CICC: Have US stocks fallen enough?

CICC: Have US stocks fallen enough? CICC released a research report stating that recent volatility in US stocks has increased markedly. Policy "noise" after the midterm elections triggered market turbulence, and the key issue is that AI itself is currently facing many "problems," such as a demand "gap" and high investment, which have been exposed ahead of time. Of course, this rapid clearing may not be a bad thing for the long-term trend, as it helps digest excessively high expectations and positioning, which is instead beneficial for regaining upward momentum when new catalysts emerge; it is just that new catalysts need to be awaited. It should be noted that the gains in US stocks so far this year have basically been contributed by earnings rather than valuation expansion, and after the recent pullback, valuations have fallen even faster. Apart from the Dow's 19.6x valuation, which is still at the 86th percentile of the past 20 years, S&P 500 and Nasdaq valuations have fallen to 17.8x and 20.3x, corresponding to the 70th and 44th percentiles of the past 20 years, while the Philadelphia Semiconductor Index has even dropped to 13.8x, only the 14th percentile of the past 20 years. CICC's main views are as follows: This year's "K-shaped divergence" in US fundamentals has caused the market to concentrate and rely on a few areas: 1) US Mag7 contributed 22% of the S&P 500's gain since 2026, and before the mid-October decline had contributed as much as 40%; 2) the five major cloud vendors accounted for 20% of S&P 500 earnings in the first three quarters of 2026 and contributed 25% of S&P 500 earnings growth; 3) the leading-stock concentration effect also remains. Since the start of the year, the S&P 500 has risen 8%, but the proportion of stocks above their 200-day moving average has fallen from 60% at the start of the year to 55%, with the divergence especially obvious in August-September. However, AI currently lacks catalysts at the model and application levels and has also run into policy uncertainty, while traditional demand is suppressed by high interest rates. In other words, the main direction is facing challenges while the traditional direction has yet to take over, which explains the overall turbulence. Therefore, the market needs two catalysts: either a major breakthrough in AI, or the Fed ending tightening or even turning to easing. Before that, CICC recommends a cautious wait-and-see approach: 1) watch whether there is short-term negative liquidity feedback, or further policy disruption (the game before the new Congress takes office early next year); 2) control positions to leave room for subsequent allocation; 3) adjust structure. Even within technology, priority can be given to cloud vendors and computing chips with low AI exposure, or low valuations and stronger pricing power. The first quarter of next year is the key point: 1) the window for a clear Fed pivot is expected to come only in early next year. Strong October nonfarm payrolls reinforced the probability of a December rate hike, and early next year inflation is more certain to fall under the support of a high base; 2) fourth-quarter results disclosed in January next year. ROIC for tech companies may continue to come under pressure, and by then new AI progress will be even more needed; 3) the new US Congress takes office on January 3 next year. Before the new Congress takes office, Trump's policies and continued remarks by newly elected lawmakers may cause ongoing disruption, and the market needs to digest this; after the new Congress takes office, the market may gradually become desensitized. So by that time, either AI will need to make progress, or Fed tightening expectations will need to recede, which may provide support or an opportunity. In this article, CICC further combines the basically fully disclosed third-quarter US stock results to answer the question of the US stock outlook from a fundamental perspective. Market DRIVE: Earnings are the market's, with AI contributing 60% of growth; growth shows an inflection point First, strong earnings growth is the main driver of US stocks. Year to date, the gains in major US stock indices have all been contributed by earnings. The S&P 500, Dow Jones, and Nasdaq rose 8%, 7%, and 6%, respectively, with forward earnings growing 29%, 17%, and 40%, respectively, offsetting valuation contraction. Third-quarter S&P 500 earnings grew 25%, down from 32% and 56% in the first and second quarters, but still above the 12% average quarterly earnings growth during 2023-2025. Second, AI-related sectors contributed 60% of earnings growth. Estimating the AI sector using communication services, information technology, and Amazon, it contributed 62% of overall earnings growth in the third quarter, down from 70% in the second quarter. Among this, information technology contributed 58%, supporting the vast majority of growth in the AI sector, while communication services weakened markedly due to a sharp slowdown in the media and entertainment sub-industry. This also makes overall earnings growth dependent on AI alone; excluding the AI sector, third-quarter earnings growth would likely fall to 15%. Traditional demand earnings were not bad in the third quarter, but under high interest rates they still dragged on earnings. By growth rate, energy (150%), semiconductor equipment (110%), real estate management and development (49%), and capital goods (48%) led, while transportation (-37%), telecom services (-34%), financial services (-14%), and insurance (-11%) lagged. By contribution, industrials, materials, utilities, and consumer staples together contributed 10% of earnings, while financials and real estate turned negative, dragging overall growth by 7.5%. Third, growth showed an inflection point, slowing for the first time since 2Q25. Third-quarter S&P 500 earnings grew 25%, slower than in the first two quarters of this year. The market has become more "picky" about results, requiring not only high earnings growth but also continuous acceleration above expectations, which also increased the difficulty of beating expectations in the third-quarter reporting season. Among S&P 500 constituents, both the proportion and magnitude of beats declined. The proportion edged down from 86% in the second quarter to 84%, while the magnitude of beats fell significantly from 27% in the second quarter to 3.5%, the lowest since the fourth quarter of 2022. Among these, communication services, where Google and Meta are located, saw the largest decline, with earnings shifting from 99% above expectations in the second quarter to 19% below expectations in the third quarter. AI direction: Revenue lags investment; returns are falling but still above cost; can withstand rates falling to 6-6.5% 1) Revenue lags investment. Capex is accelerating, free cash flow has turned negative, and external financing pressure is rising. The five major cloud vendors' combined third-quarter capex rose from US$181.5 billion to US$230 billion, with year-on-year growth accelerating from 87% in the second quarter to 118%. However, operating cash flow growth slowed markedly, from 38% in the second quarter to 19% in the third quarter, totaling US$198 billion for the five major cloud vendors, no longer enough to cover capex over the same period. This turned the five major cloud vendors' free cash flow from US$4.8 billion in the second quarter to -US$31.8 billion. Except for Microsoft, Amazon, Meta, Google, and Oracle all saw negative free cash flow in the third quarter, further increasing external financing needs. Against the backdrop that AI demand has not yet broken through the "ceiling" of existing application scenarios and the investment return outlook remains unclear, the market's attention to financing costs and funding sources has risen markedly, and concerns about credit risk have intensified accordingly, with CDS for the five major cloud vendors rising to historical highs. 2) Returns are above cost. Marginal ROIC has fallen further, but overall returns can still cover financing costs. As invested capital continues to expand and incremental operating returns fail to keep pace, the five major cloud vendors' marginal ROIC has continued to decline from a peak of 44% in the second quarter of 2024 to 6.1%, dragging overall ROIC further down to 15%. However, overall ROIC is still above the five major cloud vendors' current weighted average cost of capital (WACC) of about 10%, meaning cloud vendors can still earn excess returns. The decline in marginal returns has lowered the interest rate level AI can withstand to 6-6.5%. Previously, CICC estimated that if the five major cloud vendors maintained ROIC unchanged and assumed the gap between ROIC and WACC narrowed to the average level of 8ppt before the rise of the AI trend in 2015-2023, the corresponding risk-free rate they could withstand would be 7-7.5% ("AI: From a 'Friend' of Traditional Demand to an 'Enemy'?"). However, as marginal investment returns weakened further in the third quarter, the gap between ROIC and WACC has already converged to the historical average. Combined with return expectations for 2027, the risk-free rate level AI can withstand has fallen to 6-6.5%. 3) Cloud vendors' "asset-heavy" shift is putting pressure on capital turnover. Further decomposing ROIC into after-tax operating margin (NOPAT/Revenue) and invested capital turnover (Revenue/IC), the five major cloud vendors' after-tax operating margin has continued to rise since 2023, from 14.8% to 22.1% in the third quarter; however, capital turnover has continued to fall, from a peak of 109% in the first quarter of 2023 to 66% in the third quarter, a new low since 2015. At the same time, since the second quarter of 2025, the slowdown in margin improvement has been unable to offset the drag from capital turnover, which has also caused the five major cloud vendors' ROIC to keep declining. Technology is diverging further internally. Some hardware segments still need to digest growth expectations, while cloud vendor pricing is relatively reasonable. Some hardware infrastructure that had priced in higher growth expectations, such as network interconnection/optical modules, semiconductor manufacturing, and liquid cooling/temperature control, have seen notable valuation pullbacks recently. For example, the valuation of the semiconductor manufacturing sector quickly dropped from the 90th percentile in early October to around the 20th percentile. By contrast, cloud vendors have not been given an obvious valuation premium for incremental earnings, and their valuation adjustment room is relatively limited. Traditional sectors: Rate-sensitive segments are suppressed by high rates, and AI crowding-out is gradually showing Traditional demand has not yet formed a broad improvement. In consumption, services consumption is stronger than goods consumption. Consumer services, food & beverage and tobacco, and autos & components earnings grew 12%, 8.7%, and 7.3%, respectively, but durables & apparel and household & personal products fell 5.4% and 10.3%, respectively. The divergence in industrials is even more obvious: capital goods earnings grew 48.3% and commercial & professional services grew 16.2%, but transportation, eroded by high oil price costs, fell 37.1%. In financials, bank earnings rose slightly by 4.8% supported by net interest income, while insurance and financial services fell 10.7% and 13.6%, respectively, due to factors such as volatility in investment gains and losses. Growth remains resilient, but rate-sensitive growth segments continue to bottom out. Recently released macro data reflect that current US growth remains resilient. Third-quarter real GDP grew at a 3.4% quarter-on-quarter annualized rate, above the expected 3.2% and the second quarter's 2.2%. In addition to AI-related investment, private consumption is also strong, especially services consumption such as healthcare, dining, and accommodation. October ISM manufacturing and nonfarm payrolls also greatly exceeded expectations, and the unemployment rate fell from 4.2% to 4.1%. But high interest rates continue to suppress the release of household homebuying and credit demand. The 30-year mortgage rate has continued to rise since the start of the year and briefly rose to 7.4% in October. Real estate sales remain hovering at low levels. Although September existing home sales rebounded slightly month on month, the annualized sales volume was only 3.99 million units, and new home sales also fell to 676,000 units. On the other hand, AI with weakening returns may also begin to "crowd out" real and financial factor resources from traditional demand. As CICC emphasized in "AI: From a 'Friend' of Traditional Demand to an 'Enemy'?", the positive externalities brought by AI demand previously supported US economic growth, but as AI needs more external funding to support investment, crowding out of traditional demand is also unavoidable. AI's large-scale bond issuance and the Fed still needing to raise rates are typical examples. A larger recovery still requires a decline in interest rates to accompany it. CICC previously estimated, using real estate and non-AI companies, that traditional demand can withstand a 10-year US Treasury yield of about 4%. Although there is resilience now, high rates are still squeezing the further release of household homebuying and small and medium-sized enterprise credit demand, reflected in insufficient momentum for credit expansion in the household sector. In the short term, growth resilience and employment data leave the Fed lacking the motivation to pivot immediately, but the weakness of traditional demand also does not support a larger Fed rate hike. If AI expectations remain weak in the first quarter of next year, then the market's expectations for Fed tightening may recede, creating conditions for lower rates and a recovery in traditional demand. Outlook judgment: The market needs the catalyst of either the end of tightening or AI progress; early next year is the key node At present, the market's expectations for AI are themselves in a "gap" phase. The key problem is that there is no new narrative or new catalyst in demand and models, so it also amplifies the risk of credit and regulatory pressure. High-intensity capex continues to squeeze cloud vendors' free cash flow, causing rising concerns about the pace of AI revenue realization and investment returns. This is similar to the logic at the end of last year when AI progress was insufficient and the market worried about private equity and credit risk. Against this backdrop, the market needs two catalysts: either a major breakthrough in AI, or the Fed ending tightening or even turning to easing. If neither appears, it may affect investors' willingness to re-enter and bottom-fish, and also amplify volatility. In the short term, attention should be paid to whether there is negative liquidity feedback, or further policy disruption (before the new Congress takes office early next year, whether Trump will use this window for policy bargaining). The first quarter of next year is the key verification point, as the Congress taking office, AI realization during the earnings season, and falling inflation will all bring changes. Before that, CICC still recommends a cautious wait-and-see approach in the short term. Before the above catalysts appear, watch in the short term whether negative liquidity feedback and further policy disruption occur. The S&P 500 has now returned to near its annual line, and the earlier adjustment has already released some sentiment, but the market still lacks new upward catalysts, and factors such as liquidity sub-items and a December Fed rate hike may also bring further disruption. Therefore, CICC recommends waiting until the above catalysts appear and the investment return outlook becomes clearer before allocating. From an extreme scenario perspective, if all the additional earnings growth brought by AI expectations were "erased," forward earnings growth could fall to 10%. Assuming valuations simultaneously fall to 16-17x, this would imply the S&P 500 still has about 10% downside. In addition to controlling positions well, investors can also adjust holdings structure. Even within technology, priority can be given to cloud vendors and computing chips with low AI exposure, or low valuations and stronger pricing power. Recent market performance also reflects this divergence: segments that had previously priced in fuller growth expectations pulled back more noticeably, while cloud vendors were relatively resilient. Using the average valuation of the past 10 years as a reference, sectors such as optical modules, semiconductor manufacturing, and power grids & distribution still need earnings upgrades of 40-70% to match current pricing, while cloud vendors still have nearly 20% of earnings downgrade room.