Over the last week, a Chinese laboratory, Z.ai, has shown it can build Artificial Intelligence systems capable of matching the American frontier models that can write working software, find security flaws in complex code and reason through long technical problems. It can carry out the kinds of expert work that until recently only the most expensive US systems, such as Anthropic’s Mythos, could do reliably. It has done this, moreover, on open weights: making the models freely available to download, run and modify on one’s own machines.
Creating such AI models is not merely a tall order, but among the tallest in modern industry, demanding tens of thousands of advanced chips, robust electricity supply, years of accumulated research talent, and capital running into the billions. America has dominated this work because it has dominated the full stack on which it rests: The chip designs, the manufacturing tools, the cloud infrastructure on which models are trained and served, the small number of laboratories that produce the frontier models, a near-endless supply of capital, and the diplomatic and commercial arrangements through which the world gains access to them.
Yet the American stack, for all its capability, carries the costs of so much concentrated in so few hands. Earlier this month, Washington ordered the worldwide suspension of two of Anthropic’s most powerful models. Anthropic’s rival, OpenAI, has since said it will release its newest models only after consultations with Washington. Against this backdrop, the assumption that the American frontier is the only frontier in AI needed a reality check. China’s new models supply it.
There are, however, considerations with open models too. They cannot be switched off by an order issued in another capital, but they do not by themselves confer sovereignty. They come with serious questions — about hidden behaviours, the data they were trained on, the values their training may encode — that India would be irresponsible not to investigate with care. China’s governance values are not India’s, and engagement with, or avoidance of, technology of Chinese origin cannot be sentimental. But these are problems to be worked out, not reasons to fall back on a single foreign dependence that has just been proven brittle.
China’s Mythos moment, read calmly, is not an invitation to choose between two stacks. It is an argument for building a third — possible only with sustained capital, patient industrial policy, deep talent pipelines, and a willingness to underwrite long-shot research over decades.