JPMorgan asserts that the recent pullback in the A-share AI sector is fundamentally a process of leverage unwinding rather than a deterioration in fundamentals. The decline in margin financing activity within the IT sector, coupled with net inflows into technology ETFs, suggests the deleveraging is nearing its end. Furthermore, the balance sheets of major Chinese and US cloud giants remain robust, with debt ratios far below levels seen during historical bubble periods. The bank expects constraints on AI hardware supply to persist until at least 2028.
JPMorgan's analysis indicates the correction in China's AI-themed stocks is a healthy adjustment to remove excessive leverage, not a signal of impaired long-term investment logic for the AI ecosystem.
In a report dated July 15th, JPMorgan's China equity strategist, Zhang Xiaoning, explicitly rejected the market narrative of an impending bubble burst. The bank's stance is supported by three key factors: the health of corporate balance sheets, the continuous advancement of large language model capabilities, and the persistence of AI hardware supply bottlenecks. On liquidity, the proportion of margin financing in the A-share IT sector's turnover has dropped from a mid-cycle peak of around 12% to 8%-9%, indicating that the most leveraged positions have largely been forced out and the deleveraging process is largely complete.
Looking ahead, JPMorgan maintains its baseline year-end 2026 targets for the MSCI China Index at 100 points and the CSI 300 Index at 5,200 points. The bank advises investors to hold onto quality large-cap AI stocks during short-term volatility, anticipating that China's AI sector could regain leadership during the August earnings season.
Liquidity Indicators Point to End of Deleveraging
JPMorgan used two key liquidity metrics to assess the nature of the correction.
First, while the 5-day rolling average turnover for A-shares has retreated significantly from a cycle high of around 6.5%, it remains above the lows seen at the bottom of previous bull markets. This pattern aligns with a market "de-frothing" rather than a wholesale institutional exodus. Second, the deleveraging of margin financing is essentially finished. The drop in the IT sector's margin financing share of turnover from ~12% to 8%-9% signifies a large-scale clearing of the most aggressive leveraged positions.
Notably, ETF flows into A-shares turned positive in early July, with semiconductor and tech hardware ETFs seeing the most prominent net inflows. From July 1st to 13th, semiconductor ETFs averaged daily net inflows of 1.2 billion yuan, while tech hardware ETFs averaged 350 million yuan. JPMorgan views this as further evidence of a healthy correction, not the spread of systemic risk.
Balance Sheet Health: Far from Bubble-Era Levels
Citing Bloomberg data, JPMorgan notes that the latest debt-to-asset ratios for China's hyperscale cloud providers (Tencent, Alibaba, Baidu) and their US peers (Amazon, Alphabet, Microsoft, Meta) are around 40%. This is far below the average ~100% leverage seen among Chinese property developers at their 2021-2022 peak and incomparable to the extreme ~230% leverage of telecom operators like Global Crossing and Level 3 Communications during the 2001-2002 US dot-com bubble.
From a credit rating perspective, these hyperscale cloud providers maintain investment-grade ratings: US giants range from 'A+' to 'AAA', while Chinese giants range from 'BBB+' to 'A+'. JPMorgan argues that despite some widening in option-adjusted spreads for certain companies due to recent bond issuance, overall balance sheets remain solid, fundamentally different from financial structures preceding historical bubble bursts.
Technological Evolution and Demand: Upside Potential Remains
JPMorgan sees the continuous iteration of large language model capabilities as a core pillar supporting the sustainability of AI capital expenditure. The report states that China's leading models are striving to catch up to the performance level of leading US models projected for mid-2026. Each new, more capable generation unlocks new enterprise and consumer use cases, driving incremental infrastructure investment.
On the revenue side, annual recurring revenue (ARR) for the broader AI sector is accelerating, and order visibility for China's AI supply chain has significantly improved. JPMorgan believes the current AI investment cycle is not yet constrained by demand saturation. The virtuous cycle of technological progress creating new scenarios, which in turn drive ARR growth, continues, representing an upside option for valuations.
Hardware Bottlenecks: Supply Constraints Until 2028
JPMorgan's detailed review of the global AI hardware production timeline suggests supply constraints for key components are unlikely to ease before late 2027 to 2028 at the earliest, meaning the next two quarters will not be a window for testing the sustainability of AI capex.
Globally, TSMC's Arizona fab is slated for 3nm mass production in the second half of 2027. Micron is advancing HBM memory production in both the US and Singapore, while SK Hynix's CEO has warned of the most severe memory shortage in history in 2027. A large-scale release of global HBM supply and a genuine easing of the shortage is not expected until 2028. JPMorgan argues this constraint actually reinforces the pricing power and urgency of AI infrastructure spending.
Strategic Advice: Hold Quality Large-Caps, Watch for Rotation
Following the correction, forward price-to-sales multiples and valuation differentials across China's AI ecosystem have returned to more reasonable levels. JPMorgan advises continuing to hold quality large-cap AI stocks amid short-term liquidity fluctuations. In the near term, the bank expects the market rotation from AI to non-AI sectors seen in July may continue, potentially catalyzed by the earnings season for sectors like brokers, insurance, and healthcare.
The bank maintains its year-end 2026 index targets. JPMorgan believes China's AI ecosystem will regain market leadership during the August earnings season, supported by Q2 financial results and H2 guidance.
Comments