CICC: The Tipping Point Where AI Turns from "Friend" to "Foe" for Traditional Demand
nashnova research
CICC's latest research report asks a pivotal question: AI is shifting from a growth 'friend' that lifts the broader economy to a 'foe' that crowds out traditional demand — with major US and Chinese cloud firms' free cash flow already turning negative in Q2, pinpointing this tipping point will shape policy and investment in both countries.
How far has AI's resource drain gone?
In Q2, major US cloud firms (excluding Microsoft and Meta) saw free cash flow turn negative — operating cash flow can no longer cover ballooning AI capex, forcing them to raise equity or debt.
This means → AI investment has moved from "spending profits" to "spending borrowed money." US cloud firms' weighted ROIC of 19.9% still clears their 9.6% WACC, so borrowing still makes sense — but the cushion is shrinking.
China is more strained: Tencent, Alibaba, and Baidu's weighted ROIC is roughly 11.9%, barely 60% of US peers, and free cash flow has likewise turned negative. In plain terms = Chinese cloud firms earn less to begin with, have less room to borrow, and face a faster path to crowding out traditional demand.
How do the US and Chinese AI economies differ structurally?
CICC splits the economy into four sectors: AI production (large models, software, data centers), AI infrastructure (power, construction, metals, machinery), AI applications (finance, pharma, media), and other traditional sectors.
In both countries, AI-related and non-AI sectors split roughly 50/50 — but the internal mix diverges sharply: the US AI-production sector is 10.3% of GDP (up 2.4 pp since 2022); China's is under 5%, half the US level.
Conversely, China's AI-infrastructure sector is nearly 14%, about twice the US share. This reflects a structural tilt: China is still "building the roads," while the US is already "driving on them."
When AI is the 'friend,' how much growth does it pull?
In Q2, US real GDP grew at a 2.2% annualized rate; broad AI demand contributed 28% of that, adding roughly 0.6 pp. China's real GDP grew at a seasonally adjusted 4.77% annualized rate, with AI contributing 47% — mostly via exports, where AI-related products accounted for over 60% of net-export growth.
On the earnings side, AI-related sectors drove 53% of US equity earnings growth, versus 19% for A-shares and 2.4% for Hong Kong stocks. This means → more than half of US equity growth rests on AI; if AI investment returns disappoint, the market's foundation shakes.
Wealth effects are far stronger in the US: household equity-like assets reach 47% of balance sheets, and retail/non-institutional investors hold over 42% of US stocks. In China, household equity allocation remains low, muting the wealth channel.
When AI is the 'foe,' what does it crowd out?
Physical resources: global data-center power consumption rose from 1.0% of generation in 2020 to 1.5% in 2025; the US share jumped from 2.6% to 4.9%. Storage prices surged 110% this year, driving the US PPI electronics-components sub-index from 4% y/y at year-end to as high as 29% in June.
Financial resources: massive AI-related issuance plus external supply shocks have pushed up overall borrowing costs — the US 30-year mortgage rate hit 7.5%, small-business short-term loan rates reached 8.2%, and high-yield spreads widened notably. In plain terms = AI firms' heavy bond issuance is bidding up "the price of borrowing," and ordinary homebuyers and small businesses pay higher interest as a result.
CICC argues that AI-driven bond supply was one cause of August's "Treasury storm" — AI is competing not just for power and chips, but for capital in financial markets.
Where is the tipping point, and what should investors watch?
CICC's core thesis: identifying the "critical threshold" at which AI crowds out traditional demand reveals the boundary that economic policy can tolerate.
Three warning signals are flashing simultaneously: rising Treasury yields, CoreWeave risk exposure, and weak Chinese domestic demand — each a different refraction of the same crowding-out dynamic.
This means → the ultimate variable is the K-shaped divergence: whether the AI-linked "upper half" can keep offsetting the traditional-sector "lower half" so that aggregate growth stays positive. This reflects a bigger question — AI's own rate of return is becoming the core variable for judging policy direction and investment positioning in both the US and China.
市场有风险，内容仅供研究参考，不构成投资建议。