Sandberg, Clegg join Nscale board as this ‘Stargate Norway’ startup hits $14.6B valuation

Nscale's $14.6B Valuation and “Stargate Norway” Project: A Technical Deep Dive into Europe’s Largest AI Infrastructure Play

Executive summary
Nscale is a vertically integrated AI infrastructure company that owns the stack from renewable power sourcing through GPU clusters to orchestration software. The British startup has reached a $14.6 billion post-money valuation after a $2 billion Series C (including a $433 million pre-Series C SAFE) led by Aker ASA and 8090 Industries, with participation from Nvidia, Dell, Nokia, Blue Owl, Citadel, Jane Street, and others. The company’s flagship “Stargate Norway” joint venture with Aker targets a 100,000 Nvidia GPU cluster by the end of 2026, with OpenAI as the anchor customer. In parallel, Nscale has an expanded Microsoft deal covering approximately 200,000 Nvidia GPUs across three European and one U.S. data-center sites in collaboration with Dell. The raise is widely viewed as IPO preparation, and the addition of Sheryl Sandberg, Nick Clegg, and Susan Decker to the board signals a shift toward enterprise-grade governance and hyperscaler relationships.

Technical architecture
Nscale’s core differentiation is full vertical integration across four layers:

1. Energy layer – Direct access to low-cost renewable (primarily hydro) power in Norway and planned sites across Europe and North America. The company explicitly designs clusters to reuse waste heat for district heating or industrial processes, improving overall power usage effectiveness (PUE) beyond typical 1.2–1.3 hyperscale figures.

2. Data-center layer – Purpose-built or retrofitted facilities optimized for high-density GPU deployments. While exact rack power density is not disclosed, the scale (100k–200k H100/H200/B200-class GPUs) implies liquid-cooling readiness and 100–150 kW per rack.

3. Compute layer – Large-scale homogeneous Nvidia GPU clusters. The Stargate Norway project is explicitly scoped to 100,000 Nvidia GPUs by end-2026. The Microsoft deal adds another ~200,000 GPUs, suggesting Nscale is on track to operate one of the largest independent GPU clouds in Europe. No specific GPU generation mix (H100 vs. Blackwell) has been disclosed beyond “Nvidia GPUs.”

4. Orchestration and software layer – Nscale develops its own cluster management, scheduling, and inference-serving platform. The company positions this software as a competitive advantage against pure co-location or hyperscaler offerings, promising better utilization, multi-tenancy controls, and lower orchestration overhead than open-source stacks such as Kubernetes + Slurm + custom Ray/Spark layers.

The decision to bring the Aker joint venture fully under Nscale management centralizes delivery, governance, and operational telemetry under one roof. This removes traditional JV friction and allows tighter integration between power procurement, facility operations, and the software control plane.

Performance analysis
Public benchmark data on Nscale clusters is not yet available; the company has not released MLPerf, GPU utilization, or tokens-per-dollar figures. However, several indirect signals can be derived from the disclosed deals:

• Scale — 100k GPUs in Norway + 200k GPUs via the Microsoft/Dell agreement places Nscale in the same absolute GPU count tier as some hyperscaler regions. A single 100k H100 cluster at ~700W TDP per GPU (plus networking and cooling) requires roughly 100–120 MW of critical load, making Stargate Norway one of the largest single-site AI infrastructure projects announced in Europe to date.

• Power economics — Norway’s abundant, low-cost hydro power (often < $30/MWh) gives Nscale a structural advantage versus U.S. or continental European sites that face higher energy prices or grid constraints. Waste-heat reuse further improves total cost of ownership (TCO).

• Customer traction — OpenAI as the initial Stargate Norway customer and an expanded multi-year Microsoft deal indicate that the infrastructure has passed technical and compliance bar for two of the most demanding AI organizations. This is a strong proxy for cluster stability, low tail latency on RDMA fabrics, and effective job scheduling at scale.

Competitive context
Nscale sits at the intersection of several competitor categories:

• Hyperscalers (Azure, AWS, GCP): Offers similar GPU counts but with greater geographic diversity and managed services. Nscale’s advantage is lower power cost in Norway and potentially more flexible contract terms for large committed blocks.

• Specialized GPU cloud providers (CoreWeave, Crusoe, Lambda, Voltage Park): Nscale is larger in announced GPU count than most pure-play GPU clouds and benefits from deeper vertical integration into power and heat reuse. CoreWeave remains the closest peer in Europe and the U.S.; Nscale’s $14.6B valuation now exceeds CoreWeave’s last reported $8.5B–$9B range.

• European AI infrastructure peers: Compared with Mistral AI (valued at $6B) and Helsing ($5B+), Nscale is now Europe’s highest-valued private AI company. Unlike the model developers, Nscale’s business is capex-heavy and closer to a utility than a software product.

Technical implications for the ecosystem

1. Europe’s AI sovereignty push — By building 100k+ GPU clusters on European soil with local renewable energy, Nscale reduces reliance on U.S. hyperscalers for sovereign AI workloads. Governments and enterprises concerned about data residency and export controls now have a credible large-scale alternative.

2. Accelerated liquid-cooling and heat-reuse standards — Widespread adoption of waste-heat utilization in Norway could set a template for other Nordic and cold-climate data centers, lowering both operational cost and carbon footprint.

3. IPO path for AI infrastructure — A successful 2026 IPO would create a liquid public pure-play in AI compute infrastructure, giving investors direct exposure to GPU capex cycles without buying Nvidia or hyperscaler stock.

4. Pressure on traditional colocation providers — Equinix, Digital Realty, and others must now compete on power cost, GPU density, and software orchestration rather than just shell space.

Limitations and trade-offs

• Capital intensity — Even with $2B equity and $1.4B debt (GPU-backed term loan), building 300k GPUs at current Nvidia pricing requires tens of billions in total capex. Execution risk remains high.
• Geographic concentration — Heavy reliance on Norwegian hydro exposes the company to potential regulatory, permitting, or transmission constraints.
• Limited software transparency — No public benchmarks or open-source components have been released, making it difficult for ML engineers to evaluate orchestration quality versus established stacks.
• Blackwell transition risk — The announcement references “Nvidia GPUs” without specifying mix; any delay in Blackwell NVL72/NVL144 rack availability could slow the 2026 ramp.
• Talent — Expanding engineering and operations teams across three continents while maintaining high utilization is non-trivial.

Expert perspective
Nscale’s move from a Series B infrastructure startup to a $14.6B vertically integrated AI utility in roughly six months is remarkable. The combination of credible hyperscaler customers (OpenAI, Microsoft), strategic Norwegian energy partnership, and high-profile board additions suggests the company has solved early technical and commercial validation problems. The real test over the next 18 months will be whether Nscale can deliver the 100k GPU Stargate Norway cluster on schedule, achieve competitive utilization rates (>65–70% sustained), and demonstrate that its proprietary orchestration layer delivers measurable TCO or performance advantages. If successful, Nscale could become the de-facto “AI power company” of Europe and a template for other regions with stranded renewable energy.

Technical FAQ

### How does Nscale’s GPU scale compare to other independent providers?
Nscale’s disclosed pipeline (100k in Norway + ~200k via Microsoft/Dell) totals ~300k GPUs by end-2026. This puts it ahead of most pure-play GPU cloud companies and roughly comparable to a single large hyperscaler region. Exact real-time utilization and GPU generation mix remain undisclosed.

### What is the power architecture behind Stargate Norway?
The project leverages Norway’s low-cost renewable hydro power and emphasizes waste-heat reuse. While exact PUE, rack density, or cooling technology (direct liquid cooling vs. immersion) is not public, the scale implies multi-100-MW critical load with advanced thermal management.

### Is Nscale’s orchestration platform open source or proprietary?
The company describes it as “its platform” and a competitive differentiator, indicating a proprietary stack. No code, APIs, or performance benchmarks have been published as of the Series C announcement.

### How does the $1.4B GPU-backed debt facility affect capex flexibility?
The delayed-draw term loan allows Nscale to finance GPU purchases without immediate equity dilution. It is secured against the GPUs themselves, typical for large-scale AI infrastructure financings, and provides runway while the company prepares for a potential IPO.

### Is the Microsoft deal additive or overlapping with Stargate Norway?
The Microsoft agreement covers three European sites and one U.S. site for ~200k GPUs and appears separate from the Norway-specific Stargate project, giving Nscale geographic diversification beyond the Nordic region.

Sources

• TechCrunch — Sandberg, Clegg join Nscale board as this ‘Stargate Norway’ startup hits $14.6B valuation
• The Guardian — British AI datacentre firm Nscale raises $2bn as Sheryl Sandberg and Nick Clegg join board
• Bloomberg — Nscale Raises $2 Billion and Adds Sandberg, Clegg to Board
• CNBC — Nvidia backs AI data center startup Nscale as it hits $14.6 billion valuation
• Tech.eu and Tech Funding News coverage of the Series C

Word count: ~1,480. Analysis based exclusively on disclosed information in the provided announcement and supporting coverage as of March 2026.