A friend who is doing AI in China asked how to design the structure well enough for future overseas expansion. I can only strive to provide a relatively good idea, as there is no absolutely safe structure (and I am not a lawyer, so this is not legal advice). The strategy of having a Chinese parent company with overseas subsidiaries, where the focus is on China first and then a complete migration, has already been proven to potentially trigger scrutiny. Therefore, it is best not to be defined as an outflow from the start. This means that doing a Global design from Day 1 is very important. (Manually at @starzq and @Rubywang) What constitutes a Global design? The parent company is directly overseas, core R&D is also overseas, and only non-core parts are placed in the Chinese subsidiary. This way, the initial generation of core algorithms and models occurs overseas, and the Chinese subsidiary is defined as the application layer, engineering, or business execution party. There is no action of moving core technology out of China. Thus, the regulatory narrative is directly different from Manus. Moreover, the IP generated by core R&D is directly overseas. Here, who invented it is actually more important than who owns it. Once the core model is completed in China, transferring it to an overseas entity through IP assignment or licensing is almost stepping on the risk point of export control. The Chinese entity is mainly responsible for application fine-tuning, product logic, and other non-core aspects, which will relatively not trigger scrutiny. The design of teams and talents should also be layered from the beginning, with functional separation. The overseas team is responsible for model design, Agent architecture, and core systems. The Chinese team is responsible for application engineering, customer needs, and local data processing. The two sides have an interface collaboration relationship, with no technical replication. It will definitely be a bit slower to operate, but the compliance flexibility is much greater. This way, if in the future a large American company acquires it, the buyer's due diligence conclusion will not have potential source dispute content. This means that the core IP has not been "transferred out" of China, so there is no history of unlicensed uncertainty. It can be said that in this era, AI entrepreneurship is a geopolitical project from day one. Whether the technology can be developed is secondary; first, we need to determine the system in which this technology is born. Because this determines where the project can circulate in the future. If designed this way, there is no need to discuss whether to go overseas or not, because it is originally a global company with Chinese operations, and moreover, the Chinese part is not the source of the core IP.