The Meeting You Can't Get: China's Two-Track AI Diplomacy

Earlier this month, a bipartisan delegation from the US-China Economic and Security Review Commission traveled to Beijing, Hangzhou and Shanghai — the commission's first such trip in seven years. Government officials received them. Think tanks received them. China's leading AI companies did not. "We were not able to secure many of the meetings we sought, including with some of China's leading technology companies," vice-chairman Mike Kuiken said afterward, "and that is itself a data point."

It is worth decoding that data point, because the same July produced a very different picture. From July 17 to 20, Shanghai hosted the World AI Conference and the launch of the World Organization for AI Cooperation (WAICO), a 29-member intergovernmental body headquartered in the city. On July 23, APEC digital and AI ministers gathered in Chengdu and adopted a joint statement on AI cooperation, agreeing to reconvene in Vietnam in 2027. Beijing is not withdrawing from AI diplomacy. It is running two tracks — and deciding, meeting by meeting, who rides on which.

Not decoupling — sorting

The reflexive Western reading of the snub is that China is closing up. The evidence points somewhere more precise: China is sorting its interlocutors by function. The USCC is not a trade delegation; it is a congressionally chartered body whose annual reports have supplied the intellectual scaffolding for a decade of export controls and entity listings. Chinese analysts quoted by the South China Morning Post were explicit that Beijing views the commission as a policy actor tied to restrictions, not a neutral fact-finder.

Seen that way, granting DeepSeek or Alibaba meetings to the commission would have been volunteering evidence for the next sanctions package. The refusal was not paranoia; it was learned behavior. In the same month, Washington was investigating Chinese AI firms' access to export-controlled chips, accusing a Chinese company of extracting data from Anthropic's models, and openly discussing blacklistings — while a separate congressional probe, reported by CNBC on July 8, examined the growing use of Chinese AI models inside American companies. Beijing, for its part, is weighing counter-restrictions of its own, including curbs on Western acquisitions of Chinese firms.

The hospitality track

Contrast the treatment of everyone else. The Chengdu APEC ministerial gave middle powers — Vietnam, Malaysia, Indonesia, Korea, Japan, even the United States as an APEC member — a framework document on digital infrastructure, AI in agriculture and healthcare, and "safe, secure, reliable" AI, on Chinese soil and under Chinese chairmanship. WAICO goes further: a standing institution with founding members including Indonesia, Malaysia, Brazil, South Africa, Russia and Pakistan, plus a pledge of 5,000 AI training placements for developing countries over five years.

The design is legible. Washington's model ties frontier AI access to trusted-partner status, security alignment and export-control compliance. Beijing's counter-offer is broad access with institutional membership — open-weight models, capacity building, and a seat at a Shanghai table where nobody asks about your chip imports. For much of the Global South and a good part of ASEAN, the second offer requires fewer concessions than the first.

The visibility problem

The second-order effect of the snub should trouble Washington more than the discourtesy. The US is losing observational access to Chinese AI at precisely the moment Chinese AI is becoming load-bearing in the American economy. Chinese open-weight models now populate US startups' stacks deeply enough that Congress is investigating the dependency. Yet the channels through which American officials might actually understand these labs — site visits, technical exchanges, commission fact-finding — are the ones Beijing is closing first, while keeping leader-level theater open: Xi Jinping's visit to Washington is expected within weeks, and Marco Rubio met Wang Yi in Manila this month, conceding afterward that "great differences" would persist.

That asymmetry is the strategy. China engages at the level where it sets terms — heads of state, multilateral bodies it hosts — and starves the working level where US institutional machinery gathers the granular knowledge that feeds restrictions. America's China-watchers get communiqués instead of lab tours. Meanwhile Chinese firms lose little: the commission was never going to grant them anything, and their growth markets are increasingly the very countries filling seats in Chengdu and Shanghai.

What it means for the rest of Asia

For Asian governments, the two-track structure is simultaneously an opportunity and a trap. The opportunity is obvious: courted by both sides, states like Vietnam — next APEC digital chair — and Malaysia and Indonesia — WAICO founders and US-facing data-center hubs — can extract infrastructure, training and investment from both blocs. The trap is that the tracks are diverging. Every institution China builds and every restriction Washington layers on makes the eventual either/or sharper: whose safety standards, whose model registries, whose chips.

The July lesson is that China believes time favors selectivity. With domestic models near the frontier, a captive-scale home market and a widening Global South clientele, Beijing no longer treats access to itself as something to trade for goodwill — it treats access as the product, priced accordingly. A US commission gets nothing; APEC ministers get a statement; WAICO members get training programs; Xi's summit counterpart gets pageantry. The world's second AI superpower has stopped answering the door indiscriminately. Who it opens for, and who it leaves on the step, is now itself the message — and everyone in Asia is reading it.