CICC: AI Boom Faces Supply Expansion Test
nashnova research
CICC warns that the AI supply chain's supply-demand gap is narrowing — capex growth at the big five cloud vendors is decelerating while new capacity in optical modules, GPUs, and memory chips comes online, challenging the volume-and-price boom thesis.
AI stocks are diverging — what's behind the split?
After a sharp Q3 correction, Taiwan's TAIEX and the Nasdaq briefly hit new highs, but the Philadelphia Semiconductor Index, STAR 50, and ChiNext 50 lagged visibly — some gave back all year-to-date gains.
This means → the market is no longer bidding up AI as a single trade; it is sorting which links still face shortages and which do not.
CICC attributes the split not just to macro conditions but to quietly shifting supply-demand dynamics at different points along the AI chain.
Why doesn't demand growth guarantee returns?
CICC flags three common cognitive biases among investors. The first is the most important: demand growth does not equal investment returns.
In plain terms = the more certain the outlook, the more capital floods in to expand capacity; once new supply outruns demand, prices and margins fall. This reflects a counterintuitive rule — high certainty can actually compress the return pool.
The second bias: forecasting long-term demand depends on stacked assumptions, but supply-side indicators — capex, capacity under construction, fundraising volumes — are far easier to observe and verify. CICC suggests tracking supply-side shifts to gauge marginal changes in returns.
Does supply expansion always lead to a glut?
CICC's answer: not necessarily — it depends on competitive structure.
In a fragmented market, each firm's individually "rational" expansion decision can aggregate into industry-level irrational investment. In concentrated markets, leaders are more likely to weigh the impact of their own expansion on industry supply and pricing.
In plain terms = the real risk is not that someone is expanding — it is expansion without discipline, where no one is willing to stop first.
How did the current boom form?
Demand side: Anthropic estimates AI assistance can cut completion time on certain tasks by an average of 80%. U.S. nonfarm labor productivity has risen noticeably, AI token call volumes are growing rapidly, and cloud vendors keep raising compute spending.
The supply-side keyword is "capacity constraint" — when utilization nears its ceiling, incremental demand converts more into price increases, and profit leverage amplifies nonlinearly. This means → much of the boom has been driven by scarcity pushing up unit prices and margins, not volume growth alone.
AI spending as a share of enterprise input (excluding payroll) rose from 0.5% in 2023 to 5.7% in 2026 — still low in absolute terms, suggesting the addressable market has room to run.
How can you tell if supply expansion has gone too far?
CICC uses "relative investment intensity" (RI): first estimate a firm's normal investment level based on Tobin's Q, ROE, leverage, earnings volatility, and other factors, then measure how far actual investment deviates from that norm.
In plain terms = take a company's financial "vital signs," calculate how much it *should* be spending on expansion, then compare that to how much it actually spent — the bigger the gap, the higher the risk of overbuilding.
CICC's bottom line: whether the boom lasts hinges on how fast supply expansion erodes the capacity constraint. Which links cannot sustain high growth and which still have room for improving returns — that is the question global investors most need to answer now.
市场有风险，内容仅供研究参考，不构成投资建议。