AI’s global architecture has settled into a duopoly: closed models from a handful of American firms on one side, Chinese open-weight labs backed by the State and industrial priorities on the other. At the BRICS+ summit, Chinese President Xi Jinping proposed that member-States establish open-source AI with pooled compute, shared foundation models, and joint training programmes. Xi’s pitch is compelling. Closed models are rented through the labs that built them, on pricing and terms those labs control, and their weights — model weights being the recipe for the intelligence — are hidden and proprietary. Open-weight models publish their trained parameters; anyone with sufficient compute can download and adapt them for local use. The first arrangement is akin to renting, the second to owning but with higher infrastructure costs. The second matters most for the Global South, whose languages, public services, and research Focus are not the commercial priorities of a few Californian firms.
