NScale’s Norway AI Data Center Provided Spring Access to 2,304 Nvidia B200 GPUs
DataNews Editorial Team
NScale’s IPO materials indicate that Spring (SG) Pte Ltd, which S1 links to ByteDance, accessed 2,304 Nvidia B200 GPUs through NScale’s AI data center in Glomfjord, Norway. The arrangement used compute capacity in Norway rather than a direct shipment of the accelerators to China.
- Spring accessed 2,304 Nvidia B200 GPUs through NScale’s Glomfjord, Norway, AI data center.
- S1 links Spring (SG) Pte Ltd to ByteDance; the client was not directly named in the described NScale S-1 disclosure.
- One customer accounted for about 73% of NScale’s 2025 revenue, estimated at approximately $24 million.
- Glomfjord has about 30 MW of operating energy capacity and potential expansion to 60 MW.
- Macquarie’s $105 million loan included monitoring requirements related to Spring’s GPU use and export-control concerns.
NScale disclosed substantial customer concentration in its IPO materials: one customer contributed about 73% of its 2025 revenue, or approximately $24 million. According to S1, citing the Financial Times and related documents, that customer was Spring (SG) Pte Ltd, an entity linked to ByteDance.
Spring accessed 2,304 Nvidia B200 GPUs through NScale’s Glomfjord data center in Norway. Agreements for GPU-infrastructure services were signed between November 2024 and March 2026. The structure did not involve a direct delivery of B200 GPUs to China; the compute resources were used from Norway.
The Glomfjord facility was converted from a cryptocurrency-mining site into an AI data center. It has approximately 30 MW of operating energy capacity, with potential expansion to 60 MW. The 60 MW figure is prospective capacity rather than current operational capacity.
Macquarie extended NScale a $105 million loan, part of which was intended for the procurement of advanced Nvidia GPUs. S1 reports that the financing required NScale to monitor Spring’s GPU use for anomalous or suspicious configurations and possible export-control concerns.
For the data-center market, the arrangement suggests that GPU-dense capacity in jurisdictions with access to advanced equipment could support demand from international customers seeking remote compute access. That opportunity may, however, require operators and lenders to apply closer scrutiny to customer identity, workload configurations and end use. NScale’s early reliance on a single customer also illustrates the financial exposure that customer concentration can create for rapidly scaling AI-infrastructure platforms.
The arrangement illustrates how overseas AI data centers can provide cross-border access to advanced GPU compute without direct hardware exports. It also highlights the growing importance of workload monitoring, end-use compliance and customer-concentration risk for AI-infrastructure operators and their lenders.
- TechNews 科技新報 / ChineseSource →
This article was prepared by the DataNews editorial team based on the sources listed above.