August 14, 2026, (Inside AI) — Washington’s newest trade restrictions on China’s AI sector are unlikely to derail Beijing’s export machine, according to a fresh analysis from Citi Research. But one measure, limiting access to open-weight models, could reset global competition in ways no one fully anticipates.
The report lands just weeks before an expected leadership summit between the two powers. Its core finding: direct US-China AI trade has already shrunk so much that further curbs have limited first-order impact. The real risk sits elsewhere.
Analysts led by Yu Xiangrong in Hong Kong wrote that tariffs and export controls were “unlikely to materially disrupt China’s AI exports” because direct trade was “already suppressed at the aggregate level, limiting the potential first-order impact of further restrictions.”
Even a full decoupling, an extreme scenario, would put at most 9.2 percent of China’s exports at risk. That figure is “manageable,” the analysts argued, noting Beijing absorbed former President Donald Trump’s “Liberation Day” tariffs last year even though the US accounted for 14.7 percent of total exports at the end of 2024.
Bilateral AI trade has already largely decoupled. Chinese AI-related exports to the US fell 4.1 percent year on year between January and June, shaving 0.5 percentage points off overall export growth. Yet China remains an indirect beneficiary of the global AI boom through component sales and third-market channels.
Open-weight restrictions are the real wild card
The latest US action centers on drones. On Thursday, Trump signed a proclamation imposing tariffs of up to 100 percent on certain imported drones and components, citing national security and cybersecurity risks. That move extends a pattern of targeted tech curbs rather than broad economic warfare.
But Citi analysts flagged a different measure as the genuine uncertainty: restrictions on access to open-weight AI models. Unlike export controls on physical chips or finished goods, open-weight models are digital assets that can be downloaded, modified, and redistributed globally. Once released, they are nearly impossible to contain.
Open-weight models such as Meta’s Llama series or Alibaba’s Qwen have become foundational building blocks for developers worldwide. If Washington restricts US companies from releasing such models or limits foreign access, the ripple effects could hit innovation ecosystems far beyond China.
That would mark a shift from controlling hardware to controlling software distribution. Hardware controls target specific supply chains. Open-weight restrictions would target the collaborative layer of AI development itself, potentially fragmenting the global research community into separate spheres.
China’s indirect AI gains complicate the calculus
China’s direct AI exports to the US may be declining, but the country still profits from the broader global AI expansion. Chinese firms supply components, rare earth materials, and intermediate goods that feed AI infrastructure in other markets.
The Citi note suggests policymakers should distinguish between direct bilateral trade and systemic exposure. A narrow focus on US-China flows misses how deeply Chinese suppliers are embedded in global AI value chains.
Historical precedent supports this view. During the 2018-2019 trade war, Chinese exporters rerouted shipments through Vietnam, Mexico, and other third countries. Similar adaptation is already visible in AI-related trade, with Chinese companies expanding manufacturing footprints in Southeast Asia and Eastern Europe.
The upcoming summit will test whether both sides can agree on guardrails for open-weight models. For now, the market’s verdict is clear: decoupling is painful but survivable. The open-weight question, however, remains unanswered.
