CICC Insight said the recent surge in overseas AI capital expenditure has continued to spill over into China through the global AI supply chain. The firm said data center and AI infrastructure buildouts abroad have supported Chinese exports of servers, optical modules, printed circuit boards, communications equipment and related components, while stronger overseas orders have also improved revenue expectations for domestic AI supply-chain companies and encouraged capacity expansion and fixed-asset investment.
The report focuses mainly on changes in U.S. AI capital expenditure. Looking ahead to 2027, market consensus still points to growth in overseas AI capex in absolute terms, though the year-over-year pace may retreat from the elevated level seen in 2026. CICC said any moderation would affect China mainly through export and investment channels.
Overseas AI capex is still expanding, but growth may cool from 26Q4
According to the report, capital expenditure by five representative cloud companies — Amazon, Alphabet, Microsoft, Meta and Oracle — rose 86.5% year over year in the second quarter of 2026, extending a period of rapid expansion. Based on FactSet consensus estimates, overseas AI capex led by cloud providers is still expected to increase, but its year-over-year growth rate may slow from the fourth quarter of 2026.
CICC said the likely reasons go beyond a high comparison base. The report highlights financing constraints, physical bottlenecks and AI safety governance risks as factors that are becoming more visible.
Financing pressure is building
Over the past two years, cloud companies have expanded capital expenditure quickly, putting pressure on free cash flow and increasing reliance on external financing. With long-end interest rates staying relatively high and credit spreads widening for some cloud providers, higher funding costs could raise the return threshold for new investment.
At the same time, the large stock of capital built up in earlier rounds of spending implies heavier depreciation and amortization pressure ahead. With AI commercialization still needing to prove itself further, the gap between return on invested capital, or ROIC, and weighted average cost of capital, or WACC, has narrowed. CICC said capital allocation may shift away from the earlier push to secure computing power and scale, and toward closer scrutiny of investment returns, slowing the pace of AI capex growth.
Physical limits are affecting project rollout
The report said data center expansion is placing greater demands on grid capacity, electricity prices, water resources and land. Since 2025, disputes in some parts of the United States over data center power use, water use and community costs have intensified, leading to tighter constraints through suspended approvals and higher entry requirements. That has forced some data center expansion plans to slow.
In CICC’s view, overseas AI infrastructure growth may face a situation of strong demand but limited project execution.
Safety governance could also restrain spending growth
As AI systems gain the ability to complete more complex tasks autonomously, risk incidents in real-world operation are also increasing, raising the urgency of AI safety governance. The report noted that Anthropic CEO Dario Amodei recently said frontier model capability gains should be moderated to leave more time for third-party evaluation and risk controls.
If similar proposals turn into industry coordination or regulatory requirements, training cycles for large frontier models could lengthen, placing a marginal constraint on growth in demand for training compute, the report said.
Export channel: slower overseas AI capex may reduce export support for GDP
CICC said the most direct effect of slower overseas AI capex would come through exports. Building data centers abroad requires large volumes of servers, optical modules, PCBs, networking equipment and electronic components, and Chinese suppliers have strong participation across those segments. Because of that, the latest upswing in overseas AI investment has moved broadly in step with an improvement in Chinese exports.
Using quarterly data, the report found that growth in overseas AI capex leads China’s exports and AI supply-chain product exports by about one quarter. Its estimates show that every 10-percentage-point increase in overseas AI capex growth corresponds to about a 2.8-percentage-point increase in year-over-year growth of China’s AI supply-chain product exports, implying a transmission coefficient of about 0.28.
Under current consensus estimates, year-over-year AI capex growth at major cloud companies may fall from 89.9% in 2026 to 38.6% in 2027, a decline of about 51.3 percentage points. On that basis, growth in China’s AI supply-chain product exports may remain positive, but could slow at the margin by around 10 percentage points.
The report added that overseas AI capex is still expected to post positive growth next year, which means it should continue to support China’s AI-related exports, though the contribution may be smaller than this year.
Investment channel: the effect of softer external demand may arrive later
Beyond exports, overseas AI demand can also affect domestic AI supply-chain investment by lifting orders and capacity utilization at Chinese companies. CICC said the impact of overseas AI capex on domestic investment is essentially an investment effect induced by external demand.
Using data from A-share listed companies in the AI supply chain, and weighting fixed-asset investment by the share of overseas revenue, the report constructed an investment indicator linked to overseas demand. Its estimates show a longer transmission lag from overseas AI capex to domestic AI supply-chain investment, with overseas AI capex leading that investment by about one year. The transmission coefficient is about 0.49.
CICC said that fits the economic logic that companies need time to move from receiving overseas orders and identifying capacity gaps to expanding plants and purchasing equipment.
In terms of the marginal effect on economic growth, the report said the support from strong overseas AI capex growth in 2026 may still carry into domestic investment in 2027 because of that lag. As the slowdown in 2027 overseas AI capex is gradually transmitted, its contribution to domestic investment may decline by 2028 compared with earlier periods.
CICC’s reading: exports may feel it first in 2027, investment later in 2028
The report said the effects of slower overseas AI capex on China would not appear at the same time across channels. Exports tend to respond faster, while investment reacts with a lag.
Based on its estimates, the main impact in 2027 may first show up as weaker export support, while capacity expansion driven by earlier overseas orders may still provide a buffer. By 2028, the lagged effect through the investment channel may start to become more visible.
CICC also said its estimates are based on a static partial-equilibrium analysis designed mainly to identify the effect of changes in overseas AI demand. If domestic AI capex continues to grow quickly — especially spending by Chinese cloud providers, computing infrastructure projects and investment tied to domestic substitution — that could offset part of the drag from softer overseas demand. The report also said that if AI penetration in business operations keeps rising, using cloud providers alone to estimate domestic AI investment may understate the picture, and actual resilience in AI supply-chain investment could be stronger than the estimates derived only from changes in overseas demand.