The China Question Western Boards Can No Longer Avoid
On 10 June 2026, Nikkei Asia reported that American companies are doubling down on China as profitability rebounds, even as the slowing Chinese economy continues to weigh on sentiment and the political weather in Washington hardens against Beijing. Two days earlier, the South China Morning Post had carried an opinion piece arguing that China-UK cooperation is vital if the AI era is to put people first. Read together, the two items describe a wedge that has opened between the operating decisions of Western corporations and the public posture of Western governments — and they describe, almost in passing, where the next phase of the technology contest will actually be fought.
The argument this piece advances is straightforward. Profit is pulling Western firms deeper into the Chinese market at exactly the moment the political centre of gravity in Washington and several European capitals is moving toward decoupling. The contest over artificial-intelligence governance, export controls and standards is being conducted on terrain that those political positions cannot fully occupy, because the same firms the politicians want to constrain are also the ones delivering the tax base, the share-price support and, increasingly, the model weights that any future AI regime will have to live with. Boards that thought the China question was a 2023 problem are about to be reminded that it never went away.
The earnings tell a different story than the speeches
Nikkei's reporting, drawing on a sweep of corporate disclosures, found that US-listed multinationals are treating China less as a market to be hedged and more as a market to be harvested. The rebound in profitability is the headline; the underlying detail matters more. Companies that spent the 2023–24 window writing down China exposure, restructuring joint ventures and "China-plus-one" diversifying into Vietnam, Mexico and India are now quietly walking some of that back. Where the unit economics still work, capital is following them in. The political-risk warnings in the same filings are also real — but they read, on closer inspection, like compliance language drafted by counsel, not like strategy.
The pattern is not new. Western retailers, consumer-goods companies and automotive suppliers have spent two decades learning that the Chinese consumer is a margin story the developed-world consumer no longer is. What is new in 2026 is the extension of that pattern into the strategic sectors — chips, cloud, model APIs, electric-vehicle supply chains, biopharma contract manufacturing — that export-control regimes are nominally designed to fence off. The fence is real, but it has more gates than the rhetoric suggests, and corporate counsel have become adept at finding them.
The AI governance question London is trying to open
The South China Morning Post opinion piece, written in the register of a Chinese-policy think tank rather than a partisan outlet, made a pointed argument: that the United Kingdom, with its forthcoming AI Safety Institute work and its convening power in the multilateral system, is a natural partner for Beijing on the governance layer of the AI build-out. The argument is not that the two systems are converging. It is that the alternative — a world in which the United States, the United Kingdom, the European Union and China each build incompatible model registries, evaluation suites and incident-reporting regimes — is materially worse for everyone, including the firms that would rather not choose a side.
That is a position a British trade minister could plausibly hold. It is also, in the present political weather in Washington, a position that cannot be held out loud. The default US reading is that any governance arrangement which gives the Chinese state a seat at the standards table is, by construction, a win for Beijing. The Chinese counter-reading, aired in the SCMP piece, is that a regime designed without Chinese buy-in will simply be routed around by the largest pool of deployed AI users outside the OECD, which is the entire point of the standards exercise in the first place. Both readings have empirical content. Neither dominates.
Why decoupling rhetoric is outrunning decoupling reality
The structural reason the corporate footprint keeps widening is that the underlying economy is not decoupling. Trade in goods between the United States and China fell from its 2022 peak and has been rebuilding since late 2024, with the composition shifting toward semiconductors, lithium-ion cells, pharmaceutical intermediates and capital equipment — the very categories the export-control lists are supposed to constrain. The shift is not invisible; customs data and shipping intelligence have documented it. But the policy debate continues to use 2022 numbers as the baseline, which is how a 2026 policy framework ends up governing a 2026 economy that no longer looks the way the framework assumed.
Three things follow. First, the firms that have done the honest work of building China-specific product roadmaps — not the firms that have issued press releases about diversification, but the ones that have actually re-tooled supply chains — are now sitting on assets their competitors do not have. The 2025–26 round of earnings has been kinder to them than to the brands that talked loudly and acted late. Second, the political cost of being visibly exposed to China is rising, which means boards are now doing a second-pass review that mixes risk, reputation and regulatory exposure in a way the first pass, focused on tariffs alone, did not. Third, the places where the United States and China are most deeply entangled — AI, biotech, advanced manufacturing inputs, critical minerals processing — are exactly the places where the export-control regime is thinnest, because the controls were written for an older industrial map.
What London and Beijing actually agree on, and what they don't
The SCMP argument is not, despite its framing, that China and the UK see eye to eye on AI. They do not. They disagree on data sovereignty, on the legitimacy of state-led compute build-outs, on the role of military end-use in dual-use research, and on whether model weights count as export-controlled technology. What the argument claims is narrower and more tractable: that on the things one would have to do to make AI safe at scale — incident reporting, third-party red-team access, transparency about training data provenance, common evaluation benchmarks — cooperation is feasible and unilateralism is not.
It is worth taking that claim seriously on its merits. The 2023–24 attempt to build an export-control regime around advanced GPUs produced a real dent in Chinese frontier-training capacity and a much larger one in the revenues of the firms that designed the chips. It did not, on the evidence available, produce a regime that could meaningfully govern what happened downstream — the fine-tuning, the distil-and-deploy workflows, the model APIs that route through third-country hosting. Those are the deployment surfaces that any governance regime will actually have to police, and they are exactly the surfaces where Western firms have a continuing commercial interest in being present.
Stakes: who wins if the present trajectory holds
If the corporate doubling-down continues and the political posture does not change, three things happen. American shareholders capture the upside of Chinese consumer recovery and AI demand; American workers in the most exposed sectors continue to absorb the adjustment costs; and the export-control regime becomes a tax on the firms that comply with it, while the firms that route around it — through third-country data centres, through joint-venture structures, through licensing arrangements that put the model weight in someone else's name — collect the rents. That is not a stable equilibrium. It is the equilibrium that produces, on a three-to-five-year horizon, the next round of policy panic.
If, instead, the governance route the SCMP piece gestures at becomes the actual operating model — partial, contested, lopsided, but real — then the export-control regime evolves from a fencing exercise into a negotiating one. The firms that positioned early for Chinese deployment keep their option value. The standards bodies that include Chinese counterparts inherit a legitimacy they would not otherwise have. And the political constituency for decoupling, which exists and is not imaginary, gets a partial answer to the question of how a China-aware industrial policy is supposed to defend workers whose jobs the same policy has not yet replaced.
Neither of those trajectories is the one Washington is currently legislating for. The legislation being drafted assumes a third path in which the Chinese market is large, the Chinese deployment is real, and the Chinese firms remain the suppliers — but the Western firms have stopped participating in any of it. The Nikkei data is the clearest indication yet that the actual boards of the actual firms are not planning for that path, because it is not the path the P&L supports. That gap, between the policy the political class is writing and the strategy the corporate class is executing, is the story of the next eighteen months in US-China economic relations. It is also the terrain on which the AI governance regime will actually be built — or fail to be.
This article is a long read; the daily news file covers earnings and policy moves as they break.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- https://t.me/SCMPNews
- https://t.me/NikkeiAsia
- https://t.me/nikkeiasia