Chinese AI developers like DeepSeek and Xiaomi are shipping open-weight models monthly, offering businesses and governments software they can run on their own infrastructure without expensive licensing. In Britain, the consumer instinct is the same as with Chinese EVs: cheap is good, questions come later. But the cost structure of free AI is fundamentally different from the cost structure of cheap cars. The article, by the Guardian's economics team, frames the threat across three layers. First, surveillance: Beijing now uses AI domestically for comprehensive citizen monitoring, freeing resources previously needed to manage social unrest. Second, strategic export: open-weight models shipped globally embed political censorship that persists even when hosted on foreign servers — a finding confirmed by the US National Institute of Standards and Technology (NIST). Third, dependency: cheap AI, like cheap EVs, is designed to hollow out domestic competitors and create client-state dynamics through China's belt and road channels. The NIST evaluation is the hardest data point in the piece. Researchers ran downloaded DeepSeek models on NIST's own isolated infrastructure and found political censorship baked into the weights. This is not a server-side problem that self-hosting fixes. Ross Burley of the Centre for Information Resilience notes that running a model locally "doesn't automatically remove any political censorship. Or indeed repair weaknesses in any security defences." DeepSeek's own policy governing its online chatbot states it collects prompts, processes personal data in China, and permits disclosure to Chinese authorities. Michael Wooldridge, professor of AI at Oxford, states plainly that it is "not unreasonable to assume" data inputted into the chatbot could be shared with the Chinese state. The article draws a parallel to EVs: Beijing subsidises exports deliberately to undermine foreign domestic industries. The AI version of this strategy is more dangerous because the dependency is infrastructural — once businesses build workflows around a model, switching costs compound. The piece argues Britain should reject cheap Chinese AI, better regulate US AI, and support domestic and EU alternatives. This is the correct diagnosis but an incomplete prescription. There is no named domestic or EU model competitive with DeepSeek or OpenAI's offerings. The article's own observation — that Meta's open models and European alternatives "lag in the race" — undercuts the policy recommendation. Telling businesses to choose expensive, inferior domestic options over free, capable foreign ones requires either subsidy, regulation, or a domestic capability that does not yet exist. The military dimension is noted but underexplored: Beijing is redirecting resources toward a Pacific fleet buildup, with Taiwan — the global semiconductor chokepoint — explicitly in its sights. If Chinese AI dependency and Chinese semiconductor supply are two sides of the same coin, the strategic picture is considerably darker than a software licensing question. What the piece gets right is the core mechanism: the extraction is not price gouging but dependency creation. Free is the most expensive price when the supplier is a strategic competitor with surveillance infrastructure and no accountability to your legal system.
