Nvidia’s top-tier DGX B300 AI server, officially priced around $400,000 in the United States, now commands over $1.1 million on China’s black market, reflecting the intense scarcity created by export restrictions. These servers, designed as powerful "AI factories" with enhanced performance capabilities, are prohibited from leaving the US for China, yet smuggling networks continue to supply them at a steep premium.
The DGX B300 is equipped with cutting-edge components, including eight Blackwell Ultra GPUs, Intel Xeon 6776P processors, and over 2TB of GPU memory, delivering twice the performance of its predecessor, the DGX B200. This high-performance configuration makes it highly sought after by Chinese AI developers eager to access advanced computing power, despite import bans and the premium price.
Previously, the DGX B300 sold at around 4 million yuan (approximately $589,000), but recent reports show prices doubling to nearly 8 million yuan (over $1.1 million). Other Nvidia hardware affected by export controls also sees inflated resale costs. For instance, RTX 6000 Pro graphics cards, which serve professional AI workloads and are more capable than mainstream gaming GPUs, now fetch about $18,000 each—2.5 times their initial market price.
The US government has tightened restrictions on the sale of next-generation AI chips and systems to China, limiting access to Nvidia’s newer Blackwell platforms while allowing sales of older H200 Hopper GPUs under controlled conditions. Concurrently, enforcement efforts have targeted the illegal movement of restricted technology, although these measures have not fully curbed demand or supply on the black market.
Chinese companies continue to source Nvidia gear through unofficial channels, paying hefty markups to obtain the latest hardware. Alongside this, firms are acquiring older A100 AI systems and repurposing gaming GPUs to bolster AI processing capabilities. This activity persists despite China's domestic policies promoting homegrown chip manufacturers and mandating minimum usage quotas for locally produced inference chips within AI companies.
This push for indigenous technology has helped fuel the growth of companies like Huawei and Cambricon Technologies. While these firms cannot yet match Nvidia’s cutting-edge performance, they have made notable strides in efficiency and performance improvements, offering viable alternatives amid restricted access to American AI hardware.