Nscale Expands Its AI Datacenter Rollout With 30,000 More NVIDIA Vera Rubin GPUs, Taking the Total Beyond 130,000

Nscale has announced a major expansion of its AI infrastructure plans, confirming that it will add more than 30,000 additional NVIDIA Vera Rubin GPUs to its Narvik campus in Norway in 2027. This new deployment builds on the company’s earlier rollout of 100,000 plus Vera Rubin GPUs across Europe, taking Nscale’s total announced Rubin capacity to more than 130,000 GPUs. The latest expansion comes as customer demand for next generation AI compute continues to rise, particularly for large scale inference and emerging agentic AI workloads.

According to Nscale, the added capacity in Narvik is tied to an expanded agreement with Microsoft, with the site reaching 230MW of AI chip capacity once the project is fully deployed. The company says the Narvik campus will become one of the largest onshore infrastructure projects in Norway, underlining how quickly the region is becoming a key location for hyperscale AI buildouts in Europe. This also signals that Microsoft’s appetite for advanced AI infrastructure in Europe is still accelerating, and that Nscale is positioning itself as one of the core partners helping deliver that capacity.

The previous March 16 announcement had already placed Nscale among the first providers globally to deploy the NVIDIA Vera Rubin platform in 2027. In that earlier release, Nscale said the initial 100,000 plus GPU rollout would be spread across its sites in the United Kingdom, Norway, and beyond, with Microsoft named as a major beneficiary of the capacity. The new Narvik update effectively scales that commitment further, concentrating an additional 30,000 plus Rubin GPUs into one of the company’s most strategically important campuses.

From a market perspective, this is a significant move because NVIDIA Vera Rubin is being framed as one of the industry’s most important next generation AI platforms for inference heavy deployments. Nscale’s own statements describe Vera Rubin NVL72 as a platform aimed at frontier AI development and deployment, while NVIDIA’s Ian Buck said the system will provide critical AI infrastructure for European developers working on agentic and physical AI. In practical terms, this means Nscale is not just growing capacity for traditional training clusters, but also preparing for a wave of enterprise and developer demand centered on fast, efficient, large scale inference.

The infrastructure itself is equally notable. Nscale says these deployments will use NVIDIA Vera Rubin NVL72 systems, which package 72 Rubin GPUs into a rack scale AI supercomputing design. The company also said its broader 2027 infrastructure mix will combine advanced architectures including NVIDIA Vera Rubin and NVIDIA Grace Blackwell, showing that this is part of a larger long term compute strategy rather than a one off procurement headline. For the European AI market, that matters because access to this class of hardware has increasingly become a strategic differentiator for cloud providers, model developers, and enterprise customers trying to stay competitive.

Nscale founder and CEO Josh Payne said customer demand for advanced AI infrastructure continues to accelerate across markets, adding that the Narvik deployments reflect strong partner interest and the company’s focus on bringing the latest technology online at real scale. That wording is important because it reinforces the idea that this expansion is demand led, not speculative. It also suggests that Nscale sees Narvik as a flagship site in its effort to become a major AI infrastructure provider across Europe.

For Europe’s AI ecosystem, this announcement is a strong signal that the race for next generation compute is intensifying. With more than 130,000 Vera Rubin GPUs now tied to Nscale’s 2027 rollout plans, the company is making a serious play to become one of the region’s most important providers of high density inference and AI cloud capacity. If the deployment lands on schedule, Narvik and Nscale’s wider European footprint could become a major foundation for the next phase of AI services, enterprise inference, and agentic AI development.

Do you think Europe is finally building enough next generation AI infrastructure to compete at scale, or is demand still moving faster than supply?