Geopolitical export controls compel firms like Alibaba and ByteDance to relocate the training of LLMs, such as Qwen and Doubao, to overseas data centers to secure access to Nvidia GPUs, while broader compute supply chains are tightening, demonstrated by Nvidia reportedly stopping the supply of VRAM to GPU partners and Google restricting free access to Gemini 3 Pro features due to high demand.

Despite these constraints, AI capabilities continue to rapidly advance, exemplified by DeepSeek-Math-V2 achieving gold-level math performance and the development of 100B+ Mixture-of-Experts models like INTELLECT-3, even as the industry navigates security risks, such as the exposed metadata of OpenAI API users following the Mixpanel breach, and the legally unsettled status of applying the GPL license to trained models.