AI-Led Momentum in a Higher-Rate World: Stay Exposed, Get More Selective

Core Conclusion

The sharp bond selloff in mid-May 2026 triggered a rapid rotation that hit Momentum harder than any other style factor, but the episode is better characterised as a selective unwind of crowded AI-led winners, not a clean risk-off regime break or a durable Value-over-Growth rotation. Resilient AI earnings, contained volatility (VIX below 20), and non-excessive factor valuations (12M Momentum valuation at the 49th percentile of its historical range) argue against a prolonged Momentum correction. Investors should maintain Momentum exposure but shift from traditional 12-month Price Momentum to Earnings Window Momentum (EWM), which showed a shallower drawdown (~2% vs ~8% in the US) and a tighter link to stock-specific earnings signals.

The Correction Is a Positioning Unwind, Not a Regime Break

The 14–19 May drawdown in Momentum was driven by the long side (prior winners falling 5–6%) and a modest rebound in laggards, while Low Volatility gained only 2–3% on its short side. This pattern points to an unwind concentrated in crowded winners, not a broad defensive rotation. The VIX remained well below 20 throughout the episode, contradicting a panic-driven flight to safety. Further, the long book of 12-month Momentum is heavily exposed to the AI & Tech Diffusion and AI Infrastructure themes; the selloff was therefore a partial de-risking of the same AI, semiconductor, and high-beta names that had powered Momentum’s prior outperformance. The implied regime break is absent: the correlation between 12M Momentum and Composite Growth stays elevated, while the negative correlation with Value remains mild, confirming that the shift is a narrowing of the Growth/Momentum premium rather than a lasting rotation into Value.

Earnings Support Remains Intact

As of late May 2026, over 75% of US AI theme constituents have beaten consensus EPS, well above the ~60% beat rate in the broader MSCI US universe. This suggests the recent weakness was not driven by a fundamental deterioration in AI earnings, but by rates-driven multiple compression and positioning pressure. Meanwhile, the US 12M Momentum factor valuation percentile stands at 49% (median of historical range), implying the factor was not stretched enough to trigger a valuation-led capitulation. Should the AI earnings thesis weaken—for example, if hyperscaler capex disappoints—the fundamental cushion would erode and the risk of a deeper selloff would rise materially. For now, the data supports a contained correction.

Earnings Window Momentum Offers a More Resilient Implementation

Over the 14–19 May window, Earnings Window Momentum fell only ~2% in the US, compared to ~8% for traditional 12-month Price Momentum. The pattern held in Europe and Japan, where EWM drawdowns were roughly one-third the magnitude of 12M Momentum. EWM directly links to stock-specific earnings signals, which reduces exposure to crowded, macro-driven AI winners. In a higher-rate environment where the market is becoming more selective about AI-led momentum, EWM preserves upside participation while providing a structural buffer against concentrated unwinds.

Key Risks

A material further rise in US yields (e.g., 10Y above 5%) could trigger a second leg of multiple compression in long-duration Momentum names, especially if term premia continue to rise. A weakening of the AI earnings thesis—whether from disappointing hyperscaler capex or slower AI monetisation—would remove the fundamental cushion supporting Growth and Momentum. Under either scenario, the shallow drawdown advantage of EWM may narrow as correlation across all Momentum variants increases.

Implementation Implication

Maintain Momentum exposure but rotate from 12-month Price Momentum into Earnings Window Momentum. EWM’s fundamental grounding reduces vulnerability to crowded AI-led momentum unwinds while still capturing the AI-driven rally that the earnings data continues to support. This positioning is appropriate for a higher-rate environment where selectivity matters more than outright factor direction.

Appendix Data Summary