The great rotation: AI, deadweight loss, and the end of easy compounding

For more than a decade, equity markets have been defined by concentration. A small group of asset‑light, cash‑generative technology companies came to dominate returns, benchmarks, and capital flows. By 2024, the top 10 stocks accounted for roughly a quarter of the global index. That dominance was no accident: It was driven by a combination of ultra‑supportive macro conditions and exceptional, self‑sustaining growth in revenue, earnings, and cash flow—even as rates rose.

It hugely benefited passive investment strategies because index funds concentrated capital in the same dominant winners driving returns, while active managers—typically more diversified and valuation‑sensitive—struggled to keep pace. Passive assets overtook active over this period, reaching a majority share of U.S. equity fund assets.

Artificial intelligence (AI) is disrupting that equilibrium. The leading platforms are being drawn into a capital‑intensive investment race as a major technological shift leaves incumbents little choice but to invest to maintain their position. In 2025 alone, the largest platforms raised over USD 100 billion in debt to fund that spending. At the same time, the constraints on AI deployment are increasingly physical rather than digital, redirecting value into supply chains, infrastructure, and industrial capacity.

In other words, the conditions that sustained mega‑cap dominance are reversing. Replacing them is a broader, more diffuse pattern of value creation—one that markets, still anchored to the past, are only beginning to recognize.

From scarcity to concentration

The 2008 to 2009 financial crisis ushered in an economic regime that persisted for 15 years. Interest rates were held near zero, real yields were often negative, and policy was explicitly designed to compress risk premia. In theory, this was meant to encourage productive investment; in practice, it drove capital into financial assets, and above all into companies capable of generating reliable growth without needing much capital to do so.

The result was a steady concentration of both returns and investor attention. Mega‑cap technology platforms—Google, Apple, Microsoft, Amazon, and Meta—became the defining compounders of the era. The technology sector swelled from 11% of the MSCI All‑Country World Index in the early 2010s to over 26% by the end of 2025; the U.S. weighting in the index rose from just over 50% to 63% over the same period.

Over time, this dynamic extended beyond a handful of stocks to an entire style of investing as investors piled into “quality” and “compounder” businesses. In a world of persistently low nominal gross domestic product (GDP) growth, companies that could self‑fund above‑average growth commanded an enormous scarcity premium. This bid up valuations across the quality spectrum—not just the mega‑caps, but also the broader cohort of high‑return, high‑margin, asset‑light businesses that populated growth and quality portfolios globally.

As returns concentrated in a handful of dominant stocks, so did the structure of the market. By 2024, the largest names accounted for an unprecedented share of global indices, with a small group driving a disproportionate share of performance. This proved devastating for active management: most managers underperformed, capital flowed steadily into passive strategies, and the cycle reinforced itself as index funds allocated ever more to the same winners. What began as a narrative hardened into orthodoxy.

Yet this dynamic masked an important distinction. As Cremers and Petajisto showed, highly‑differentiated managers were not so much disproven as crowded out by concentration, while benchmark‑hugging strategies thrived by default.1 That logic now looks set to run in reverse. Assuming concentration declines, proximity to the index becomes a liability rather than a safeguard, as it leaves passive strategies and low‑active share managers structurally exposed to a narrow set of companies just as the opportunity set begins to broaden. Today’s benchmarks remain heavily concentrated in U.S. technology. Much of this change can be traced to the economics of AI.

The cost of competing in AI

The emergence of generative AI locked hyperscalers into a near–Nash equilibrium in which each company had to invest heavily, as cutting capital expenditure (capex) unilaterally would risk a significant—and potentially permanent—loss of AI market share. Practically every CEO described AI as the largest transformation since cloud computing, and perhaps since the internet. Nobody could afford to stop. The financial consequences of this were enormous and escalating.

That spending is already reshaping the financial profile of the sector. Capital expenditure among the leading hyperscalers is rising at an extraordinary pace, absorbing a growing share of cash flow and, in some cases, pushing free cash flow (FCF) generation toward zero or below. Balance sheets that once required little external funding are beginning to rely on debt markets: the big five raised USD 108 billion in bonds in 2025, with USD 1.5 trillion in projected debt issuance ahead. Alphabet alone has seen its long‑term debt quadruple to USD 46.5 billion by the end of 2025.

This is a profound shift—but not necessarily a terminal one. AI is, at its core, a destroyer and redistributor of deadweight loss: It eliminates inefficiencies across knowledge work, services, and distribution that previously sustained margins elsewhere. The hyperscalers are compelled to invest precisely because the prize—capturing and reconcentrating those rents at platform scale—is existential. They may ultimately succeed. But the transition itself is unavoidable: balance sheets must absorb leverage, free cash flow must compress, and the frictionless compounding that benchmarks still price in gives way to a capital‑intensive contest with uncertain timing.

When the constraint becomes physical

If AI were purely a software story, this transition might still favor the incumbents. But the binding constraints are no longer digital—they are physical.

The surge in demand for AI compute is running ahead of the infrastructure needed to support it. Bottlenecks are emerging across the supply chain, most notably in power. Amazon doubled its power footprint in 2025 and still could not meet demand, while Goldman Sachs estimates a U.S. data center power shortfall that could reach 40GW by 2028. Grid connections, transformers, and permitting timelines now determine how quickly AI can scale.

This shift changes where value accrues. The expected winners are not the hyperscalers deploying capital, but the companies supplying the inputs—memory, power, and infrastructure. It is the inverse of the post‑global financial crisis (GFC) era, when asset‑light platforms captured the gains and suppliers were left behind.

Such shifts tend to be gradual and then sudden. For a time, markets continue to reward the familiar winners even as the underlying drivers of value move elsewhere. Eventually, the mismatch becomes too large to ignore.

A broader economic expansion

At the same time, the effects of AI investment are propagating beyond the technology sector. Between 2010 and 2024, the U.S. economy was defined by an unusual asymmetry: financial asset prices compounded while real fixed investment stagnated. Capital flowed into buybacks and digital platforms rather than factories and power plants. That dynamic is now reversing, driven by three converging forces.

AI infrastructure is emerging as an industrial catalyst of historic scale. In early 2025, technology investment contributed more to U.S. GDP growth than consumer spending—something not seen since earlier waves of technological transformation.

Its impact extends far beyond data centers. Building AI capacity requires power systems, materials, and construction at scale, pulling demand into sectors largely bypassed in the digital era. The USD 364 billion of hyperscaler capex in 2025 alone is estimated to support more than 600,000 jobs, with multiplier effects spreading across manufacturing, utilities, and logistics.

The manufacturing renaissance reinforces this broadening. Construction spending on manufacturing has tripled from USD 76 billion in 2021 to USD 230 billion in 2025, rising from 6% to 14% of total private construction.2 The CHIPS Act has catalyzed roughly USD 630 billion in semiconductor investment, while a further USD 1.2 trillion in U.S. manufacturing—spanning chips, electronics, pharmaceuticals, and industrial infrastructure—was announced in just eight months of 2025.

This surge is feeding through the real economy. Industrial power demand is rising sharply—one Taiwan Semiconductor Manufacturing Company semiconductor fabrication plant in Arizona requires 2.85 GWh per day alone—and U.S. electricity demand could grow by as much as 50% by 2040.3 Fiscal policy is adding an unusually strong tailwind, with deficits above 6% of GDP alongside rate cuts and targeted manufacturing incentives. As Bank of America notes, the U.S. is shifting from globalization toward a more regionalized, reliability‑driven model.

Brynjolfsson, Rock, and Syverson’s “productivity J‑curve” provides the bridge between today’s investment and tomorrow’s growth. General‑purpose technologies require substantial upfront investment before productivity gains appear, and early evidence suggests AI is following this pattern.

If this holds, today’s wave of physical investment is the start of a broader productivity expansion. As growth spreads, the scarcity premium that sustained the quality and compounder cohort for 15 years erodes. More sectors participate, more companies grow, and the advantage of a narrow group of self‑funding compounders diminishes. The likely beneficiaries sit at minimal weights in today’s benchmarks.

Pressure on the old winners

If the hyperscalers are being forced to spend to capture new rents, the more vulnerable cohort sits one layer down: asset‑light software and services firms whose moats were built on human expertise—and whose margins represent exactly the deadweight loss AI is designed to eliminate.

Advances in AI are lowering the cost of performing a wide range of knowledge‑intensive tasks, from coding to customer support—suggesting that many knowledge‑intensive services can be delivered with dramatically fewer people. If large language models replicate significant portions of legal analysis, code generation, financial reporting, and marketing content, competitive moats erode rapidly. At the same time, hyperscalers desperate to monetize capex are pushing aggressively into enterprise software with distribution and AI capabilities incumbents cannot match.

This pincer—commoditization of knowledge work from below, platform competition from above—puts structural pressure on the sector that drove the quality factor’s outperformance for a generation.

The return of dispersion

Together, these forces—declining hyperscaler returns on capital, supply chain bottlenecks redirecting value, a manufacturing renaissance, AI disrupting asset‑light incumbents, and fiscal policy supporting real investment—set the stage for the most significant equity market regime change since the GFC.

Passive strategies are maximally exposed to companies undergoing potentially the greatest fundamental transition: 63% U.S., over 26% information technology,4 with outsized concentration in names shifting from compounders to capital‑intensive builders with uncertain return timelines. If value creation shifts toward utilities, industrials, materials, and infrastructure, the benchmark becomes a structural disadvantage. The winners of a physical investment boom—power equipment manufacturers, memory producers, construction firms, and industrial suppliers—sit at minimal weights precisely because they remain small relative to the mega‑cap platforms dominating the index.

Wilmington Trust has shown that active managers performed better during periods of higher dispersion and lower concentration—both dynamics likely to strengthen as AI reshapes value creation. Cremers identified three pillars of active success: skill, conviction, and opportunity. Capital intensity becomes a feature, not a bug, and international diversification regains relevance. The most differentiated managers—those punished most during the concentration era—should be best positioned for what follows.

Risks and uncertainties

This argument is not without vulnerabilities. Hyperscalers may yet capture the deadweight loss they are redistributing—if their platforms reconcentrate value at scale, today’s capex will look prescient and the rotation more cyclical than structural. The manufacturing renaissance may also fall short: construction activity has lagged announcements, constrained by labor shortages and rising costs. At the same time, large fiscal deficits risk triggering a bond market response that could choke the investment cycle.

AI’s productivity impact may prove more modest than expected, while some of the investment leaks abroad through imports, reducing the domestic multiplier. There are also historical reasons for caution: the infrastructure build‑out of the late 1990s preceded a sharp correction before delivering longer‑term gains.

Regulatory backlash could further slow the cycle while entrenching incumbents, as compliance costs favor scale. And, as ever, timing remains uncertain—similar arguments could have been made in 2018 or 2022 and would have been premature. Regime changes are only clear in retrospect.

The cycle turns

While these risks are real, they do not alter the underlying shift now underway. The post‑GFC era created a self‑reinforcing cycle of extraordinary durability: low rates favored quality growth, mega‑cap tech delivered it, benchmarks concentrated, passive captured flows, concentration deepened further. Active managers underperformed, capital shifted to passive, passive flows bought the largest stocks, and active managers underperformed again. It survived a trade war, a pandemic, and the most aggressive tightening in four decades.

The AI mega‑cycle may be the force that breaks it—by transforming platforms into capital‑intensive builders, by redirecting value to a broad physical supply chain, by catalyzing a manufacturing renaissance, and by threatening the asset‑light models that defined the quality factor.

For investors, the message is not that technology is finished but that the locus of value creation may be shifting in ways passive strategies cannot capture. The expected beneficiaries of the next decade sit at negligible weights in today’s benchmarks. The irony is exquisite: the most technology‑centric disruption in decades—one whose ultimate winners may include the very platforms now forced to transform—may revive the fundamentally human discipline of active investment management, just as everyone stopped believing in it.