caps just posted their best first half on record, financials and healthcare are leading, and roughly two-thirds of index stocks are higher since June. The conclusion writes itself: AI is losing its stranglehold on the U.S. stock market.

Every one of those facts is true. That is exactly why the conclusion is worth distrusting. The market did not loosen its grip on AI. It changed which AI names hold the grip — and in the process concentrated something far more fragile than stock returns.

The broadening that looked like escape

Start with what actually happened in 2026, because the believers deserve a fair hearing. By mid-July the seven biggest tech platform stocks were up just 5.5% on a cap-weighted basis, on pace for their worst year since 2022, with Microsoft down more than 20%. Meanwhile the equal-weight S&P 500 climbed more than 13%, comfortably ahead of the standard index's 8.5%. Small-cap stocks returned 22.9% in the first half, their best such stretch on record, and all eleven sectors in the small-cap index finished positive. Healthcare rose 14% and financials 12% from June. This is not a mirage. A genuine rotation happened.

The trouble is what the rotation was made of. Count the drivers and "escape from AI" dissolves into "a new corner of AI." Semiconductors alone contributed 4.25 percentage points of the Morningstar large/mid index's 10.6-point gain this year — about 40% of the entire move (semiconductors drove 40% of the index's gain). The individual leaders were not old software platforms. They were Micron, up more than 200%; SanDisk, up over 600%; AMD, Intel, and the equipment makers Applied Materials and Lam Research, all up in triple digits. The reason is that "agentic" AI needs ordinary computing infrastructure — server CPUs and memory — on top of the GPUs everyone already bought. Micron has grown so fast it has become one of the largest companies in its index.

Diversified by count, concentrated by cause

Now the denominator move, because this is where breadth lies to you.

The celebrated statistic — how many stocks are rising — counts securities. A rising security count looks like diversification. But diversification is not how many tickers move together; it is how many independent futures are paying for them. And the futures funding this rally are not independent at all.

The four largest hyperscalers — Microsoft, Amazon, Alphabet, Meta — are on track to spend on the order of $770 billion on capital expenditures in 2026, roughly equal to 100% of their operating cash flow. They are financing the buildout by borrowing — net debt up about $170 billion since the start of 2025 — and by trimming buybacks and letting share counts drift higher. Memory shortages, exploding power demand, and the grid buildout that money pays for: that is the engine under the equal-weight rally, the small-cap record, the industrial strength, and even much of the utilities and energy trade. The small caps celebrated for "broadening" the market are, in large part, beneficiaries of the very same AI capex.

So the market is not diversified into the same future the way an index with 500 names should be. It is secretly one trade — call it "the AI buildout keeps growing" — spread across a supply chain and measured in hundreds of tickers. When people say AI is losing its stranglehold on the market, what they usually mean is that the mega-cap platform stocks no longer drive returns. That is true and it is not the same thing as the market depending less on AI.

Where the concentration actually lives now

Here is the uncomfortable half of the trade. A stock-return concentration (a few big platform names carrying the index) has been swapped for a spending-concentration (the whole market carried by a handful of capex budgets). One is visible in every chart. The other is invisible unless you ask who pays for the earnings.

The data confirms the switch. Goldman Sachs calculates the S&P 500's aggregate return on equity hit a record 22% in early 2026 — but only because the mega-cap tech weight drags the average up. The median stock in the index has seen its ROE fall in recent years, hit by higher interest expense and lower leverage. The seven largest tech names enjoy a collective 44% ROE, and consensus already implies that number drops by about seven percentage points in the next year as depreciation and equity flood in. The entire celebrated "broadening" has been running on an earnings engine that lives in a very small building.

What would break the new grip

The honest response to "AI is losing its stranglehold" is: prove it with the denominator, not the count. The broadening is only real if it survives its funding source slowing down. That is the testable catalyst.

Watch one number above all: hyperscaler capex growth, because every leg of this rotation sits on top of it. The failure condition is not a dip in Nvidia's price. It is a hyperscaler saying out loud that AI returns do not justify the next step up in spending, cutting the buybacks further, or letting the debt-funded buildout stall. When the four companies funding roughly the whole trade build less, the memory makers lose their pricing power, the equipment names lose their orders, and the small-cap and power names that look like "breadth" turn out to have been the capex forecast wearing a different hat. The narrow index the market escaped would look like the hedge it once provided.

None of this tells you AI is a bubble about to pop, or that Nvidia is doomed. Nvidia is still the largest company in the world, and its forward multiple remains demanding. The point is narrower and more useful. The market did not trade four stocks for five hundred independent ideas. It traded a few platforms for a few budgets. Breadth by security count is a headline; breadth by source of funds is the economics. The crowd that is celebrating the former has stopped looking at the latter — and it is the latter that decides whether the rotation is a step toward safety or a bigger step toward the same bet.