Nscale Closes $2B to Build Europe's AI Backbone

Nscale today announced $2 billion in Series C funding, the largest round ever raised by a European startup. The company is now valued at $14.6 billion. This capital isn't going toward general hiring or sales expansion - it goes directly into building Stargate Norway, a 100,000-GPU AI gigafactory in Narvik being constructed with OpenAI as the anchor customer.

What Nscale Actually Builds

Nscale isn't a cloud reseller. They own the full stack - from the data center facilities and power agreements to the GPU clusters, networking, and the orchestration layer that developers actually touch. Founded in 2024 by Josh Payne, a former Australian coal miner who ended up building supercomputing infrastructure in London, the company has moved faster than almost anyone in the industry expected.

The Compute Stack

The platform covers three main use cases: training, fine-tuning, and inference. For training and fine-tuning, developers access distributed GPU clusters via managed SLURM or Kubernetes. For inference, Nscale runs a serverless endpoint layer that's drop-in compatible with the OpenAI SDK - swap the base URL to https://inference.api.nscale.com/v1 and your existing code works unchanged.

from openai import OpenAI

client = OpenAI(
    api_key="NSCALE_API_KEY",
    base_url="https://inference.api.nscale.com/v1",
)

response = client.chat.completions.create(
    model="meta-llama/Llama-4-Scout-17B-16E-Instruct",
    messages=[{"role": "user", "content": "Explain co-packaged optics in 2 sentences"}],
    temperature=0.7,
)
print(response.choices[0].message.content)

Nscale is also an official Hugging Face Inference provider, which means models published on the Hub can be routed through Nscale's GPU fleet with no additional setup.

The Data Centers

The Norwegian facilities are the crown jewels. Glomfjord sits in a small industrial town just above the Arctic Circle, next to a hydroelectric plant that's been running for over a century. The ambient climate handles a significant fraction of cooling load without additional energy. Keflavik runs more than 4,600 NVIDIA Blackwell Ultra GPUs - current-generation hardware - on Icelandic geothermal power.

The Glomfjord hydroelectric dam, which has supplied power to the region for over a century, now feeds Nscale's AI compute cluster operating above the Arctic Circle. Source: nscale.com

The $2 Billion Round

The Series C was led by Aker ASA - the Norwegian industrial conglomerate that already co-owns Stargate Norway with Nscale - and 8090 Industries. That combination is remarkable: it's not just financial capital, it's strategic alignment with the entity that controls the land and power agreements for the Norway facility.

The board additions are worth noting. Sheryl Sandberg joins as a director - she was Meta's COO for fourteen years and was involved in Google's early ad infrastructure scaling. Nick Clegg, former UK Deputy Prime Minister and now a General Partner at Hiro Capital, also joins. Susan Decker, CEO of Raftr and a former Yahoo president, rounds out the three new seats.

These aren't just advisory names. European AI infrastructure has a regulatory complexity problem - data sovereignty rules, energy permitting, cross-border GPU export compliance - and adding people with deep experience in exactly that kind of institutional navigation is a meaningful choice.

Stargate Norway - The Real Bet

Stargate Norway, the planned 100,000-GPU AI gigafactory in Kvandal, just outside Narvik, announced in July 2025 as a joint venture between Nscale, Aker ASA, and OpenAI. Source: nscale.com

The Stargate Norway announcement came in July 2025. The $2B round today is the capital that makes it real.

OpenAI as Anchor Customer

OpenAI is the initial offtaker under its "OpenAI for Countries" programme - meaning they've committed to buying compute from the facility, with the option to scale their usage over time. For context, Nscale and Aker each committed about $500 million to the initial 20MW phase (out of a total $1 billion for that phase), and the offtake agreement with OpenAI de-risks the revenue side of that investment. Stargate Norway is also OpenAI's first European data center.

This is a different dynamic than the $50 billion OpenAI-AWS deal, where OpenAI is buying capacity from an existing hyperscaler. Nscale is building infrastructure specifically for OpenAI's European needs, with Aker's Norwegian government relationships providing power and land at a scale the hyperscalers can't easily replicate on that timeline.

The Renewable Energy Stack

The full build targets 230MW of capacity - expandable to 520MW - all on Norwegian hydropower. At that scale, this isn't a marketing claim. Norway generates roughly 90% of its electricity from hydro, and Narvik specifically has excess generation capacity due to limited local demand. Nscale's deal structures direct power purchase agreements rather than relying on RECs or carbon offsets.

The engineering approach also includes reusing GPU heat to supply local low-carbon enterprises in the region, which is increasingly a condition of permitting in Scandinavian markets.

Where It Falls Short

Nscale is operationally real - they have running clusters, paying customers, and a live API. But the story being pitched to investors is still mostly a construction project.

The 100,000-GPU Stargate Norway target is for end-2026. The initial 20MW phase is the near-term commitment, and "initial phase" at 20MW is smaller than what some well-funded US inference providers operate today. Building at the speed and scale being promised, in an Arctic location with demanding permitting requirements, isn't a solved problem. The Ayar Labs silicon photonics bet shows how much capital is being directed at AI interconnect challenges that Nscale will still need to solve at scale.

The investor list also reflects where the risk sits. Jane Street and Citadel aren't strategic AI infrastructure players - their presence suggests the round includes meaningful yield or equity upside for purely financial allocators, which sometimes suggests the round size required more buyers than were naturally available.

Competition is structural. AWS, Azure, and Google have existing relationships with every major AI lab. Their latency to market on European capacity is narrowing. Meta's multi-billion GPU deals with AMD and NVIDIA (which we covered here) give hyperscalers leverage that's harder to match even with sovereign differentiation.

Nscale's actual advantage is narrow but real: they can offer European AI customers compute that's physically in Europe, on renewable power, with data sovereignty guarantees that a US hyperscaler can't credibly match under current regulatory pressure. Whether that wedge is wide enough to sustain a $14.6 billion valuation depends on how quickly the EU mandates what "sovereign AI infrastructure" actually means in practice.

Sources:

• Nscale Series C Press Release
• Stargate Norway Announcement - Nscale
• Nscale Sets European Record - Techzine
• Stargate Norway Blog - Nscale
• Nscale Inference API Docs