Summary
- Nscale's IPO filing shows one customer accounted for 73% of its $33m 2025 revenue.
- Reporting identifies that customer as a ByteDance-linked Singapore entity using 2,304 B200 GPUs at Glomfjord.
- Nscale expects customer concentration to reduce as larger contracts enter service.
Nscale generated 73% of its 2025 revenue from a single customer that the Financial Times has identified as connected to ByteDance, bringing the commercial role of the UK company’s Norwegian AI infrastructure into sharper view as it prepares for a New York listing.
Nscale’s registration statement reports $33m of revenue for 2025 and discloses that one customer accounted for 73% of the total. The main prospectus does not name that customer.
Financial Times reporting links it to Spring (SG) Pte Ltd, a Singapore company associated with ByteDance. Supporting financing documentation identifies a cloud-services contract involving Spring and infrastructure containing 2,304 Nvidia B200 GPUs at Nscale’s Glomfjord data centre in Norway.
The arrangement is also tied to a $105m financing involving Macquarie. The customer contract formed part of the security underpinning capital used for the GPU deployment, illustrating the extent to which long-term compute agreements can be converted into financing for expensive physical infrastructure.
DataCentral has already reported on Nscale’s IPO filing, which shows the company attempting to fund a pipeline measured in gigawatts while its operating business remains capital intensive.
Customer concentration meets infrastructure financing
A customer representing nearly three-quarters of annual revenue would be material for almost any business. It is particularly significant for an AI infrastructure provider because the capacity serving that customer is built from specialised assets with high upfront costs.
GPUs, networking, power equipment, liquid-cooling systems, and facility capacity have to be financed before revenue can be recognised. Providers therefore depend not only on demand existing, but on customer contracts remaining durable enough to support the capital structure behind the deployment.
Nscale’s prospectus explicitly identifies customer concentration as a risk. Its business has since expanded substantially, with major commitments involving Microsoft and Anthropic, and the company indicates that the proportion of revenue attributable to its largest customer should fall as those newer contracts contribute.
The Glomfjord arrangement also illustrates the role of Norway in the European accelerated-compute market. Facilities with established grid connections, cooler operating conditions, and access to low-carbon electricity have become attractive locations for GPU clusters that can serve customers outside the country without the equipment itself leaving Europe.
The underlying hardware is important. Nvidia’s B200 accelerators sit within a class of equipment subject to US export-control rules when supplied to certain destinations or users. The Financial Times reported that the Norwegian arrangement complied with the applicable rules. Policymakers in the US continue to debate how export controls should apply where restricted customers obtain remote access to advanced compute located in third countries.
That policy debate is separate from whether the Glomfjord contract complied with rules in force when it was signed. The infrastructure consequence is that data centre location can affect not only power and latency, but also which regulatory regimes govern access to high-end computing hardware.
The IPO exposes more of the operating model
Nscale’s public filing is providing considerably more detail about a business that has expanded rapidly from a comparatively small revenue base. The company reported $140.6m of revenue in the first half of 2026, alongside a net loss exceeding $1bn, much of which related to non-cash items.
It says it currently has 55MW of compute capacity operating or leased and a much larger development pipeline attached to contracted demand. The relationship between those contracts, site delivery, GPU procurement, and financing will be central to how investors assess the company.
Customer concentration should decline if the larger Microsoft and Anthropic commitments enter service as planned. That transition, however, replaces one form of concentration risk with a much larger execution challenge: turning multi-billion-dollar contracts into commissioned power, cooling, racks, networks, and GPUs on schedule.
The ByteDance-linked contract is therefore useful less as an isolated customer story than as a view into how European AI infrastructure has been financed. A relatively concentrated revenue stream helped support the purchase and deployment of expensive accelerators in Norway. Nscale’s next phase will require that model to work across far more capacity and a much broader customer base.
