Export controls have spent three years pointing one way. Washington restricted the sale of advanced chips, then chipmaking equipment, then the tools used to design them. Now Beijing is preparing a list of its own covering key technologies in artificial intelligence and solar manufacturing, according to a report from Golem. The precise scope is not settled and no decree has been published, but the direction is clear: the licence list is being turned around.
China did not arrive at this instrument overnight. Gallium and germanium went under export licensing in August 2023. Graphite followed that December. Antimony was added in 2024, and in December of that year the gallium, germanium and antimony routes to the United States were closed outright. The April 2025 rare-earth controls were the ones that broke through: they stalled car factories in Europe and India within weeks. The October 2025 expansion added a de minimis rule that pulled foreign-made goods into the regime whenever they contained even trace amounts of Chinese rare earths. Each step was narrower than the last on paper and wider in practice, because the bureaucracy deciding licences has enormous discretion and no obligation to be quick about it.
AI and solar are a different kind of target from rare earths, and why matters for what comes next. Rare earths are a chokepoint because separation capacity outside China is thin and takes years to build. Solar is not a chokepoint in the same sense — the supply chain is more substitutable — but China's position in it is far more complete. Polysilicon, wafers, cells, modules: the country dominates every stage, and the equipment that makes them is concentrated in a handful of Chinese firms. A control regime aimed at solar equipment or process know-how would not be about withholding a material. It would be about withholding the ability to build a factory.
The AI side is stranger and more revealing. Control over model weights is difficult to enforce in a world of open-weight releases and downloadable checkpoints. What is enforceable is the physical and organisational layer: the chips, the interconnects, the packaging, and increasingly the engineering know-how and tooling that turn a research result into a production system. A technology-export catalogue in the Chinese sense is largely about technology transfer and intellectual property — who may take know-how out of the country, and under what conditions — rather than about goods crossing a border. That is a slower, quieter instrument than a rare-earth ban, and it tends to be aimed at joint ventures, overseas subsidiaries and engineers rather than at shipping containers.
The counterargument is the one that has now been borne out twice. Every control accelerates the diversification it was designed to prevent. The 2025 rare-earth shock did not teach the world to stop needing Chinese magnets; it triggered recycling programmes, non-Chinese separation capacity, stockpiling and a wave of substitution research that would otherwise have waited another decade for funding. A solar equipment regime would do the same to thin-film and non-Chinese module supply chains, at precisely the moment those chains are finally being subsidised into existence. And an AI technology regime would push third countries — the ones currently choosing between Chinese open-weight models and American closed ones — toward the American stack, or toward building their own.
That tension is the thing to watch, not the headline. Three questions decide how much this matters. First, is this a goods regime with licences and quotas, or a technology catalogue governing transfer and know-how? The second is enforceable in weeks; the first takes years and mostly produces press releases. Second, does solar mean equipment, or just process documentation? Equipment is the version with teeth. Third, does it land before or after the next American round — because a tit-for-tat spiral is a different story from a unilateral squeeze.
For developers and hardware planners outside China, the practical consequence is the one the rare-earth shock already taught: single-source assumptions are a schedule risk, not a procurement detail. The list is not published yet. The pattern behind it is.
