JPMorgan: A-Share AI Pullback Is Deleveraging, Not a Bubble Burst
Alina Collins
JPMorgan said on July 15 that the recent A-share AI selloff is a leverage flush, not a fundamental breakdown — and China's long-term AI investment case remains intact. This means → the bank reads the drop as wringing out excess, not heading for the exit.
How far has deleveraging gone?
Margin-trading turnover in A-share IT stocks fell from a mid-cycle peak of roughly 12% of total volume to 8%–9%. This means → the most aggressively leveraged positions have largely been forced out, and the deleveraging process is essentially complete.
Five-day rolling turnover has pulled back sharply from a cycle high of about 6.5%, yet it remains above the lows seen at prior bull-market troughs. In plain terms = the market is deflating froth, not staging an institutional exodus.
ETF flows confirm the picture: from July 1–13, semiconductor ETFs saw average daily net inflows of RMB 1.2 billion, and tech-hardware ETFs drew RMB 350 million a day — overall flows have turned net-positive since early July.
Compared with past bubbles, are these companies healthy?
Chinese hyperscale cloud operators (Tencent, Alibaba, Baidu) and their US peers (Amazon, Alphabet, Microsoft, Meta) all carry debt-to-asset ratios of roughly 40%.
For contrast: Chinese property developers averaged about 100% leverage at their 2021–2022 peak; some US telecom operators hit roughly 230% during the 2001–2002 dot-com bust. In plain terms = today's cloud giants carry less than half the debt load of China's property sector at its worst — let alone dot-com-era extremes.
On credit ratings, US cloud majors sit between "A+" and "AAA"; Chinese cloud majors between "BBB+" and "A+" — all investment grade. This reflects a fundamentally different balance-sheet profile from the run-up to past bubble collapses.
Can AI spending be sustained? What underpins it?
JPMorgan views the continuous iteration of large language models — the core technology enabling AI to understand and generate human language — as the primary support for sustainable AI capex. China's leading models are closing in on US frontier-model performance as of mid-2026.
Each stronger generation unlocks new enterprise and consumer use cases → drives incremental infrastructure spending → accelerates recurring revenue across the AI sector. In plain terms = better models do more, earn more, and justify more spending — a self-reinforcing loop.
Hardware supply constraints add a second pillar: TSMC's Arizona Fab 2 enters 3 nm mass production in H2 2027; SK Hynix's CEO has warned of the worst-ever memory shortage in 2027; large-scale global supply of HBM — high-bandwidth memory, a critical AI-chip component — is not expected until full-year 2028. This means → key components will stay tight until at least late 2027 to 2028, and the scarcity actually strengthens pricing power and urgency for AI infrastructure spending.
What is JPMorgan's market outlook?
The bank holds its year-end 2026 targets unchanged: MSCI China index base case at 100 points, CSI 300 index base case at 5,200 points. It recommends staying with quality large-cap AI names through short-term volatility.
Near-term, the July rotation from AI into non-AI sectors may persist; earnings season for brokerages, insurers, and healthcare could provide a tactical catalyst.
The key test: August earnings season. Whether Chinese AI companies can reclaim market leadership on the back of Q2 results and H2 guidance will be the critical checkpoint for the thesis that this pullback was merely a leverage flush.
Content is for reference only, not financial advice.