The price of Nvidia’s flagship DGX B300 AI server has more than doubled on China’s gray market, reaching over 8 million yuan, approximately $1.1 million, as tightening US export controls and stepped-up enforcement actions dramatically reduce the supply of smuggled hardware. The price surge, reported by the Financial Times and corroborated by Nikkei Asia, is the most visible indicator yet that the US government’s escalating campaign to restrict China’s access to advanced AI computing infrastructure is beginning to bite in ways that go beyond corporate procurement decisions.
The DGX B300, Nvidia’s most powerful AI training system, is built around the Blackwell Ultra GPU architecture and is not available for sale in China under current export control regulations. Despite this prohibition, a substantial gray market has developed, with hardware routed through third countries, most commonly Singapore, Malaysia, and the United Arab Emirates, before making its way into Chinese data centers. The US Commerce Department’s Bureau of Industry and Security (BIS) has been progressively tightening enforcement of these routes, leading to a significant contraction in supply.
The Anatomy of China’s AI Hardware Black Market
The mechanics of China’s AI hardware gray market are well understood, even if the participants are difficult to prosecute. Hardware is typically purchased by shell companies or complicit intermediaries in jurisdictions that are not subject to US export controls, then re-exported to China through a chain of transactions designed to obscure the ultimate destination. The process adds cost at every step — for logistics, for intermediaries who take a cut, and for the legal and compliance risks participants assume.
As enforcement has intensified, the number of willing intermediaries has shrunk, and those who remain have raised their prices accordingly. The result is a gray market that is simultaneously more expensive and more difficult to access, a dynamic that is forcing Chinese AI companies to make increasingly difficult choices about how to allocate their hardware budgets.
For the largest Chinese technology companies such as Alibaba, ByteDance, Tencent, and Baidu, the calculus is shifting toward domestic alternatives. These firms have the scale and the engineering talent to deploy Huawei’s Ascend processors and to invest in the software optimization work required to make domestic hardware competitive for their specific workloads. For smaller companies and research institutions that lack these resources, the doubling of gray market prices represents a genuine barrier to accessing the compute they need.
The Enforcement Escalation
The price surge reflects a meaningful escalation in US enforcement activity over the past six months. The Commerce Department has issued a series of new guidance documents and enforcement actions targeting the transshipment networks that have historically supplied China’s gray market. Several Singapore-based trading companies have been added to the Entity List, and BIS has been working with allied governments to coordinate export control enforcement — a development that has significantly reduced the number of safe transshipment routes.
The Netherlands’ decision to join the Pax Silica alliance this week, alongside the US, Japan, South Korea, and several other European nations, is part of this broader effort to build a coordinated multilateral framework for semiconductor export controls. While Pax Silica is primarily focused on equipment rather than finished chips, its expansion signals a growing consensus among Western governments that unilateral US controls are insufficient and that a coordinated approach is necessary to achieve the desired effect.
The Demand Side: Why Chinese Companies Still Pay the Premium
The persistence of the gray market, even at dramatically elevated prices, is a testament to the undiminished demand for Nvidia’s hardware among Chinese AI developers. Despite the rapid progress of domestic alternatives, Nvidia’s CUDA software ecosystem remains the industry standard for AI development, and the performance gap between the DGX B300 and the best available domestic alternatives, while narrowing, has not yet closed entirely.
For Chinese AI companies that are competing in global markets or developing models intended for international deployment, the ability to train on the same hardware as their Western competitors carries both technical and reputational value. The willingness to pay $1.1 million for a server that retails for roughly $500,000 in the US market reflects the strategic importance that these companies attach to maintaining access to frontier compute.
This dynamic is unlikely to resolve quickly. The VanEck SMHC China Semiconductor ETF launched this week to give international investors exposure to China’s domestic chip build-out, reflecting growing confidence that the domestic ecosystem will eventually close the gap. But “eventually” is not “now,” and in the interim, the gray market will continue to serve as a pressure valve for Chinese companies that cannot afford to wait.
The Geopolitical Feedback Loop
The DGX B300 price surge illustrates a broader dynamic that has come to define the US-China technology competition: every escalation in export controls produces a countervailing response that partially offsets its intended effect. The restrictions accelerate domestic Chinese investment in semiconductor capabilities, raise the cost of hardware for Chinese companies, and enrich the intermediaries who facilitate circumvention.
None of this means that export controls are ineffective, the evidence suggests they are imposing real costs on China’s AI development. But it does mean that the relationship between policy intent and policy outcome is more complex than a simple linear model would suggest. As the price of smuggled Nvidia hardware continues to climb, the pressure on Chinese companies to accelerate the transition to domestic alternatives will intensify, and the timeline for that transition may be shorter than either side currently anticipates.