CITIC SEC: The market is awaiting the emergence of a new narrative for AI futures, and investors need to return to reasonable expected returns.

We categorize the market into three performance themes: the North American AI chain, the overseas expansion to non-US markets, and domestic demand: 1) For the North American AI chain, after a two-week correction, the odds of the price increase chain have clearly diminished. Recently, there have been some marginal changes in AI narratives, but they are still insufficient to drive a revaluation of the entire sector; we need to be patient for the emergence of new long-term narratives; 2) Regarding the overseas expansion to non-US markets, the trade dispute between China and Europe may continue to be tense for a while, suppressing valuations in many high-end manufacturing sectors. At the same time, as we approach the intensive disclosure period for interim reports, the impact of exchange rate losses should not be underestimated; 3) For purely domestic demand products, the relative advantage of domestic computing power is clear, but considering that valuations reflect optimism about long-term narratives, the sentiment in this sector may struggle to remain independent of the North American AI chain. Overall, many sectors still exhibit an upward industrial trend, but the factors restricting valuation expansion are increasing, and the pace of growth may slow. The instances of stock prices doubling in the second quarter are extremely rare even over a ten-year horizon, and investors need to revert to reasonable expected returns. Market opportunities may increasingly manifest as valuation recoveries supported by performance; the subsequent process of market declining volatility is also a process of optimizing the shareholding structure and re-pricing quality assets, laying a solid foundation for the continuation of the mid- to long-term market trend. The oversold repair process of the AI price increase chain is nearing its end. We still need to be patient for the emergence of new long-term narratives. 1) After a two-week correction, the attractiveness of the AI price increase chain's odds has declined. If profit expectations are not significantly revised upwards, the odds of the AI price increase chain will gradually diminish alongside valuation recovery. In previous reports, we mentioned that the low points during the first wave of the super cycle's peak adjustment will usually not see a dynamic P/E lower than that at the cycle's outset. On July 29, the median dynamic P/E of related domestic AI price increase chain samples had already reached 1.18 times that at the cycle outset. If there is a rebound back to the previous high forming a double top, it corresponds to an expected return of about 88%. If it adjusts downwards, the maximum foreseeable adjustment limit (returning to the dynamic P/E level at the cycle start) corresponds to an expected loss of about 15%, indicating that at the end of July, the expected odds/downside space of the AI price increase chain had a ratio of 6 to 1. However, as of August 14, following a notable correction, the median dynamic P/E of related domestic AI price increase chain samples had risen to 1.5 times the level at the cycle's commencement. Under the same calculation, the expected repair odds/downside space is only 1.2 times, and the attractiveness has significantly diminished. Although the industrial trend and pricing logic have not shown significant changes, considering the floating losses at higher levels and the floating profits from this round of rebound, the prospect of successfully playing for a double top and anticipating a one-time valuation repair is challenging. What we should genuinely participate in is the logic behind creating new highs, rather than merely speculating on the top of cyclical stocks. 2) There is a marginal change in the AI narrative, but it may still be insufficient to lead to a revaluation of the entire sector. Recently, new industrial narratives around AI for Science and RSI have emerged, with hopes that these applications can bring about greater computing power consumption and commercialization scenarios. However, reviewing the AI market this year shows that there is a significant time lag between the explosion of technology or products and the initiation of capital market trends. Market explosions only begin with a massive consensus on products or technologies and a substantial divergence in stock price movements. The reasoning model o1 is set to launch at the end of 2024, with a notable increase in token consumption and the North American exchanges expanding scheduled for March to April 2025. However, during that time, market attention was still focused on DeepSeek and trade frictions, and the relevant market didn't truly start until June 2025, accelerating only during the earnings forecasts in July. Similarly, the Coding Agent (Claude Code + Opus 4.6) exploded in early February 2026, and OpenClaw (Little Lobster) rapidly spread at the end of February, while the GLM-5 launched during the Spring Festival showcased some coding capabilities of cutting-edge models. However, the North American computing power chain did not start its rally until the end of March, and the domestic AI chain only saw a broad increase in May and entered an acceleration phase in June. Thus, the widespread dissemination and rapid increase of technology and product penetration need to combine with the significant expectation gap in capital market pricing to create a noteworthy market trend. For AI for Science, RSI, and similar directions, it is entirely reasonable to wait until products truly explode and their penetration rates are factually verified before making any investments, without rushing. When there is high consensus in the capital market but the technology and products themselves are not mature, the chances for investors to miss a significant market movement are very low. Of course, there are also some narratives marginally catalyzing computing power demand recently. The strong performance of Grok 4.6 has been attributed by some investors to its larger model scale. The market seems to have resumed telling the "miracles come from robust training" story, combined with a phase of rebounds in U.S. and South Korean stocks, while Amazon and Microsoft have each reached new highs, and Nvidia is nearing its previous high. The narrative of increasing computing power demand from the training end received some positive feedback in stock prices. However, as long as the models do not exhibit non-linear growth in capability, larger models may still not resolve market concerns regarding the commercialization of AI applications at the backend. These marginal changes are more likely to support rebounds in core stocks rather than elevate the entire sector to the next level. Risks related to overseas expansion to non-US markets remain. This year is still in the phase of expectation correction. 1) The trade dispute between China and Europe may continue to be tense for a while. September and October may be a critical verification window for achieving phase results in China-Europe economic and trade negotiations. It will gradually become clear whether relevant measures can ease tensions and whether negotiations can achieve substantial progress during this period. Currently, the anti-subsidy tariffs on electric vehicles and procurement restrictions on medical devices have been implemented against China, and there are already cases concerning price commitments and foreign subsidy reviews. The use of tools for economic coercion could represent a potential escalation. Meanwhile, the EU Carbon Border Adjustment Mechanism (CBAM), forced labor regulations, and the Industrial Accelerator Act will extend restrictions from border tax burdens to the entire business chain through carbon emission accounting, supply chain due diligence, product tracing, and localization requirements. Such measures may not drastically lower exports in the short term but will continually increase compliance costs for companies, supply chain adjustments, and overseas establishment costs, consequently lowering profit margins and capital returns, becoming important constraints on the long-term valuations of European-exposed companies. Trade frictions in Europe differ from those with the United States, as they do not solely involve simple trade barriers but impose trade thresholds indirectly through compliance requirements, which have a greater impact on complex supply chains like photovoltaics, textiles, electronics, and automobiles. 2) As we approach the intensive disclosure period for interim reports, it is advisable to focus on the impact of the appreciation of the renminbi and resulting exchange rate losses. According to data from the State Administration of Foreign Exchange, in the first six months of this year, banks' net foreign exchange settlements for clients totaled approximately USD 320.5 billion, equivalent to 1.37 times the total volume expected for the entire year of 2025. This surge stems from both the formation of new trade surpluses creating a "flow pool" and the release of previously unconverted funds, forming a "stock pool." When the appreciation rate of the renminbi against the U.S. dollar exceeds the China-U.S. bond yield spread, the existing settlement trend may further intensify, in turn reinforcing expectations for renminbi appreciation. As of now, among the 384 non-financial listed companies that have disclosed their interim reports, the financial expense rate for Q2 2026 rose by 0.47 percentage points year-on-year, with companies having less than 30% of their revenues from abroad experiencing an increase of 0.26 percentage points, while those over 30% saw increases ranging from 1.04 to 1.74 percentage points. Based on the disclosed interim reports, the substantial rise in financial expenses has primarily been due to exchange rate losses. The next two weeks will enter a period of dense interim report disclosures, after all, overseas expansion has been one of the most significant performance clues in the A-share market over the past two years, and the implications of exchange rate losses need to be closely monitored. In purely domestic demand products, the advantage of domestic computing power is clear. However, the sentiment in this sector may not remain completely independent of the North American chain's beta. This year, the overall fiscal net deficit has been significantly lower than expected. According to data from the Ministry of Finance and Wind, in the first half of the year, general fiscal revenue reached 13.63 trillion yuan, a year-on-year increase of 1.0%, while general fiscal expenditure was 18.20 trillion yuan, a year-on-year decrease of 2.9%. The accumulated deficit narrowed from 5.25 trillion yuan in the same period last year to 4.57 trillion yuan this year, a decrease of 679.3 billion yuan, representing a reduction of 12.9%. Public budget revenue increased by 4.74% year-on-year in the first half, slightly outpacing GDP growth, marking the first such occurrence since 2023. General fiscal expenditure shifted to negative year-on-year growth starting from May of this year, marking the first instance since October 2024. The cautious fiscal approach has somewhat dampened investors' expectations for the pace of inflation recovery, reflected in persistently declining long-term bond yields. Investment in domestic computing power remains one of the few prosperous directions among domestic demand products, but current pricing is also relatively adequate. Following a rebound, from June until now, the domestic semiconductor sector's gains have significantly outperformed North American and South Korean markets, with semiconductor materials and equipment and innovative chips showing cumulative increases of +22.8% and +1.3% respectively, clearly beating Philadelphia Semiconductor and KOSPI150 which saw declines of -3.2% and -19.1%. Since August, the domestic chain has become a direction for funds retreating from certain North American chains, and we estimate that the occupation level has not significantly decreased, which may also restrict the flexibility of subsequent valuations. The upward trend remains unchanged, but the pace has slowed down. Since the second quarter, market volatility has noticeably increased: the number of companies whose stock prices have doubled in a single quarter is at a historically rare level. In Q2 2026, the number of listed companies on the Shanghai and Shenzhen stock exchanges that doubled in price reached 275, the highest since Q1 2007. Such episodes of doubling in stock prices are a rarity every ten years and are not the market norm. Whether it relates to export chains targeting the U.S. market (like the AI industry chain) or export sectors focused on non-U.S. markets (such as new energy vehicles, engineering machinery, cross-border e-commerce, and cyclical products), despite the upward industrial trend, there remain many factors constraining valuation expansion, making it difficult to achieve systematic and significant valuation uplift in the short term. The upcoming phases of the year may likely see market opportunities increasingly manifesting as phase valuation recoveries for products with performance support. The market is likely to experience a period of "withdrawal," transitioning from the illusion of instantly doubling valuations back to seeking reasonable expected returns as the norm. This process will inevitably be accompanied by a sustained decline in overall market volatility. In terms of allocation, within the technology sector, it is advisable to take advantage of rebounds in AI price rising products to timely adjust to core assets (such as gas turbines, wafer fabrication platforms, and semiconductor equipment), placing greater emphasis on "quantity certainty" while being cautious about "price explosiveness." For non-technology sectors, key allocations should focus on energy and chemicals, non-ferrous metals, innovative pharmaceuticals, and leading brokerages with overseas expansion potential. Risk factors Increased friction in the fields of technology, trade, and finance between China and the U.S.; domestic policy intensity, implementation effects, or economic recovery failing to meet expectations; macro liquidity both domestically and internationally tightening beyond expectations; escalation of conflicts in regions such as Ukraine and the Middle East; slower-than-expected digestion of real estate inventory in China. This article is selected from the WeChat public account "CITIC SEC Research," edited by GMTEight: Chen Yufeng.