The Capability Premium.
A 2026 capability bifurcation in public-market benchmarks.

The 2023–2025 enterprise AI investment cycle deployed roughly $30–40 billion of corporate capital into pilots, tools, and transformation programs. The aggregate productivity result is well-documented: most of it produced no measurable impact. MIT Project NANDA finds that 95% of enterprise AI pilots fail to deliver measurable P&L impact.1 National Bureau of Economic Research analysis of 6,000 executives across the U.S., U.K., Germany, and Australia confirms the vast majority report little impact from AI on operations.2 Through 2024, Danish administrative data covering 11 AI-exposed occupations finds essentially zero aggregate effects on earnings or hours worked, despite widespread worker-reported adoption.3

Aggregate AI productivity has not arrived — yet. But the dispersion underneath the aggregate has. Morgan Stanley Research's 2026 outlook documents that the 21% of S&P 500 companies citing measurable AI benefits are seeing cash-flow margin expansion at roughly 2× the global average.4 BlackRock's 2026 outlook frames the same phenomenon from the allocator's perspective: "the productivity math is powerful, and potentially margin-expanding in a way that compounds. But it is unlikely to be evenly distributed. Companies that can translate AI into durable cost advantages and cash-flow resilience should separate from those that can't."5

The market is now pricing this dispersion explicitly. Q1 2026 PitchBook data documents the median 21.2× EV/Revenue for AI-native software versus 5.5× for legacy SaaS — a 285% premium that widened, not narrowed, through the 2024–2026 repricing.6 Within software specifically, AI vertical conglomerates fell 37% in Q1 2026 even as AI semiconductors held flat at −2%, in "a market that has stopped rewarding AI as a theme and started demanding proof of monetization."6

For the investor — PE managing partner, growth-equity sector lead, public-market analyst, family-office CIO — this poses a sharp question. Which firms in any given sector are emerging from the AI implementation J-curve with capability that compounds, and which are not? The price action says the market is sorting them, but the price action is a lagging indicator. Multiples expand or compress only after fundamentals do; the investor's edge is in identifying the firms ex-ante — before the multiple has fully repriced.

Three patterns emerge across the benchmark record. First, the dispersion is intra-sector, not just cross-sector. April 2026 multiples data documents that within horizontal SaaS — ostensibly a single category — design and engineering software trades at premiums while sales automation sits well below average, "as gen AI threatens to fundamentally replace traditional CRM workflows rather than augment them."7 Within infrastructure SaaS, data infrastructure commands the highest multiples while cloud infrastructure trades at a notable discount despite identical underlying technology stacks. Investors evaluating sector exposure have to evaluate capability exposure within the sector, not the sector itself.

Second, the market is rewarding evidence rather than narrative. AI vertical conglomerates fell 37% in Q1 2026 even as the broader S&P 500 fell 4.6%, in "a fundamental repricing phase".6 EV/revenue multiples for vertical software conglomerates compressed to roughly 4.5× despite continued 10–21% revenue growth — growth without margin trajectory ceased to clear the underwriting bar. The market is now paying for repeatable monetization and discounting opacity.8

Third, the dispersion mechanism is rooted in the productivity J-curve framework. Brynjolfsson, Rock, and Syverson9 establish that productivity falls before it rises because complementary intangible investments — in process redesign, governance, and human capital — take years to compound. The 2026 benchmark dispersion is the empirical residue of a structural bifurcation: firms that owned the learning are emerging with margin expansion; firms that outsourced it remain in pilot purgatory and trade accordingly.

Two interpretive notes are essential. First, the quartile assignment is sector-relative, not absolute. A Q1 firm in a slow-moving regulated industry exhibits the signal pattern at lower amplitude than a Q3 firm in a fast-moving consumer technology category — absolute margin expansion is not directly comparable across industries with different capital structures. Comparability discipline applies: peers in the same size class, same geography, same business mix, with definitions reconciled across the peer set.

Second, the framework is predictive, not retrospective. The signals are visible in disclosures that lead the financial statements by 12–24 months. A Q1 classification today is a hypothesis about emergence over the next 12–24 months, falsifiable by subsequent margin trajectory. The investor's edge is precisely in the lead time — by the time the financial statements have confirmed the bifurcation, the multiple has repriced. The signal set exists to identify the bifurcation before the multiple has caught up.