Rumors are crackling across the Pacific: China’s DeepSeek is allegedly training its next wave of large language models on smuggled Nvidia Blackwell chips, the crown jewels of US AI hardware.
The unverified report claims DeepSeek tapped a constellation of phantom data centers in undisclosed countries, ferrying dismantled Blackwell GPUs into mainland China under the radar. These chips are the most powerful engines currently available for building frontier AI models, and they’re locked behind strict U.S. export controls.
Nvidia has blasted the allegations as unsubstantiated and “far-fetched,” but says it will investigate any credible tip. Still, the accusations land in fertile ground. US prosecutors have already exposed multimillion-dollar smuggling pipelines, including a recent DOJ bust involving more than $160 million in H100 and H200 GPUs, proving that banned chips have been slipping into China for years.
With the black market swelling, Nvidia is reportedly rolling out new location-verification tech to track its GPUs and choke off unauthorized use.
A history of chip smuggling
Tensions between China and the US on chip sales have been taut since October 2022. That month, the US government imposed sweeping export controls designed to slow China’s technological advancement, including denying access to the most powerful chips, especially Nvidia’s A100, that are essential for training cutting-edge AI models.
The severe restrictions, designed to give the US a leg up in the AI race, have forced China to create less-powerful chips specifically for the Chinese market to comply with the rules. However, DeepSeek has achieved world-class models with comparatively less hardware, placing pressure on US hardware manufacturers to justify the cost of their newest high-end chips in terms of performance gains.
Despite trade barriers, US chip manufacturers, including Nvidia and AMD, rely on massive revenue streams from legal sales in China of older chips, which fund the research and development that has since given them the global lead. It is a precarious relationship that risks slowing China at the expense of innovation while inadvertently boosting China’s own branded tech in a push for self-sufficiency.
Splitting the global AI market
The ongoing cat-and-mouse game on chip development has already impacted not only US-China relations but also the trajectory of the AI infrastructure bonanza. Nvidia’s vow to add chip-tracking software to stymie smuggling is set to usher in a new era of digital enforcement, creating an effective logistical and software hurdle for China’s black-market access to newer chips.
Hardware scarcity will inevitably spur China’s own innovation, accelerating breakthroughs in AI training algorithms and model builds, which will lessen dependence on raw chip power. It is a development that paves the way for domestic companies like Huawei to produce competitive chips that may not outpace US manufacturers in power, but most crucially, in cost.
However, the potential impact on the data-center building spree is most potentially stark.
Should hardware access remain uncertain, China is incentivized to build its own AI infrastructure ecosystem in collaboration with friendly nations in Southeast Asia and the Middle East, where the US and Europe are currently trying to gain a foothold. This physical construction of a parallel infrastructure creates an intensely competitive landscape that places the US, China, and their respective allies in direct battle for global market share.
Whether proven or not, the allegations against DeepSeek irrevocably confirm the intense technological war of attrition between Washington and Beijing. For the global tech industry, it is an expensive and logistically complex conflict that could split the global AI market in two, effectively fragmenting the supply chain.
Check out TechRepublic’s coverage for added context on how shifting US export rules could redefine Nvidia’s position in China.
