AI Stocks: P/E Collapsed, Data-Usage Bubble Grows

Lead

The price-earnings multiple for a basket of AI-exposed equities has collapsed materially from its 2024–25 zenith even as a distinct usage-driven metric—AI compute and data consumption—continues to expand, according to Capital Economics analyst John Higgins quoted in Fortune (Mar 29, 2026). Higgins describes the divergence as a “rare” kind of bubble in which valuation multiples have corrected but the real-economy proxy for adoption has not, introducing a complex risk-reward calculation for institutional investors. The P/E contraction reported by Capital Economics is estimated at roughly 45% from peak levels, while compute demand for large-scale model training and inference reportedly rose approximately 120% year-over-year in 2025 (Capital Economics via Fortune, Mar 29, 2026). These dual signals—cheapening multiples on one hand and accelerating usage on the other—are reshaping capital-allocation debates across technology, cloud infrastructure, and semiconductor sectors.

For portfolio managers and allocators focused on technology, the immediate question is whether falling multiples reflect a permanent reassessment of future cash flows or a near-term repricing driven by macro and liquidity dynamics. Benchmark comparisons are already stark: the aggregate AI basket's trailing P/E has moved well below the broader market's forward P/E, which stood near 18x as of late March 2026 (Bloomberg, Mar 27, 2026). Yet cloud-capacity indicators and vendor order-books point to continuing secular demand: firms operating hyperscale data centers reported capacity utilization metrics and GPU order growth consistent with double-digit to triple-digit annual increases in AI-related compute workloads in 2025 (company filings and industry reports, 2025–2026). This article dissects the metrics, contrasts valuation and usage trends, and assesses implications for equity and infrastructure exposures.

Context

The recent sell-off in AI-focused equities followed a concentrated rally between mid-2023 and mid-2025 that pushed forward-looking multiples to historically elevated levels. Capital Economics' read of the market assigns much of the multiple compression to increasing rates volatility and a pullback in investor expectations around near-term margin expansion for software incumbents. That correction reduced the headline P/E for the basket by roughly 45% from its peak (Capital Economics via Fortune, Mar 29, 2026), bringing it closer to long-term average multiples for large-cap technology names.

Historically, technology waves have featured similar patterns: hype-driven premium indices reprice once consensus growth expectations are questioned, even as underlying adoption can persist. The dot-com cycle in 2000 is the archetype: multiples evaporated while usage of the internet continued to increase; true monetization and a multiyear recovery followed only for select survivors. The current episode differs because compute capacity and data ingestion—measurable inputs to AI deployment—are rising at rates that previous software revolutions did not require at scale. Capital Economics flags this distinction and labels the phenomenon 'rare' because the user base and infrastructure demand can sustain value capture outside of headline multiples for a period.

Macro conditions also matter. The 2024–26 tightening cycle and then partial easing altered discount rates and liquidity provisioning for growth assets. As of March 2026, real yields and cost-of-capital dynamics remain an important backdrop for how investors price long-dated AI cash flows. In short, the context is one of corrected valuations against an accelerating demand curve—forcing investors to adjudicate between valuation reset permanence and structural revenue tailwinds.

Data Deep Dive

Capital Economics provides three headline data points shaping its thesis: a c.45% collapse in aggregate P/E from the 2024–25 peak (reported Mar 29, 2026), an estimated c.120% year-over-year increase in AI compute demand in 2025 (Capital Economics estimate, Mar 2026), and rising data-center capital expenditure growth rates reported by major cloud providers in FY2025 (company filings, 2025). The 45% P/E decline is a valuation metric that captures market sentiment; the 120% compute demand increase is an operational metric that captures real activity. The divergence creates a cross-asset dispersion where equity prices may no longer reflect immediate infrastructure spend trends.

To ground those macro numbers, consider two industry-level comparisons. First, the AI compute demand increase (c.120% YoY in 2025) contrasts with a typical pre-AI cloud growth rate of mid-teens annual percentage points—this is a multiple rather than an incremental change versus historical norms. Second, the AI-basket P/E now trades at a discount vis-à-vis the broader market's forward P/E (~18x, Bloomberg, Mar 27, 2026), suggesting either heightened risk premium or a reassessment of revenue durability. These comparisons paint a picture of structurally higher resource intensity per unit of AI revenue relative to prior software cycles.

Data from vendors and hardware suppliers corroborate the compute narrative. GPU and accelerator demand—an observable proxy for large-model training and inference—showed sustained order-book strength through 4Q 2025 and into early 2026 (company earnings releases, 4Q FY2025–1Q FY2026). Channel checks reported by multiple industry analysts indicate multi-quarter backlog for high-end accelerators. Those operational data points support the idea of an underlying usage bubble even as market multiples have retrenched.

Sector Implications

The valuation-usage bifurcation has differentiated implications across sectors tied to AI. For semiconductor suppliers and data-center infrastructure firms, rising compute demand implies multi-year top-line visibility and capex reallocation that could underpin earnings upgrades. If compute demand continues at double-digit or higher growth—Capital Economics' ~120% YoY estimate—these suppliers can convert utilization into pricing power and margin expansion, provided supply constraints and competitive dynamics are favorable.

By contrast, software providers and public AI application companies face a different stress test. They must demonstrate unit economics at scale while contending with slower revenue recognition cycles and heightened price competition. The market’s downward re-rating has been more pronounced in the application layer, where multiples contracted faster than for hardware or cloud platform suppliers. Relative performance vs. peers is instructive: hardware firms have outperformed software peers on a sector-relative basis since mid-2025, implying investors are repricing exposure toward asset-light backbones and away from high margin promises that may be tougher to realize.

For large-cap index investors and passive funds, the divergence raises tracking and composition questions. Indices heavy on AI application names have experienced greater volatility and deeper drawdowns than indices dominated by infrastructure constituents. Institutional clients with mandate constraints are increasingly examining exposure not only to 'AI' as a label but to subcomponents—chipmakers, cloud operators, systems integrators—each with distinct cash-flow profiles and balance-sheet resilience.

Risk Assessment

The primary risk is a classic mismatch between expectations and realizations. If compute demand decelerates from the elevated 2025 pace—due to macro weakness, regulatory clampdowns, or faster-than-expected improvements in model efficiency—then firms that have scaled capacity on the bet of sustained demand could face inventory and utilization write-downs. Conversely, if compute demand continues to outstrip supply, price inflation for accelerators and cloud services could compress gross margins downstream and reignite valuation rerating for infrastructure names.

Secondary risks include regulatory and antitrust pressures that could materially change market structure. Government interventions in data access, cross-border model deployment, or export controls on advanced accelerators would disproportionately affect the revenue forecasts baked into current valuations. Market liquidity and investor sentiment cycles are also salient: a continued de-risking by large passive flows could amplify price dislocations in the short run even if fundamentals remain intact.

Finally, technology advancement risk—improvements in model efficiency, software stack optimization, or new accelerator architectures—could lower compute per unit of output, undermining the 'usage bubble' thesis if such gains accumulate rapidly. This technical deflation would reduce the infrastructural scale required, shifting value capture upstream toward software and services rather than hardware or raw compute providers.

Fazen Capital Perspective

Fazen Capital views the current configuration as a potential strategic opportunity for differentiated active allocations rather than a binary bubble-or-no-bubble call. The coexistence of compressed multiples and rising compute demand implies higher dispersion in future returns across sub-sectors. In practice, this favors investment frameworks that separate exposure to (1) monetizable software cash flows, (2) capital-intensive infrastructure and hardware, and (3) idiosyncratic platforms that control data and go-to-market distribution. We recommend scenario analyses that stress test capex, margin, and adoption timelines across these buckets. Our contrarian insight is that a continued decoupling—valuations down, usage up—can persist for multiple quarters, rewarding patient, research-intensive active strategies that can identify firms able to convert usage growth into sustainable monetization.

For institutional investors, the tactical implication is to prioritize exposure to balance-sheet-resilient suppliers and diversified cloud providers while demanding clearer go-to-market economics from application-layer companies. Fazen Capital also emphasizes the importance of operational due diligence: inspection of customer concentration, contract structures, and gross-margin dynamics is necessary to separate transient winners from structurally advantaged firms. See our related sector work on AI infrastructure and software monetization on the Fazen insights hub: Fazen insights and a recent thematic piece on sector rotations: sector analysis.

FAQ

Q: Could the current pattern mirror the dot-com cycle where valuations collapsed but adoption continued? How is this different?

A: Yes, there are parallels—both episodes show valuation compression ahead of adoption normalization. The key difference is resource intensity: AI requires materially more compute and storage per unit of revenue than early internet applications did, making hardware and infrastructure a larger share of the economic value chain. That amplifies the intermediate-term impact on capital spending and vendor revenues even if equity multiples are correcting.

Q: What leading indicators should investors monitor to time exposure to AI infrastructure or equities?

A: Monitor vendor order backlogs for accelerators, cloud providers' capacity-utilization rates, capex guidance in quarterly filings, and pricing trends for spot GPU rentals. On the demand side, measure enterprise adoption via contract wins disclosed in filings and developer activity metrics on major model platforms. These indicators offer earlier signals than headline equity multiples and can help distinguish transient sentiment-driven corrections from fundamental demand shifts.

Bottom Line

Valuations in AI equities have retraced significantly—Capital Economics estimates a c.45% P/E drop from the 2024–25 peak—while operational metrics such as compute demand remain elevated (~120% YoY in 2025), creating a multifaceted investment landscape where active, granular assessment of sub-sector economics is essential. Institutional investors should prioritize differentiation between monetizable software franchises and capital-intensive infrastructure suppliers when constructing exposure.

Disclaimer: This article is for informational purposes only and does not constitute investment advice.