Shenwan Hongyuan Group: Slowing organic demand for US equities in 2027; focus on downstream AI applications and low-beta defensive sectors

Shenwan Hongyuan Group released a research report stating that in 2026, US equities played out around two main threads: AI development and the Middle East conflict. Major indices rose about 10%, the market moved from K-shaped divergence toward convergence, energy led gains for the full year, and global developed-market funds' US equity allocation coefficient reached the historical 100th percentile. Looking ahead to 2027, slowing AI investment growth will drag on the economy and marginally lower its growth contribution; monetary policy remains relatively tight but easing expectations may build earlier than expected; and political-cycle risk rises after the midterm elections. On market pricing, the AI hardware supply shock has gradually played out, with semiconductor valuations falling from 25x to 14.58x, but the compute supply-demand imbalance and the risk of a negative AI financing loop have not yet been fully priced. Structurally, the report recommends focusing on low-beta defensive healthcare and consumer staples, AI application and data collection themes, and policy opportunities in power and new energy after the midterm elections. The main views of Shenwan Hongyuan Group are as follows: Since 2026, US equities have played out around two main threads: changes in AI's own development and the Middle East conflict 1) For the full year, excluding small caps, large-cap growth, CKH HOLDINGS, and large-cap value all rose as earnings digested valuations, with major index gains concentrated around 10%. 2) The market went through a process from K-shaped divergence to convergence, with gains rotating from AI compute in the first half to AI applications in the second half and then to consumer, while energy led for the full year. 3) In the first half, US equities mostly rose in tandem with Asia-Pacific markets; in the second half, US equities significantly outperformed non-US markets. As of 2026Q3, global developed-market funds' US equity allocation coefficient had also reached the historical 100th percentile. Looking ahead to 2027, US investment-driven endogenous demand will slow, and the sensitivity of potential demand improvement to liquidity and policy will rise 1) On the economy, after AI investment growth slows, its marginal contribution to the economy will be revised down, awaiting the gradual release of pent-up non-AI demand. US industrial and consumer capex is already below the historical 20th percentile, and household credit has also fallen to a low level. Marginal relief in real interest rates will help improve non-AI manufacturing and ease "carbon-based" consumption. 2) On monetary policy, the lagged impact of AI investment on the economy still marginally supports Fed tightening; watch for upside liquidity risk from changes in the monetary policy framework, while oil price uncertainty remains a potential constraint. A historical review shows that from the first rate hike to the inflection point in liquidity expectations takes at least 9 months, with only 1997 taking less than 1 month, when US equities, after fluctuating higher, met the Southeast Asian financial crisis and the Fed subsequently began cutting rates. At present, AI's impact on end-service prices is somewhat deflationary but the aggregate effect is limited. If the Fed subsequently emphasizes its assessment of AI's long-term deflationary trend, this may bring forward the build-up of easing expectations. The impact of political-cycle risk on the market will be greater than in 2026 Under the baseline assumption, the White House's substantive impact from congressional interference on diplomacy, tariffs, and immigration policy is limited, and there is a basis for bargaining with Congress on industrial policy (AI, healthcare, new energy, etc.). 1) Lowering the cost of living is a potential basis for bipartisan cooperation, but differing positions mean the substantive impact should not be overestimated. The inflation issue was both the reason Trump won the 2024 election and the reason the Democrats swept Congress this round. After this round of elections, the market's first reaction was to price in gains for consumer, retail, and airlines, while construction engineering, power equipment, and semiconductors led declines. Historically, however, the probability of reversal within three months for sector performance on the first trading day after midterm elections is relatively high, with continuity over the next 12 months relatively higher, though the impact is only slightly above random probability. Going forward, watch the room for bipartisan negotiation on domestic infrastructure supply such as energy diversification, power, and housing, as well as on external tariffs and the US-Iran conflict. 2) On AI buildout, the two parties have a cooperative-competitive relationship, and rising construction costs plus tighter regulation are a high-probability outcome. The two parties will likely reach consensus on the rising power and water costs brought by data center construction itself. However, AI regulation will likely tighten somewhat, with Democrats more cautious than Republicans on "AI frontier safety" and "compute trade" issues. After 1998, based on the Microsoft antitrust case, Congress held multiple hearings that caused fluctuations in market sentiment. 3) US populism is intensifying, and a phased escalation of partisan conflict could shock market risk appetite. The divergence between US capital and labor compensation is currently at its largest since 1960, and distributional conflict may make it harder for the two parties to reconcile on the debt issue. After the president lost Congress in 1994, two government shutdowns in Q4 1995 triggered market turmoil. And in 2010 the president faced a divided Congress, and throughout 2011 the battle over the debt ceiling and fiscal contraction led to downward revisions in corporate earnings. From the perspective of market pricing, the supply shock from AI hardware to short-term data center construction has gradually played out As of 2026/10/30, the forward valuation of US semiconductor products and equipment was 14.58x (versus a high of 25x at the start of the year), with past cycle bottoms at roughly 10-12x. Earnings expectations for 2027 are still being marginally revised up, but growth is close to stalling. Historically, share prices tend to gradually bottom after earnings expectations decline. For the Nasdaq, after including Anthropic and Open AI, the static PE valuation is 30.2x, at the historical 47.1st percentile, and the 2027 forward PS valuation is 5.25x, at the historical 69.4th percentile. Changes in earnings themselves remain key to the market's subsequent direction. However, potential changes in the supply-demand imbalance in 2027 that could disrupt compute prices have not yet been fully priced. Under a neutral assumption, US compute supply will add 50GW in 2027, a higher supply growth rate than this year. On the demand side, Anthropic's revenue grew 10x this year and may slow to 2-3x next year, implying a marginal slowdown in demand growth. Qualitatively, deep AI customers' token spending grew sharply in the first half and relatively stagnated in the third quarter, and AI revenue growth depends on broader diffusion of paying users. This year, token prices fell 50% amid compute progress and model competition, while current GPU rental prices remain firm; this is the core indicator to watch for changes in the supply-demand landscape going forward. Watch tail risk from macro second-derivative shocks triggered by rising financing constraints AI capital expenditure officially shifted from cash-flow funding to primarily external financing in the third quarter. Based on the current capex pace, North American cloud companies' combined cash flow is expected to be unable to turn positive in 2027. The bond market has already begun pricing project default risk. As of 1105, Coreweave's CDS (>1200bp) has exceeded Lehman crisis-era CDS (600bp), while the average CDS of AI technology companies has risen from 60bp to the current 133bp; at that time, bank CDS rose from 60bp to a peak of 200bp. For the macro economy, after bank credit risk deteriorated in Q3 2007, cross-industry credit spreads widened, and from 2008 onward broad fundamentals weakened substantially. Industrial credit risk has now emerged, but Baa spreads are still narrowing; the magnitude of the macro shock may be smaller than in 2008, but the signal and logical path still matter. Watch for a possible negative feedback loop: rising financing costs marginal projects cut capex growth declines supplier earnings revised down collateral (GPU) valuations fall financing costs rise further. On timing, the market in the short to medium term remains in a digestion phase after the momentum reversal, in a transition period for liquidity expectations from tightening to easing, and in the long term awaits the emergence of AI's positive externalities US equities entered a digestion phase after the mid-2026 momentum reversal. Judging by the pace of digestion, it is not yet sufficient, and the market is consolidating through breadth expansion. After the tech momentum reversal in 2000, the S&P 500 fluctuated at highs for two quarters, during which financials, industrials, utilities, and energy remained resilient, until fundamentals deteriorated, overall earnings were revised down, and the market corrected. According to bottom-up expectations, S&P 500 earnings growth will slow from 32.9% in 2026 to 16.3% in 2027, with the low point in earnings growth likely in 2027Q2; watch for changes in liquidity as earnings expectations slow. In the long term, with AI as a productivity tool, AI spending by the top 500 US wealth companies currently accounts for 0.1%. By scenario, according to Nvidia's disclosed classification of final compute consumption, CSP internal use accounts for 42%, frontier models 18%, traditional enterprises 12%, edge computing 7%, and AI-native companies 22%. Among these, enterprise AI and edge computing correspond to enterprise-grade agents and Siasun Robot&Automation scenarios, respectively, and their shares still have room to rise. Structurally, based on earnings growth expectations, the sectors with the most obvious improvement in 2027 are healthcare and industrials Other sectors will see marginal slowdown in growth next year; information technology will slow from a high level, while telecom and energy may enter negative growth territory. From the perspective of potential catalysts, the report recommends looking for opportunities around the low-beta market style, AI themes, and policy preferences after the midterm elections. 1) Within the low-beta market style, focus on defensive healthcare and consumer. As areas that were previously underweighted by the market, healthcare has already seen right-side capital inflows, while consumer is still seeing marginal outflows. By segment, as AI for Science accelerates, pharmaceutical companies' R&D demand continues to expand, with the life sciences industry benefiting more. Medical equipment earnings have passed their peak and offer some valuation attractiveness; like hospitals and health insurers, they depend on policy support from the Democrats after the midterms. 2) On AI themes, the value of data collection in downstream scenarios may become more prominent, and opportunities on the application side deserve more attention. In terms of AI application penetration, from tool payment to workflow replacement to macro-level diffusion, US vertical applications are entering the second stage. In the primary market, financing is more active in enterprise process support, industrial manufacturing and operations/maintenance, and medical technology. However, no company has yet replicated the early ARR growth rate of the coding track (vertical application ARR is around US$10 billion). Currently, whether for physical AI or enterprise-grade AI, penetration improvement is at a development stage where insufficient data limits model performance; software companies focused on data collection and data distribution deserve attention. On the To C side, overall PB valuations for edge consumer electronics are at a mid-level; as cost pressure eases, revenue and profit margins are expected to improve marginally next year. In addition, companies that control user traffic entry points (such as social and retail) remain directions where AI revenue could grow non-linearly. 3) On policy direction, after the Democrats take power, opportunities in power investment and new energy may again deserve attention, while defense and banks face headwinds. Democrats advocate clean energy, but restoring tax credits requires presidential cooperation, making near-term delivery difficult. If Democrats become more proactive at the regulatory level, such as on fuel vehicle emissions standards, energy storage and solar may see marginal valuation repair. In addition, grid investment and nuclear investment are long-term directions where the two parties have consensus. Risk warnings: 1) Watch for the risk of compute oversupply in the US, with the negative feedback from the US equity wealth effect to the real economy; 2) during the evolution of AI technology, slow wealth redistribution due to changes in corporate organizational forms, rising regulatory and compliance costs after increased social negative externalities, and difficulty in effectively expanding macroeconomic demand; 3) the Middle East conflict lasting longer than expected, raising stagflation pressure, while US fiscal expansion is already relatively constrained, and a selloff in the bond market triggers a liquidity crisis; 4) escalation of geopolitical protection measures raises corporate operating uncertainty and shocks earnings expectations.