Apple and Nvidia Now Produce 65% of the Magnificent Seven's Free Cash Flow — and That Is Why the Group Stopped Trading as One

Correlations collapse when a group stops sharing a single driver. The dispersion inside the Magnificent Seven shows exactly that: the best and worst performers are 37.6 percentage points apart in 2026, and the group's own ETF trails the index it is supposed to lead by five points. A single-factor basket with that kind of lag no longer moves as one trade.

The group is also losing its grip on the index it dominates. The seven were about 35% of the S&P 500 at Sept. 30, 2026, yet the other 493 companies rose roughly 12.3% in 2026 through Oct. 23 while the seven rose about 9.7% on a weighted basis — and the equal-weighted S&P 500 beat the cap-weighted index, 13.0% against 11.4%. Only three of the seven were still among the top ten contributors to index performance as of July 10. Index concentration has split into self-funding cash generators and externally funded builders, and the two cohorts now pull against each other.

The popular explanation is that the group has divided into AI funders and AI monetizers. The filings say something sharper and less comfortable. Rank the seven by how much capital they spend to earn a dollar of revenue, and that order almost perfectly predicts the order of their 2026 returns. Capital intensity predicts the 2026 ranking better than business model does.

Rank-correlating the two columns above gives roughly minus 0.8 for the six names that actually book AI revenue today, and roughly minus 0.5 across all seven. Tesla is the outlier that proves the rule: it spends only 14.9% of revenue on capital, but none of that spending is attached to booked AI revenue, and it is the worst performer in the group. Spend without revenue gets punished, and so does revenue without a funding plan. The second point is the one investors keep missing: Apple spent $10.0 billion of capital in twelve months and produced $136.7 billion of free cash flow — almost exactly what Microsoft, Alphabet, Amazon and Meta were left with after $525.8 billion of capex.

This is where the correlation break becomes a funding story. The four companies running the largest AI buildouts spent $525.8 billion over the trailing twelve months — 79% of the $663.1 billion of operating cash flow they generated between them. The gap is now being closed in the capital markets, and each company has chosen a different currency.

Balance sheets do not de-correlate on their own. Something has to change in what investors believe, and on Oct. 8, 2026 it did. OpenAI told investors its annualised revenue had reached roughly $50 billion at the end of September — about $20 billion below the figure that had been circulating. Nvidia, Oracle and CoreWeave fell together that day. The market began pricing AI revenue claims to the last verifiable number.

Put the three disclosures together and the mechanism is clear. Demand is real and contracted. What is no longer reliable is the assumption that contracted demand converts into cash for the buyer of the compute — and, for the seller of it, the assumption that the buyer can keep paying. When a trade rests on one shared assumption and that assumption cracks, the names that look most dependent on it should fall hardest and least together. That is precisely a correlation break. The group is now priced on counterparty risk rather than on AI exposure, and counterparty risk is not a common factor — it is a set of bilateral ones.

The transmission is mechanical. Every dollar a hyperscaler spends on chips, memory, power and cooling becomes revenue for a supplier that mostly does not have to fund the customer. That is why TSMC grew revenue 41% year over year in the September quarter, why Vertiv could raise full-year guidance to $14.65 billion on Oct. 28 with organic orders up more than 100%, and why Micron turned $37.4 billion of revenue into $133.2 billion in a single fiscal year. The upstream gets paid for the buildout before anyone has to ask whether the buildout pays for itself.

For passive holders this changes what concentration risk actually is. The seven are still roughly 35% of the S&P 500, but they no longer carry one exposure. They carry two that have turned partly against each other: a self-funding group that keeps raising free cash flow, and a funded group whose earnings quality now depends on the credit markets staying open. Index concentration has become a barbell between two different balance sheets, not a single bet on one AI trade.

Here is the thesis. The Magnificent Seven has not split into funders and monetizers; it has split into companies that can pay for the AI buildout out of operating cash flow and companies that must pay for it in the capital market. Apple and Nvidia sit on one side — 2.2% and 2.4% of revenue in capital spending, $263.7 billion of combined trailing free cash flow and 41% of the group's market value. Meta, Amazon and Alphabet sit on the other, having committed $410 billion of capital and added roughly $197 billion of net debt between them.

The correlation break is the symptom; the financing regime is the disease. That is why the group's ETF can trail the S&P 500 by five points while Apple beats the index by almost fifteen, and why the next 48 hours of earnings — Alphabet and Microsoft on Oct. 28, Amazon on Oct. 29 — matter more for relative positioning than three years of AI narrative did. The group that no longer moves together still, very largely, holds one macro risk: the availability of capital.