Nscale: European Sovereign AI Cloud & Stargate Norway

Executive Summary

Nscale is a London-based AI infrastructure company and Europe's fastest-growing AI hyperscaler.¹ Spun from crypto miner Arkon Energy in 2024, it builds vertically integrated AI data centers.² Total raised: $1.1B Series B (Europe's largest) plus $433M pre-Series C SAFE.³

Anchor partnerships define its trajectory. Stargate Norway with Aker and OpenAI targets 100,000 NVIDIA GPUs on renewable energy.⁴ A $14B GPU supply contract with Microsoft covers 200,000 GB300 GPUs across Europe and the US.⁵ The company plans a 2026 IPO.⁶

Its sovereign cloud platform offers serverless inference, fine-tuning, and bare-metal GPU access. All infrastructure operates under UK and European law with full data isolation.⁷ Nscale claims 80% cost savings versus hyperscalers (per Nscale marketing materials; independently unverified).⁸

Nscale is not Northern Data Group, despite surface similarities. It spun out from Arkon Energy in May 2024.⁹ The crypto-to-AI pivot mirrors Crusoe Energy.

Why This Matters Now

Nscale validates three theses relevant to MARA's strategy:

• Sovereign AI demand is real. A $1.1B Series B proves institutional capital backs sovereign compute as a multi-billion-dollar market.
• Vertical integration wins. Owning power, data centers, and software creates cost advantages hyperscalers cannot match.
• Speed beats scale. Nscale went from stealth to $14B in contracted deal value in 18 months by securing GPU supply and long-term offtake.

Company Profile & History

Origins: From Crypto Mining to AI Hyperscaler

Nscale began as Arkon Energy, a Bitcoin miner founded by Josh Payne. Arkon ran renewable-powered data centers in Norway, the US, and Australia, raising $110M by late 2023.² In May 2024, Payne spun out Nscale to pursue AI cloud services, reusing data center assets and power expertise.⁹

Corporate Overview

Attribute	Detail
Legal Name	Nscale Global Holdings Limited
Headquarters	London, United Kingdom
Founded	2024 (spun off from Arkon Energy)
CEO & Founder	Josh Payne
Employees	~160 (as of early 2026)10
Contracted Deal Value	$14B (Microsoft GPU supply contracts, not recurring revenue)11
Status	Private; IPO targeted H2 2026

Leadership Team

Name	Role	Background
Josh Payne	Founder & CEO	Ex-CEO Arkon Energy. Background in capital markets and supercomputing.12
Donald Morrison	Co-Founder & COO	Co-founded company. Operations lead.
Alex Sharp	President, Data Centers	Chartered engineer. 30+ years. Designed 54 data centers across Europe.12
David Power	CTO	Technical architecture and platform.
Nidhi (surname not publicly disclosed)	Head of AI Infrastructure	Former Head of AI Infrastructure at Microsoft. Managed 2M GPUs.12
Jing (surname not publicly disclosed)	SVP Global Operations	Former Alibaba Cloud, Ant Group, Lazada. Supply chain and commercial strategy.12

Corporate Structure

Nscale Global Holdings Limited is the UK-registered parent. Key subsidiaries:

• Nscale Ltd: UK operating entity.²⁵
• Stargate Norway JV: 50/50 with Aker ASA for Kvandal.
• US Operations: Cedarvale (Barstow, Texas) for Microsoft GPU supply.

Each facility operates under local law, reinforcing the sovereign cloud narrative.

Founding Director Exit

A founding director exited in late 2025, shortly after the $1.1B raise.²⁵ Investors should monitor whether this signals strategic disagreement or routine scaling-stage transition.

Funding & Financial Profile

Funding History

Round	Date	Amount	Key Investors
Arkon Energy (pre-spinoff)	2022-2023	~$140M	Various; crypto infrastructure investors
Series A	Dec 2024	$155M	Sandton Capital, early backers13
Series B	Sep 2025	$1.1B	Aker ASA (lead), NVIDIA, Dell, Nokia, Fidelity, G Squared, Blue Owl, Point72, T.Capital1
Pre-Series C SAFE	Oct 2025	$433M	Blue Owl, Dell, NVIDIA, Nokia3
Total Raised		$2.69B

Strategic Investor Significance

Investor	Strategic Value
Aker ASA	Norwegian industrial conglomerate. 50/50 JV on Stargate Norway. Energy infrastructure and Nordic network.
NVIDIA	$683M+ invested. Priority GPU allocation (GB200, GB300). DGX Cloud Lepton partnership.14
Dell	Hardware supply chain and server infrastructure.
Nokia	High-bandwidth networking for GPU clusters.
Fidelity	Institutional validation from one of the world's largest asset managers.
Point72	Steve Cohen's hedge fund. Pre-IPO credibility.

Technical Infrastructure

Data Center Portfolio

Location	Capacity	GPU Count	Status	Partner
Kvandal, Norway (Stargate Norway)	230 MW (expandable to 520 MW)	100,000 NVIDIA GPUs	Phase 1 under construction; target end 20264	Aker ASA, OpenAI
Loughton, Essex, UK	50 MW (expandable to 90 MW)	45,000 GB200 GPUs initially; 23,040 GB300 from Q1 2027	Live Q4 202615	Microsoft
Cobalt Park, UK (Stargate UK)	TBD	Up to 31,000 NVIDIA GPUs over time	Planning phase16	OpenAI, NVIDIA
Sines, Portugal	TBD	12,600 GB300 GPUs	Delivery from Q1 20265	Microsoft (Start Campus)
Barstow, Texas, US	240 MW	104,000 NVIDIA GPUs	Q3 2026 opening5	Microsoft (Cedarvale)
Glomfjord, Norway (legacy)	~20 MW	Legacy; original Arkon site	Operational	None (owned)

Technology Stack

GPU Deployment Roadmap: 300,000 GPUs Globally

Nscale targets 300,000 NVIDIA GPUs globally by end of 2027.¹⁶ The UK receives 58,640 GPUs. Another 4,600 GB300 GPUs feed NVIDIA DGX Cloud/Lepton marketplace access. Renewable energy powers the Nordic sites; US facilities use grid with PPAs.

Stargate Norway: Deep Dive

Stargate Norway is Nscale's flagship project and OpenAI's first European Stargate deployment.²³ Located in Kvandal, northern Norway, it targets 100,000 NVIDIA GPUs by end of 2026.

Key Design Decisions

• Renewable power: 100% hydroelectric. Zero carbon emissions.
• Liquid cooling: Direct-to-chip, closed-loop. Critical for GB200/GB300 GPUs. Eliminates water consumption from evaporative cooling.
• Waste heat recovery: GPU heat redirected to regional low-carbon enterprises.
• Joint venture: 50/50 with Aker. ~$500M equity each for initial 20 MW phase.⁴
• Phased build: 230 MW initial, expandable to 520 MW.

Norway and UK sites benefit from cold climates. Nscale has not published PUE figures, but liquid cooling plus cold climates suggests sub-1.2 PUE.

Geographic Strategy: Why These Locations?

Location	Why Chosen	Strategic Rationale
Narvik, Norway	Abundant hydroelectric power. Cold climate. EEA data adequacy.	Flagship sovereign European facility. Renewable narrative. OpenAI's first EU Stargate.
Essex, UK	Proximity to London. UK government AI Growth Zone designation.	UK's AI supercomputer. Microsoft partnership anchor. Regulatory access.
Cobalt Park, UK	North East England AI Growth Zone.	Government-designated zone. Political goodwill. Stargate UK branding.
Sines, Portugal	Start Campus. EU data residency. Renewable energy access.	EU mainland presence. Microsoft delivery point. Diversifies beyond UK/Nordic.
Barstow, Texas	Cheap power. US market access. Microsoft Cedarvale project.	Proves global execution. US market diversification. IPO narrative.

Products & Services

Service Portfolio

Product	Description	Pricing Model
Serverless Inference	Managed endpoints. Llama 4, DeepSeek R1, Qwen models. OpenAI-compatible API. Auto-scaling.17	Pay-per-token. Free credits for new users.
Fine-Tuning	Serverless. No ML expertise needed. Full parameter and LoRA support.18	Pay-per-run. $2 minimum.
Bare-Metal GPU	H100, GB200 for training. Dedicated nodes with InfiniBand.	Per-GPU-hour.
Sovereign Cloud	UK/EU law governed infrastructure. Local personnel. No foreign data access. Full workload isolation.7	Enterprise contracts.

Sovereign Serverless: Isolation by Design

Built for sovereign compliance from the ground up.¹⁹

• Workloads never share data space across organizations.
• Infrastructure governed by UK/EU law exclusively.
• Operated by local personnel. No foreign data access paths.
• Full OpenAI API compatibility for zero-effort migration.

Supported Models

Model Family	Variants Available	Use Case
Meta Llama	Llama 4 Scout 17B, Llama 3.3 70B Instruct, Llama 3.2 11B Vision, Llama 3.1 8B17	General purpose, vision, code generation
DeepSeek	DeepSeek R1 Distill 70B, R1 Distill 8B, R1 Qwen variants	Reasoning, code, math
Alibaba Qwen	Qwen family models	Multilingual, code, general purpose

The catalog focuses on open-source and open-weight models. Customers control model weights and audit behavior. Nscale supports Llama, Mistral, and other open-source families. This overlaps directly with MARA's target LLM stack. Both platforms compete for sovereign Llama inference workloads.

Pricing Comparison (Estimated)

Service	Nscale (Est.)	Hyperscaler Equivalent	Savings Claim
Serverless Inference (per 1M tokens)	Competitive with Together AI, DeepInfra	AWS Bedrock, Azure AI	Up to 80% (per Nscale marketing; unverified)8
Fine-Tuning (per job)	$2 minimum	$10-50+ on hyperscalers	Not verified
Bare-Metal GPU (H100/hr)	Market rate (~$2-3/hr)	$3-10/hr on hyperscalers	30-70%

Nscale has not published per-token pricing. Free credits and $2 fine-tuning minimums suggest aggressive developer acquisition. Real monetization comes from enterprise contracts and GPU supply deals.

Platform Architecture Differentiators

OpenAI API Compatibility

Fully compatible with OpenAI's API format. Developers redirect existing code without changes. Enterprises locked into OpenAI can move to sovereign infrastructure instantly.¹⁹

Multi-Region Scheduling

Workloads route to specific regions for compliance. UK government customers run inference only on UK GPUs. Critical for data residency requirements.

DGX Cloud Lepton Integration

Part of NVIDIA's DGX Cloud Lepton marketplace. Developers discover Nscale capacity through NVIDIA's ecosystem, expanding reach beyond direct sales.¹⁶

Customer & Partnership Analysis

Anchor Partnerships

Partner	Deal Structure	Value
Microsoft	200,000 GB300 GPU supply contract across US and Europe. Loughton UK, Start Campus Portugal, Cedarvale Texas. Total $23B with prior commitments.5	$14B GPU supply contract (new) / $23B (total)
OpenAI	Initial offtaker for Stargate Norway (100K GPUs). Stargate UK with up to 31K GPUs. Part of "OpenAI for Countries" program.4	Multi-billion (offtake)
Aker ASA	50/50 JV on Stargate Norway. ~$500M equity commitment from each partner. Lead Series B investor.4	$1B+ combined
NVIDIA	$683M investment. Priority GPU allocation. DGX Cloud Lepton marketplace integration.14	$683M investment
Armada	LOI for hub-and-spoke sovereign AI deployment. Nscale provides hyperscale hubs; Armada provides edge.20	TBD

Customer Segments

• Hyperscalers: Microsoft dominates. GPU supply contracts drive near-term contracted value.
• AI Labs: OpenAI uses Stargate for sovereign European deployments. Others access via DGX Cloud Lepton.
• Enterprise & Government: Regulated industries (finance, healthcare, defense) needing sovereign inference.¹⁶

Contract Concentration Analysis

Nscale's customer base is highly concentrated. Microsoft accounts for the majority of contracted deal value. The $14B GPU supply contract (expanded to $23B total) dwarfs all other partnerships.

Customer	Est. Contracted Value Share	Risk Level
Microsoft	>80%	Critical concentration
OpenAI (via Stargate)	~10-15%	Moderate (offtake, not guaranteed)
Enterprise / Other	<5%	Diversification opportunity

Microsoft provides guaranteed utilization. But if Microsoft renegotiates, delays, or shifts in-house, Nscale's financial model is at risk. IPO investors will scrutinize this.

Stargate Brand Power

"Stargate" branding gives Nscale outsized credibility with European governments. This borrowed credibility accelerates deal cycles.

However, Stargate branding is not exclusive. OpenAI deploys Stargate projects with multiple partners. Crusoe builds Stargate Abilene. The brand helps today but does not create a moat.

Armada Partnership: Hub-and-Spoke Model

In February 2026, Nscale signed an LOI with Armada for global sovereign AI deployment.²⁰ Nscale provides hyperscale hubs; Armada deploys edge containers. This validates that sovereign AI needs both centralized and distributed infrastructure.

Enterprise Adoption Barriers

• Platform maturity: Launched in 2024. Enterprise teams prefer multi-year track records.
• Support: ~160 employees cannot match AWS or Azure 24/7 global support.
• Ecosystem lock-in: AWS/Azure tooling creates switching costs beyond the inference API.
• Certifications: SOC 2, ISO 27001, and sector-specific certifications take time.

Competitive Positioning

Neocloud Competitive Landscape

Dimension	Nscale	Nebius	Crusoe	CoreWeave
HQ / Focus	London. European sovereign AI.	Amsterdam. Ex-Yandex team. EU/US cloud.	San Francisco. Clean energy AI.	New Jersey. US GPU cloud leader.
Total Raised	$2.69B3	$4.6B+	$3.8B+	$12B+ (incl. debt)
Key Customer	Microsoft ($14B GPU supply contract)	Microsoft ($17.4B), Meta ($3B)	OpenAI (Abilene Stargate)	Microsoft, Meta
Power Pipeline	1.3 GW	~1 GW (2026 target)	45 GW (claimed)	~1.5 GW
Sovereignty	Core differentiator. UK/EU law. Full isolation.7	EU-focused. GDPR compliance.	US-focused. Clean energy angle.	US-focused. No sovereign pitch.
Services	Inference, fine-tuning, bare-metal, sovereign cloud	Inference, training, bare-metal	GPU cloud, managed Kubernetes	GPU cloud, managed Kubernetes, inference
IPO Plans	H2 20266	Listed (NBIS on Nasdaq)	Filed S-1 (delayed)	IPO completed 2025

Differentiation vs. Hyperscalers

Nscale positions against AWS, Azure, and GCP on three axes.

• Cost: Claims 80% savings vs. hyperscalers (per Nscale marketing materials; independently unverified).⁸ Driven by purpose-built facilities, direct power procurement, and vertical integration.
• Sovereignty: Infrastructure governed by UK/EU law. No US CLOUD Act exposure. Critical for defense, healthcare, finance, and public sector.
• AI-native: Purpose-built for GPU workloads. No legacy VM overhead. Every infrastructure dollar goes to AI compute.

The 80% savings claim (per Nscale marketing materials; independently unverified) likely compares on-demand GPU pricing against Nscale's bare-metal rates. For reserved instances, the gap narrows. MARA should validate against its own cost structure.

European Market Positioning

Nscale and Nebius are the two dominant neoclouds targeting Europe. Both deployed 14,000 Blackwell GPUs for UK data centers.²¹ Their positioning differs:

SWOT Analysis

Strengths	Weaknesses
Largest European AI infrastructure raise ($1.1B Series B) NVIDIA strategic investor with priority GPU allocation Stargate brand association with OpenAI Vertical integration from power to API layer Renewable energy positioning (Norway hydro) Experienced hyperscaler talent (ex-Microsoft, Alibaba)	Extreme customer concentration (Microsoft >80%) Crypto mining origin may face IPO scrutiny NVIDIA-only GPU strategy (no hardware diversification) Small team (~160) for 1.3 GW buildout ambition No published uptime SLA or latency benchmarks Limited track record (founded May 2024)
Opportunities	Threats
EU AI Act drives demand for compliant infrastructure Government sovereign AI mandates accelerating IPO could unlock additional growth capital Armada partnership extends reach to edge/Middle East Enterprise diversification beyond hyperscalers	Hyperscalers build sovereign regions themselves GPU supply glut could undercut pricing Microsoft renegotiation or delay risk Nebius competing for same European sovereign deals Execution risk on multi-continent buildout

Market Position Map

Key Milestones & Timeline

Upcoming Milestones to Watch

Milestone	Expected Date	Significance for MARA
Portugal GPU delivery (12,600 GB300s)	Q1 2026	First large-scale GB300 deployment in Europe.
Cedarvale Texas campus opening	Q3 2026	Proves US execution capability.
Loughton UK data center live	Q4 2026	UK's largest AI supercomputer.
Stargate Norway Phase 1 (100K GPUs)	End 2026	Flagship project. Delay signals execution risk.
Potential IPO	H2 2026	S-1 reveals margins, customer concentration, unit economics.
Loughton GB300 expansion (23K GPUs)	Q1 2027	Signals ongoing Microsoft commitment.

Strategic Threat Assessment

Threat to MARA: Medium-High

Factor	Assessment	Impact
Scale Overlap	Gigawatt campuses vs. modular containers. Different tiers.	LOW
Sovereign Inference	Direct competition for regulated industry workloads.	HIGH
European Market	Nscale dominates UK/Nordic sovereign narrative.	MEDIUM
Cost Position	Claims 80% savings (per Nscale marketing; unverified). Real vertical integration advantages.	HIGH
Customer Overlap	Nscale: hyperscalers. MARA: enterprise. Limited overlap today.	LOW
Technology	GB300 GPUs. Serverless sovereign inference.	MEDIUM

Key Risks for Nscale

Risk Category	Description	Likelihood
Customer concentration	Microsoft accounts for >80% of contracted deal value. Renegotiation, delay, or cancellation would be existential.	HIGH
Execution at scale	1.3 GW across three continents with ~160 employees. Talent, supply chain, and construction are all risk vectors.	MEDIUM
Crypto origin scrutiny	Arkon Energy pivot will face IPO scrutiny. Analysts will examine asset transfers and accounting.	MEDIUM
NVIDIA dependency	100% NVIDIA GPU strategy. Supply disruption or pricing change cascades directly.	MEDIUM
Geopolitical	Sovereignty narrative is fragile. US export controls or trade disputes could disrupt operations.	LOW
GPU oversupply	If demand plateaus or GPU supply catches up, cost advantages narrow and utilization drops.	LOW (near-term)

MARA Strategic Response

Key Lesson: Contracted Value vs. Platform Revenue

Nscale secured $14B in Microsoft GPU supply contracts before building most infrastructure. Contracted value comes from GPU supply, not cloud platform. Serverless inference is a future growth driver. MARA should not benchmark against Nscale's headline figures.

Key Data Points for Quick Reference

Report prepared by MinjAI Agents. Intelligence current as of February 20, 2026. Sources verified against public filings, press releases, and credible technology media.

What We Don't Know

Unknown	Why It Matters	How to Monitor
Inference platform margins	GPU supply margins vs platform margins are very different. We don't know either.	Watch IPO S-1 filing (H2 2026) for segment-level financials.
Serverless inference performance	No latency benchmarks, throughput metrics, or uptime SLA data published.	Test the platform directly or monitor developer community feedback.
Customer concentration depth	Microsoft >80% of contracted value. Renegotiation risk is existential.	Monitor Microsoft earnings calls for Nscale mentions.
Founding director exit reasons	A founder leaving after $1.1B raise is atypical. Could signal strategic disagreements.	Monitor leadership announcements and industry sources.
Crypto-to-AI asset quality	IPO investors will scrutinize data center assets from Arkon Energy era.	Watch IPO S-1 for asset valuations and depreciation schedules.