Morgan Stanley reviews the AI selloff: it's valuation being hit, not earnings! But don't rush to buy the dip.

Notice that this global AI stock pullback is less a collapse in fundamentals than a pure "valuation bubble squeeze." In a report released on November 8, Morgan Stanley's quantitative equity research team reached a rather counterintuitive conclusion: this decline in the AI sector was almost entirely driven by valuation compressionwhile consensus earnings expectations have not been revised down, but are instead still steadily moving higher. The intensity and structure of the decline From September 30 to the close on November 5, the AI & Tech Diffusion index compiled by Morgan Stanley fell 26.6% cumulatively, while the MSCI ACWI global index fell only 3.7% over the same period; extending the Japan- and Asia-listed constituents to the close on November 6, the portfolio's decline widened further to 28.4%, a 33% drawdown from its historical peak on October 6. The pullback unfolded in two stages: the first wave in mid-October, and the second starting at the end of October and deepening markedly after the U.S. midterm elections on November 3. Valuation compression "contributed more than the entire decline." The report decomposed prices for 91 constituents with positive earnings in both periods: equal-weighted average FY2 consensus EPS rose by about 3 log points across regions, while P/E ratios contracted by about 37 log points overall (ranging from 26 to 48 log points across regions). In other words, the valuation de-rating was not only sufficient to explain the entire decline, but more than sufficient. Put differently, investors may have priced in the risk of slowing AI infrastructure growth before consensus expectations reflected it. Regional divergence: Japan saw the harshest valuation de-rating On a percentile basis relative to their own valuation history (since September 2019), U.S. and European AI stocks have fallen from the middle of their ranges to the bottom fifth (median percentiles of 17 and 13, respectively), while Japan plunged from the 82nd percentile to the 28th percentile, making it the most typical example of "resilient earnings, compressed valuations"; Asia/emerging markets, by contrast, remain near their own historical median (53rd percentile) and have the highest share of downgrades7 of 26 constituents (27%) had FY2 estimates cut by more than 0.5%. How thick is the valuation "safety cushion"? Morgan Stanley ran a stress test: if a hypothetical 20% cut were applied to 12-month forward EPS, the share of constituents trading above their own historical median P/E would rise from the current 27% to 55% across the portfolio. The median "break-even downgrade magnitude" each region can absorb is approximately: U.S. 24%, Europe 23%, Japan 16%, while Asia/emerging marketsbecause more than half of their stocks are already above their own medianshave a negative median buffer, leaving almost no safety cushion. "Adopters" outperform, but earnings momentum remains upstream Since the end of September, the AI Adopters portfolio has returned -3.2%, while the AI Enablers portfolio has returned -29.0%, a gap of about 26 percentage points, and the gap remains significant after stripping out market beta. But it is worth noting that this style rotation is "valuation-driven" rather than "earnings-driven": the median forward P/E of Enablers fell from 27.4x to 18.6x, now almost on par with Adopters' 18.1x, yet their net earnings upgrade breadth still reached +48%, far exceeding Adopters' +2%, and 35% of Adopters had FY2 estimates revised down. Capex becomes the focus The report also found that the stocks with the largest upgrades actually fell the most (rank correlation -0.42); capex-intensive companies saw average valuation de-rating of about 40 log points after earnings, significantly higher than the 25 log points for other companies. Holding earnings upgrades constant, every 10% increase in capex expectations corresponds to about 4 percentage points of relative underperformance (t-statistic -2.8), indicating that the market is increasingly scrutinizing the returns on incremental AI investment. Three main lines of response Morgan Stanley's core conclusion is: the valuation reset has improved the entry point for AI, but earnings delivery remains key. Strategically, first, avoid "indiscriminate dip-buying" in upstream AI, and watch for the lagged impact of tighter data center regulation and project delays on orders and earnings; second, selectively increase exposure to AI Adopters, but prioritize names with resilient earnings expectations rather than adding broadly; third, prioritize regions with valuation safety cushions (the U.S. and Europe have the thickest, Japan has seen the most thorough valuation de-rating), and keep a close eye on earnings revision trendsespecially the 33 constituents whose latest earnings have not yet been incorporated into analysis. At the macro level, the Iran conflict and energy market volatility have amplified risk aversion, while the renewed rise in yields remains another valuation risk independent of earnings.