Citi: U.S. and China May Play the "Trade Card" Again Before September Summit, AI Exports Become Core Bargaining Focus
Nashnova编辑部
Citi notes both sides are escalating restrictions ahead of the September summit, with the battleground expanding from chips to the entire AI supply chain — AI-linked goods now account for roughly a quarter of China's total exports, and any clampdown on technology access would hit an entire export and manufacturing ecosystem.
Why are both sides escalating again before the summit?
With China's US visit roughly six weeks away, both sides are ramping up restrictions — a pattern highly consistent with previous pre-summit cycles: move first, negotiate later, and grab leverage before sitting down.
US moves: the FCC banned foreign-made advanced robots and grid-tied power inverters; three days later, 43 Chinese entities were added to the UFLPA (Uyghur Forced Labor Prevention Act) list — the largest single expansion since the list was created.
China's response: sanctions on firms providing UFLPA compliance and audit services, plus tighter export controls on drones and other dual-use goods. This means → the broadest round of Chinese trade retaliation since the October 2025 truce.
Citi's takeaway: the friction has spread from chips and semiconductors to robots, drones, power equipment, data centers, and AI models — the main theater is now AI and its infrastructure.
How important are AI exports to China's economy?
Citi classified 203 eight-digit customs codes as "AI-related," spanning upstream semiconductors, midstream data-center construction, and downstream consumer electronics. The result: AI-linked products now account for roughly a quarter of China's total exports.
Growth is striking: this basket grew 45.7% YoY in H1 2026, with June alone at 59.8%, contributing 9.9 percentage points to China's overall 17.6% export growth.
The driver is price, not volume: the June telecom-manufacturing export price index surged 54.2% YoY, far above CPI or PPI. In plain terms = China is not shipping more units — each unit sells for much more.
Profits followed: H1 2026 telecom-manufacturing profits jumped 96.9% YoY, versus just 11.1% for other industrial sectors. This means → AI exports are the core engine of China's export growth and industrial profits, not a side story.
Has direct US-China AI trade already decoupled?
At the direct-trade level, largely yes. China's AI-related exports to the US fell 4.1% YoY in H1 2026, dragging overall export growth by 0.5 percentage points.
China's share of US AI-goods imports dropped to 6.6%, below its 8.4% share in non-AI goods. Excluding consumer electronics, its share in upstream and midstream AI imports is just 5.5% and 4.1%, respectively.
But indirect exposure remains large — China is deeply embedded in regional supply chains that ultimately serve US AI demand. This means → the surface has decoupled, but the pipes underneath are still connected.
Where is the indirect exposure hiding?
Taiwan is a clear example: Taiwan's imports from mainland China rose 37.7% YoY in H1 2026, while exports to the US contributed 19.2 percentage points to Taiwan's own 47.1% overall export growth. In plain terms = mainland Chinese components get processed in Taiwan and flow into the US AI supply chain.
Printed circuit boards (PCBs) make it even more concrete: in 2025, mainland China supplied 66.9% of Taiwan's PCB imports, equivalent to 39.2% of Taiwan's total PCB supply including local production.
Per the OECD's 2022 inter-country input-output tables, Chinese intermediate inputs account for 14.1% of Mexico's electronics output, 12.5% of Vietnam's, and 12.2% of Thailand's — Citi believes these shares have likely risen since.
Third-country data-center construction is another channel: ASEAN economies have data-center capacity under construction equal to 27.8% of operating capacity; China's investment in Asian electronics and IT hit its highest level since 2000 in 2025.
Tariffs versus tech restrictions — which is the real killer?
Citi draws a clear line between two types of risk. Conventional trade friction — tariffs and routine export controls — likely has limited impact: semiconductor supply chains span multiple countries, direct US-China AI trade has already largely decoupled, and Chinese firms have built some overseas production capacity.
Optical modules are a case in point: despite reports that the US plans to restrict Chinese optical-module imports, China's direct exports of these goods to the US account for just 7.8% of its total optical-module exports — a ban would not hit an artery.
Technology restrictions are a different matter. Potential measures include curbing Chinese firms' access to cloud-based AI chips, restricting external access to Chinese open-weight AI models, and further tightening advanced-chip access. This means → tariffs target "volume"; tech restrictions target "capability" — only the latter can reshape the landscape.
What to watch at the September summit?
Citi projects roughly $7.6 trillion in global AI capex from 2026 to 2030 and believes trade friction is unlikely to alter the overall trajectory of AI demand.
But technology and chip-access restrictions are the "true unknown variable." In the near term, tighter restrictions could slow China's model development and overseas data-center expansion, weighing on tech exports.
Over the longer term, sustained denial of US chips and technology may accelerate China's push for a self-sufficient tech stack — domestic AI chips, servers, operating systems, and cloud platforms across the board. This reflects a deeper logic: the harder the restrictions, the more complete the decoupling, and the greater the chance of a fully parallel technology ecosystem emerging.
Whether the September summit can draw a boundary around chip and technology-access restrictions is the market's most critical watch point.
Content is for reference only, not financial advice.