Shenwan Hongyuan Group Co., Ltd. has released a research report indicating that after the oversold rebound, disruptions in the AI industry have resurfaced, compounded by concerns over elevated U.S. Treasury yields, triggering a secondary bottoming-out trend. The firm maintains its medium-to-short-term outlook, projecting the rebound could extend until late September, supported by steady policy implementation. The report suggests that during the September rebound, structural selection should focus on sectors capable of recovering a larger portion of their declines. Within the AI chain, greater attention should be directed toward non-institutionally held positions, with domestic compute chains and AI small-cap stocks providing "medium-term cover for short-term moves." Additionally, within the global compute chain, sectors with later cyclical downturns and fresh near-term catalysts—such as memory and PCB—are poised for high elasticity, warranting exploration of new compute inflation sub-segments and targets.

Short-term market trajectory aligns with the firm's "four-step" scenario for AI chain resurgence. Following the oversold rebound, AI industry disruptions reappeared alongside high Treasury yield concerns, resulting in a secondary dip. The medium-to-short-term outlook remains unchanged: the rebound may persist until late September, with steady policy measures potentially triggering a convergence of optimistic expectations across long, medium, and short horizons—likely marking the peak of this rebound cycle. The positive capital feedback loop is unlikely to return to end-June levels quickly, and AI chain new highs require fundamental expectations to surpass those of June. Compared with trends since 2025, this AI chain revival demands higher-level catalysts and likely a longer gestation period. Consequently, September may not see new highs, and further corrective waves could emerge, with markets anticipating a lengthening of tech consolidation from monthly to quarterly scales.

The current market performance largely mirrors the firm's projections, reaffirming the "four-step" path to AI chain strength: first, oversold rebounds restore fundamental research effectiveness; second, as floating gains and losses return to breakeven, rebound resistance naturally increases. This week's short-term disruptions in large-model ARR, combined with high Treasury yield concerns, have realized the secondary bottoming-out. However, post-correction buying interest has emerged, making a rebound through late September highly probable. Steady policy measures could fuel a convergence of optimistic expectations, potentially marking the rebound's zenith. Third, by the Q3 2026 earnings season, valuation digestion will be preliminarily complete, restoring earnings growth support for share prices, with overseas compute chains likely to experience differentiated trends—some high-certainty leaders may resume uptrends. Fourth, AI chain new highs hinge on major industrial catalysts and fresh consensus formation around AI trends. It is highly probable that September's oversold rebound peak will fail to breach previous highs.

The positive capital loop cannot swiftly revert to end-June conditions; new highs necessitate fundamental expectations surpassing June levels. Since 2025, each mid-cycle correction in AI trends—marked by floating gains turning negative—has confirmed new industrial trends within a quarter. This cycle's higher-level correction, potentially linked to AI Coding penetration reaching elevated levels, requires stronger catalysts and likely longer waiting times. By late September, a new AI industry consensus may not yet form, preventing market breakthroughs and opening fresh corrective waves. Markets will then anticipate a lengthened tech consolidation, shifting from monthly to quarterly duration.

Elevated U.S. Treasury yields stem from both medium-term and short-term factors. The rise in Treasury yields is not a novel development; the market's concentrated near-term pricing centers on tech correction waves, layered with medium-term high-rate concerns, amplifying adjustment depth—this skews toward small-wave characteristics. The firm summarizes the medium-term issue as: the Federal Reserve's monetary easing faces constraints, requiring a marginal weakening of AI capital expenditure's economic support. A return to a favorable investment clock quadrant will take time and require fundamental expectations to navigate turbulence. Rational market projections suggest a phase of concentrated risk release may precede a new tech upcycle.

Discussing high Treasury yield impacts, the report notes both medium-term and short-term drivers. Medium-term factors include: recurring U.S.-Iran conflicts and resource nationalism spreading, fueling medium-term inflation and passive rate-hike fears; weakened U.S. fiscal discipline with short-borrowing-long-lending practices accumulating medium-term risks; and Trump's politicization of dollar circulation mechanisms. Short-term factors encompass increased Treasury issuance pressure, bond financing by AI supply chains diverting allocation demand, and alternative reserve assets like gold. Treasury yields have risen for some time, not a new shift; the sudden spike in market attention stems from tech industry disruptions intersecting with high-yield concerns, intensifying near-term corrections. This attention surge itself reflects small-wave traits. Medium-term concerns are concentrated, while short-term factors are transient, with subsequent mitigation waves likely.

The medium-term issue behind rising Treasury yields, per the firm, is: high financing levels in the U.S. AI supply chain lack direct, effective tools for managing potential credit risks. Price tools face constraints, and existing policy stances favor cautious use of quantity tools. This week, the U.S. Treasury's quasi-QT has shown some effect, but more effective measures would involve QE or even more breakthrough targeted easing tools. Meanwhile, the Fed's transition from rate-hike expectations to actual cuts requires numerous preconditions, including diminished AI capital expenditure support for overall growth, corresponding to compute inflation contraction and emerging compute deflation. Thus, before a favorable investment clock quadrant emerges—also preceding an environment conducive to AI application-side expansion—fundamental and liquidity expectations may face repeated disruptions. Consequently, a phase of concentrated risk release before a new tech upcycle is a rational projection. Near-term, high Treasury yields may only induce a one-off risk appetite shock, but within late September's corrective wave, medium-term issues behind high yields could remain significant disruptors.

September rebound continuation calls for structural focus on sectors with maximum decline recovery. Within the AI chain, prioritize non-institutionally held positions, with domestic compute chains and AI small-caps providing "medium-term cover for short-term moves." Globally, within the compute chain, sectors with later cyclical downturn inflection points and fresh short-term catalysts—such as memory and PCB—are likely to deliver high elasticity, necessitating exploration of new compute inflation sub-segments and targets. Post-September, markets may anticipate lengthened tech consolidation waves, extending the outperformance period for non-tech sectors and increasing absolute return windows for high-dividend assets.

In non-tech tracks, innovative drugs and CXO represent areas already demonstrating positive capital loops. Precious metals consensus is rapidly forming, potentially driving similar positive capital dynamics. Opportunities in industrial metals and basic chemicals also merit attention. Over the medium term, with increased correction waves, high-dividend assets may offer repeated opportunities. The report reiterates that the CSI 800 Index weight minus Q2 2026 public fund holdings weight may serve as an effective approach for identifying high-dividend assets, with a focus on banks, non-bank financials, food and beverage, and utilities. Risk warning: overseas economic recession may exceed expectations.