While Blackwell is banned from being exported to China, it appears that domestic AI giants have found their workarounds to access NVIDIA's latest technology.
Chinese AI Giants Are More Inclined Towards the 'GPU Rental' Model Over Getting Hopper Chips, Claims Analyst
There's a dire need for computing power among Chinese AI giants. While the nation is making efforts to establish domestic AI chip providers, companies like Tencent are looking for alternative methods to fulfill their needs. One prominent option is using rental compute services, since the U.S. export controls don't have 'strict' boundaries set for such a venture, which is why the Financial Times reports that Tencent now has access to NVIDIA's B200 AI chips, through a Japanese neocloud Datasection. The firm is reportedly holding rental contracts worth more than $1.2 billion from Tencent alone.
Since then, Datasection has gained more than $1.2bn in contracts with one large customer to access a significant portion of its 15,000 Nvidia Blackwell processors. That customer, which has a relationship with Datasection via a third party, is Tencent, according to people familiar with the matter.
The deal allows the Chinese tech giant to use a legal but geopolitically fraught strategy to access advanced AI chips at a time Washington restricts the export of Nvidia’s top hardware to China.
- The Financial Times
Datasection has access to NVIDIA's latest Blackwell AI chips, including both the B200 and the B300 AI chips, which are deployed across multiple facilities, primarily in Japan and Australia. Despite all the efforts, we do know that Chinese AI giants cannot train capable frontier models with the lack of NVIDIA's AI chips, which is why firms like Tencent, Alibaba and Baidu are known to have been utilizing the "rental compute" loophole to gain access to Team Green's latest hardware, which cannot be exported directly to China due to U.S. restrictions.
Interestingly, renting compute power is becoming so common among Chinese companies that FT claims that an analyst at Bernstein Research claims that firms might find it more compelling to continue this arrangement, rather than buying NVIDIA's AI chips in China, since the computing capabilities they get through this means are significantly higher than what's available in the domestic market. Even after the H200 is available to China, the B200 and B300 AI chips outperform the Hopper option by a significant margin, which shows that Beijing isn't devoid of cutting-edge American technology.
About the author: Muhammad Zuhair is a hardware and technology reporter for Wccftech, specializing in the semiconductor industry and the complex interplay between technology, manufacturing, and geopolitics. His coverage focuses on the corporate strategies and technological roadmaps of industry giants like TSMC, NVIDIA, Samsung, and Intel. Zuhair's expertise lies in deconstructing complex topics such as fabrication nodes (e.g., 2nm process), the economic impact of policies like the CHIPS Act, and the strategic development of AI infrastructure from NVIDIA, AMD and Intel.
Follow Wccftech on Google to get more of our news coverage in your feeds.
