What happened

At a briefing on 19 August 2026, Chinese foreign ministry spokesperson Lin Jian opposed dividing countries into rival AI camps and argued that countries should be able to choose their partners. Reuters recorded the remarks and reported them that day.

The response followed reporting about a proposed US message to countries participating in AI cooperation initiatives. It was a diplomatic statement, not a new rule governing every company’s software purchases.

The new development was China’s public response. For executives, it adds a reason to examine whether the flexibility promised in an AI strategy exists in practice.

Why it matters

Choice at the point of purchase is different from choice after a system becomes embedded in daily work. A business might begin with several possible suppliers and gradually build a process that only one can support.

That dependence can sit in unexpected places. Perhaps the source documents remain portable, but the workflow depends on a proprietary search index. Perhaps the application can call another model, but its quality checks were designed around one provider’s behaviour. Perhaps staff know how to operate the current interface but lack a documented alternative.

An executive review should therefore ask for demonstrations. Export a sample of the organisation’s own information. Reproduce one useful task through another approved route. Identify the work needed to restore acceptable quality.

The objective is to understand the switching cost before anyone is forced to pay it.

The bigger shift

I would distinguish three forms of control in an AI operation.

First, control over information: the organisation knows where its material sits, who can access it and how it can be recovered or deleted.

Second, control over the process: staff understand the steps, the review points and the conditions under which work should stop.

Third, control over the supplier relationship: a named owner tracks contractual changes, service limits and issues requiring specialist advice.

No single hosting choice settles all three. Running software locally can still involve external updates, components or support. Using a cloud service can still permit substantial operational flexibility. The relevant question is which dependencies exist and whether they are acceptable for the work.

A review of AI strategy before tool selection helps keep that assessment tied to business priorities rather than national labels alone.

My take

I would resist presenting digital sovereignty as a product feature that procurement can buy and then forget. For a business, practical autonomy is something to demonstrate repeatedly.

Choose one important workflow and organise a short continuity exercise. Give the team a plausible scenario, such as temporary loss of a supplier account. Ask them to identify essential records, restore access through an approved alternative and explain the compromises.

Record what the exercise reveals: missing documentation, inaccessible evaluation data, unclear ownership or a manual method that nobody has used recently. Fix the most consequential gap before adding more integrations.

The diplomatic debate may develop in several directions. A company does not need to predict the outcome to improve its own preparedness. It needs to know which decisions it still controls, which depend on others and what it would take to change course.