THE PRODUCTIVITY PARADOX

Vanguard's 2026 outlook carries a title that should stop you in your tracks: "AI exuberance: Economic upside, stock market downside." Read that again. The U.S. economy could grow at 3% on the back of AI investment while returns for growth stocks remain muted over the next five to ten years. The economy wins. The market doesn't. I've spent weeks sitting with this prediction and I still find it genuinely strange. We're not used to a world where the most transformative technology in a generation is bad news for the companies that built it.

But the logic, once you trace it, is almost elegant in its cruelty. AI accounted for 39% of GDP growth in 2025. That's not a typo. During the dotcom peak in 2000, the internet accounted for 28%. We've already blown past that benchmark, and most organizations haven't even finished their first real deployment. LPL research shows productivity running consistently above 2%, driving 2.5% growth into 2026. Ninety-six percent of organizations investing in AI report measurable productivity gains. The technology works. That's not the question anymore.

The question is who captures the value.

I keep returning to electricity. It took roughly thirty years for electrification to show up in national productivity statistics. When it finally did, the gains weren't captured by General Electric or Westinghouse. They were captured by the factories, the retailers, the logistics companies, the countless businesses that used electricity to do what they were already doing, only better and cheaper. The power companies became utilities. Essential, yes. Profitable enough. But not the locus of value creation. The interesting question is whether AI follows the same arc, and whether we're watching it happen in compressed time.

The current market structure suggests we should care about this deeply. Five companies hold up roughly 30% of the S&P 500. That's the greatest concentration in half a century. The entire investment thesis of the past three years has been built on the assumption that the companies building AI will be the companies that capture its economic value. Vanguard is suggesting the opposite. The more successful AI becomes, the more it diffuses. The more it diffuses, the more it becomes infrastructure. And infrastructure, historically, is a terrible place to look for outsized returns.

Goldman Sachs seems to have noticed. In February 2026 they launched an "AI-free stock index," essentially a bet on the companies that use AI rather than the companies that build it. It's a quiet admission that the value migration might already be underway. If you're constructing a portfolio around AI's economic impact, maybe you should be looking at the factories, not the power plants.

There's a parallel story unfolding in the labor market that complicates this further. The standard fear narrative goes like this: AI makes workers productive, companies need fewer workers, mass layoffs follow. But that's not what's happening. Only 17% of organizations are using AI productivity gains for headcount reduction. The vast majority are reinvesting — expanding output, entering new markets, doing more with the same people rather than the same with fewer. Kevin Warsh, Trump's nominee for the Federal Reserve, argues that the productivity surge gives room for rate cuts without stoking inflation. The economy grows not by replacing workers but by making each hour of work worth more.

This is the part I find hardest to reconcile. If AI productivity gains are being reinvested rather than extracted, that's genuinely good for the economy and for workers. But it makes the investment case for AI companies even harder, because it means the surplus is flowing to the users, not the builders. Every company that uses AI to expand its margins is a company that doesn't need to pay monopoly prices for the privilege.

Axios captured it well in a piece this week about the range of possible futures. The outcomes for unemployment and the stock market aren't correlated in the way we expect. You can have full employment and a flat market. You can have productivity growth and compressed margins for AI providers. The scenarios where both workers and AI stocks do well require something specific: sustained pricing power by a small number of companies over a technology that, by its nature, wants to become commodity infrastructure.

I don't know how this resolves. I genuinely don't. The historical pattern is clear. The economic data is clear. But markets have a way of defying historical patterns for longer than anyone expects, and the five companies at the center of this trade have more capital, more talent, and more data than any corporate entities in human history. Maybe they build moats that electricity companies never could. Maybe foundation models don't commoditize the way generating capacity did. Maybe.

But I keep coming back to that Vanguard headline. The most bullish possible outcome for AI as a technology — the one where it genuinely lifts the entire economy — is the one where the value disperses so widely that no single company captures it. We might be investing in the power companies when we should be investing in the factories. And the strange thing is, the better AI works, the more likely that becomes.