China's sellers: AI hardware may experience high volatility, rather than confirming a crash

Semiconductor analyst Jukan shares a commentary translated into English by him, authored by a salesperson from a local Chinese sell-side institution. The individual states: In mid to late September, hardware quickly transitioned into a downward wave, with China and South Korea adjusting first, experiencing deeper declines and longer durations, followed by Japan, Taiwan, and the semiconductor ETF SOXX weakening subsequently; there is significant divergence in the market regarding whether AI is entering a mid-cycle collapse, but there is less controversy about the industry still advancing.

The individual summarizes the views of interviewed investors: High long-term interest rates in the U.S., capital expenditures beginning to rely on external financing and competing for funds in the market, multi-layered leverage accumulation, and tight funding are the main reasons for this round of decline, with companies like CoreWeave experiencing a noticeable deterioration in financing structure and cost burden. Model invocation prices are decreasing, but capital expenditures and supply chain costs are difficult to reduce quickly, making short-term investment returns hard to realize. ASML and TSMC's guidance this season is strong, but near-term pressures cannot be ruled out; the bullwhip effect and differences in value chain levels may cause volatility to be greater than expected.

The individual also mentions that use cases beyond coding are emerging in scenarios such as SaaS and AI drug development, which may gradually alleviate investment return concerns and expand demand. His personal judgment: Before more evidence emerges, the current situation resembles a high-volatility phase rather than a collapse that has already begun; the decline may not yet be over, and hardware leadership may give way to applications. Even if control of both houses changes in the midterm elections, the impact may mainly be limited to constraints on certain data center projects, and the U.S. earnings season needs to verify the performance of large technology companies.