Chinese laboratories are reportedly repurposing US-designed graphics processing units for artificial intelligence applications, according to recent reports. These efforts involve converting discontinued GeForce RTX 5090 cards into AI accelerators by transplanting their GPUs and memory onto custom circuit boards. This practice highlights the continued reliance on American hardware for China's expanding AI infrastructure.
The repurposing of RTX 5090 cards suggests that NVIDIA's Blackwell-series hardware is still reaching China despite export restrictions. Technicians in Chinese facilities are removing the GPU chip and 32GB VRAM from these cards to deploy them on custom boards tailored for AI servers. This process has been observed in large inventories of RTX 5090 GPUs appearing in Chinese factories.
Such activities raise questions about the effectiveness of current export controls and the potential indirect role of companies like NVIDIA in facilitating access to advanced computing resources. The situation underscores the challenges in restricting high-performance hardware for AI development, even as geopolitical tensions persist.
This trend aligns with broader concerns about China's growing dependence on foreign technology for its AI ambitions. Reports indicate that China's access to AI computing power is expected to double in 2027, with US-designed chips accounting for about 65 percent of that capacity. This continued reliance on foreign hardware highlights the complex interplay between technological advancement and international trade policies.
