Nvidia’s B300 systems fetch over $1 million on China’s underground market

· Source: Dataconomy · Field: Finance & Economics — Economic Analysis & Policy, International Business & Trade, Capital Markets & Investment Management · Depth: Fundamental Awareness, quick

Summary

Nvidia's restricted AI chip prices have more than doubled on China's black market over the past six months, with the DGX B300 server, containing eight B300 GPUs, now fetching over 8 million yuan (approximately \$1.1 million) compared to \$550,000 in the U.S. This surge follows China's January block on H200 chip imports and U.S. Commerce Department's late May closure of loopholes allowing Rubin, Blackwell, and AMD's MI350x chips to reach Chinese firms via offshore subsidiaries. Despite Beijing's efforts to promote domestic alternatives and a crackdown on grey-market smuggling, strong demand for Nvidia hardware persists among Chinese enterprises, leading to severe compute scarcity, evidenced by B300 server rental prices reaching 190,000 yuan monthly. Nvidia's shares fell over 4% to around \$200 amid concerns about its China revenue, as the company has not generated significant H200 sales there despite export license approvals.

Key takeaway

For executives overseeing AI infrastructure, the escalating black market prices for Nvidia chips in China highlight significant supply chain vulnerabilities and geopolitical risks. You should re-evaluate your hardware procurement strategies, considering the potential for further export restrictions and the necessity of diversifying your compute sources. Proactively explore alternative suppliers and domestic options to mitigate future disruptions and ensure continuous access to essential AI capabilities.

Key insights

US export controls and persistent Chinese demand have created a highly inflated black market for Nvidia's restricted AI chips.

Principles

In practice

Topics

Best for: Entrepreneur, Investor, Executive, Policy Maker

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Dataconomy.