Summary of Key Points
The United States previously banned the direct export of NVIDIA’s top-tier chips to China, but Chinese companies have circumvented this ban by using the computing power of these chips remotely through overseas data centers in Southeast Asia and other regions. Now, the U.S. is attempting to regulate remote access to computing power through the Remote Access Security Act (RASA), but it faces significant implementation challenges. Chinese companies obtain advanced computing power through three main pathways: direct import of lower-end chips, remote access from Southeast Asia, and domestic alternatives. The surge in data center infrastructure in Southeast Asia has provided a critical support for remote access. The crux of this situation is that U.S. regulations targeting “physical chips” are no longer keeping up with the business trend of “computing power as a service.” China needs to accelerate the development of its own computing power infrastructure in the long term.
1. Why can remote use of computing power bypass chip bans? – Regulations focus on ownership, not usage
The U.S.’s original export control rules only restricted the sale of chips to Chinese companies (ensuring they owned the chips), but not the remote use of computing power from other sources (similar to renting computers from internet cafes without purchasing them). For example, after the release of the AI model Kimi K3 by Moonshot, the White House accused it of using NVIDIA’s GB300 chips through facilities in Thailand, exploiting the loophole that allows control over ownership but not usage. Industry experts point out that many Chinese AI models have made rapid progress by renting computing power overseas, indicating that such loopholes weaken the effectiveness of U.S. bans.
2. The three ways Chinese companies obtain computing power: each with its own purpose
- Direct import: Only lower-end chips can be legally imported. Although the route has not been completely blocked, it has been narrowed, limiting the amount of computing power available.
- Remote access from Southeast Asia: Companies like ByteDance and Alibaba rent computing power through cloud service providers in Thailand and Malaysia. For instance, ByteDance collaborates with Aolani in Singapore to use computing power in Malaysia, and the service provider emphasizes that customers do not own the chips, complying with regulations.
- Domestic alternatives: Local computing power solutions such as Huawei’s Ascend series are expanding. Although their performance may not match NVIDIA’s top-tier chips, stricter external regulations increase the urgency for independent development, which is crucial for future autonomy.
3. Why has Southeast Asia become a “computing power hub”? Infrastructure development provides the foundation
Southeast Asia is rapidly building data centers, meeting the demand and supply needs: China needs computing power, while the region seeks investment and market opportunities. Countries like Malaysia, Thailand, and Indonesia have planned 31 large-scale data centers (with over 100 megawatts of capacity), of which only two are currently in operation, with more to come. Chip manufacturers like NVIDIA are also establishing presence in the region, making it a viable hub for remote access.
4. The challenges in implementing the RASA Act
The U.S. House of Representatives has passed the RASA Act to regulate remote access to computing power, but implementation faces three major issues:
- Industry resistance: Cloud service providers must conduct customer identity verification (KYC), which increases costs and is unlikely to be willingly complied with.
- Implementation difficulties: Computing power in the cloud does not have the same physical clearance processes as chips, making it hard to verify who is using it remotely.
- Policy lag: The act has not yet been approved by the Senate, and even if it is, specific regulations need to be established, creating a gap between policy and reality.
5. Implications for China: Seek alternatives in the short term, rely on autonomy in the long term
In the short term, using multiple pathways to obtain computing power (such as remote access from Southeast Asia) is a practical option. However, in the long run, China must accelerate the development of its own computing power infrastructure to avoid being constrained by the U.S. Additionally, the regulation of cross-border computing power services has become a new issue for Chinese policymakers. The future competition in computing power will not only be about technology but also about the rules that govern it.