Why 2026 is the first year AI stocks decoupled

For three years, the critique of AI-stock valuations was easy to dismiss because the earnings kept showing up. Revenue grew. Margins grew. Guidance kept rising. Every quarter the bears pointed at the multiple, and every quarter the fundamentals caught up enough to justify another leg.

2026 is the first year that stopped.

What changed — and this is the conversation institutional allocators have been having privately since late 2025 — is that the share of S&P 500 forward earnings attributable to AI-exposed names reached the point where the valuation math no longer works on consensus assumptions.

The headline number is that a single sector now produces 42% of the forward earnings of the largest US index. The Magnificent Seven plus their immediate AI-infrastructure satellites — Nvidia, Microsoft, Alphabet, Meta, Amazon, Apple, Tesla, plus ARM, Broadcom, AMD, TSMC ADR, and the hyperscaler power-and-cooling supply chain — now earn more than financials, healthcare, and consumer staples combined.

This concentration is not, by itself, the reason valuations cracked. The specific reason is subtler. Through 2024 and most of 2025, AI capex guidance kept rising at the hyperscalers. Every quarter, the big four (MSFT, GOOG, META, AMZN) told the market they were spending more on AI infrastructure than the previous quarter — and the market rewarded them for it, because the implicit story was that revenue would follow.

In Q4 2025, the story changed. Capex guidance for 2026 came in roughly flat. Two of the four actually guided slightly lower. The market read that — correctly — as the first admission that the second derivative of AI infrastructure demand was no longer positive. The spending will continue. It will not keep accelerating.

The valuation math, once you remove the acceleration assumption, is unforgiving. At a 27x forward P/E on the Magnificent Seven cluster, you needed 15% per annum earnings growth to justify the multiple at a normal equity risk premium. Consensus 2026 earnings growth for the cluster is now closer to 10%.

That is the decoupling. Not a crash. A re-rating. Prices rose roughly 3% YTD in 2026 against index earnings that grew approximately 9%. Multiples are compressing quietly while the indexes tread water.

Three implications for the allocator in 2026.

First, the equal-weighted S&P 500 is going to outperform the cap-weighted S&P 500 this year for the first time since 2022, and the gap will be noticeable. If you are indexing, check which index.

Second, the "Mag 7" as a trade is over. That does not mean individual names are wrong. It means the concentrated bet on the cluster behaving as one asset class is no longer a cheap way to own growth.

Third — and this is the one most investors are missing — the broadening of earnings into industrials, energy, and select healthcare is real and under-priced. The companies that sell the picks and shovels of AI physical build-out (power infrastructure, cooling, copper, grid) are trading at sub-15x multiples with a multi-year tailwind that has not yet showed up in their price.

The AI story is not over. Its cheap-multiples phase is. Allocators who understand the difference between those two statements will look smart in two years.

— The Editor's Bureau