In-Depth Financial Review Highlights for September 18th

Zhipu Unveils Its Pioneering RSI Breakthrough, Constructing GLM with GLM Leveraging 100,000 Domestic Computing Cards

Zhipu has achieved a significant milestone by releasing its first recursive self-improvement (RSI) accomplishment, successfully training the GLM model using 100,000 domestically produced computing cards. This accomplishment heralds the commencement of a self-sustaining cycle between domestic computing capabilities and models, transitioning AI development from manual optimization to model-driven self-iteration.

Chief Analyst's Insight: RSI (Recursive Self-Improvement) stands as a pivotal concept in contemporary AI discourse, as it signifies models' initiation into the design and training processes of subsequent generations. The speed of iteration is no longer linearly tied to the workforce of engineers. Yet, the truly remarkable aspect lies not in the concept itself but in the prerequisite of 100,000 domestic cards: the central query regarding domestic computing power over the past two years has revolved around the ability to acquire cards that can effectively train high-quality models. Zhipu's success in scaling to 100,000 cards and establishing a self-sustaining cycle provides engineering-level validation of their usability. The long-term implications for the industry chain are profound: on one hand, it will bolster demand for integrated systems encompassing chips, interconnects, storage, and liquid cooling, rather than standalone chip purchases; on the other hand, accelerated model self-iteration suggests that computing power consumption on the inference side may outpace that on the training side, benefiting vendors specializing in inference optimization, optical interconnects, and power supply/distribution.

For the secondary market, such events act as catalysts for short-term sentiment and thematic trading, but prudence is advised due to the potential disparity between conceptual excitement and actual orders. In the medium term, the key metrics to monitor are the actual utilization rates of domestic clusters and the declining trend of unit token costs, which serve as true indicators of the self-sustaining cycle's viability. Today, the Sci-Tech Innovation 50 (STAR 50) surged nearly 4%, with the market voting with its feet, but sustainability hinges on whether leading model companies can replicate this achievement.

Saudi Arabia Accelerates Oil Pipeline Repairs and Seeks Two-Week Ceasefire, International Oil Prices Continue to Decline

Saudi Arabia has initiated extensive repairs on a major oil pipeline damaged by attacks and has increased maritime crude exports through Oman's waters via the Strait of Hormuz. Reports indicate that Saudi Arabia has requested Oman to mediate a two-week ceasefire with the Houthis. WTI October crude futures closed down 0.51% at $101.91 per barrel, while Brent crude futures fell 0.95% to $104.82 per barrel, with WTI briefly dipping below $100 during intraday trading.

Chief Analyst's Insight: Oil prices have been a significant driver of global risk asset volatility. The chain reaction of rising inflation expectations, U.S. bond yields surpassing 5%, and growth stocks facing valuation corrections over the past week can be traced back to concerns over Middle East supply disruptions. Saudi Arabia's concurrent actions to repair pipelines, bypass the Strait of Hormuz, and mediate a ceasefire directly undermine the foundation of the risk premium. For existing analytical frameworks, this represents a critical reversal of key variables: falling oil prices not only reduce inflation expectations but also provide the Federal Reserve with leeway to avoid further rate hikes, benefiting both U.S. Treasuries and growth stocks.

The implications for A-shares are twofold: firstly, at the cost level, lower oil prices benefit sectors such as aviation, transportation, mid-to-downstream chemicals, and utilities, while diminishing profit elasticity for the oil and petrochemical sector. Secondly, at the style level, easing imported inflation pressures allow domestic monetary policy to remain accommodative, reducing the likelihood of simultaneous external and internal tightening. However, the two-week ceasefire is a temporary measure, and tail risks to transportation through the Hormuz and Red Sea persist. Crude oil's high volatility will continue to disrupt markets, so a single-day decline should not be linearly extrapolated as a sustained trend.

AI is Revolutionizing Ad Budget Allocation Logic, Increasing Wallet Share Concentration Among Leading Platforms

Morgan Stanley's survey reveals that 72% of AI users reallocate existing ad budgets rather than increasing total spending, with 74% of advertisers concentrating cross-platform ad expenditures. AI exacerbates platform differentiation: Douyin and Taobao/Tmall lead in share acquisition, while Tencent and Meituan benefit from integrated transaction closures. Baidu and Weibo face structural pressures.

Chief Analyst's Insight: The survey's core conclusion is that AI's short-term impact on the advertising industry is reallocation, not expansion. With total budgets remaining constant but structures undergoing dramatic shifts, the divergence between winners and losers will intensify compared to growth periods. The key variable for value capture is not model capability but commercial control—who owns user transaction intent and outcome data. Three pathways emerge: predictive AI enhances monetization efficiency of existing shelf spaces (stock efficiency improvement); search and Agent interface upgrades reshape candidate sets, directing ad budgets to higher-value interaction entry points; and transaction-layer upgrades to merchant OS represent true growth, converting ad revenue into commissions and recurring income. This explains why platforms like Alibaba and Meituan, with checkout closures and outcome data, hold advantages—their monetization extends beyond impressions to transaction fees.

For A-shares, caution is warranted: pure ad platforms' traffic advantages are diluted by Agent entry points, while e-commerce and local services with transaction/fulfillment closures gain relative advantage. Meanwhile, the blurring of ad-service boundaries in the Agent era creates structural opportunities for companies with merchant-side IT capabilities and data assets. Overall, AI will not uniformly raise all platforms' ceilings but accelerate value concentration toward companies controlling transaction endpoints.

- END -