A regulatory filing has shed new light on how ByteDance, the Chinese technology giant behind TikTok, secured access to more than 2,000 Nvidia B200 GPUs through a data center operated by UK-based neocloud provider Nscale in Norway. The disclosure underscores the increasingly complex pathways that Chinese firms are using to obtain advanced AI hardware amid tightening export controls.
The filing reveals that Spring, a Singaporean subsidiary, contributed 73 percent of Nscale's 2025 revenue, highlighting the outsized role that a single client entity played in the neocloud company's financial performance. The arrangement effectively allowed ByteDance to circumvent direct procurement restrictions by routing its compute needs through a European data center operator with a Singapore-based intermediary.
The Nvidia B200, part of the Blackwell generation of AI accelerators, represents some of the most sought-after silicon in the industry, capable of training and running large-scale AI models. Access to over 2,000 of these chips represents a substantial compute footprint, one that could support significant AI research and model development workloads. The scale of the deployment signals that ByteDance is continuing to invest heavily in AI infrastructure despite geopolitical headwinds.
The structure of the deal raises fresh questions about the effectiveness of export control regimes designed to limit Chinese access to cutting-edge American semiconductor technology. By leveraging a Norwegian data center and a Singaporean subsidiary, the arrangement illustrates how global cloud infrastructure can be used to obscure the ultimate end-user of advanced chips. Industry observers note that such arrangements are likely to draw scrutiny from regulators in both the United States and Europe.
For Nscale, the revenue concentration tied to Spring represents both a significant endorsement of its neocloud capabilities and a potential risk, given the geopolitical sensitivities involved. The company has positioned itself as a provider of high-performance AI compute across multiple regions, and the Norway facility appears to have served as a key node in this particular arrangement.
The disclosure is likely to intensify debate over the adequacy of current export controls and the role of third-country data centers in the global AI supply chain. As demand for advanced AI compute continues to surge, the lines between commercial cloud services and strategic technology transfer are becoming increasingly blurred.
