Coalition Strategy

Strategic Considerations

Analysis of coalition dynamics, current datacentre landscape, infrastructure requirements, and supply chain leverage. Essential context for programme planning.

---

---

1. Gives and Gets Framework

Coalition partnerships require clear articulation of what each party contributes ("gives") and what they receive ("gets"). This framework, common in MoD and large programme delivery, ensures mutual benefit and sustained commitment.

Coalition Partner Gives and Gets

Partner	Gives (Contributes)	Gets (Receives)
🇬🇧 United Kingdom	Programme secretariat and coordination role Existing Crown Hosting infrastructure (sites supporting high-classification workloads per GOV.UK framework) Five Eyes intelligence sharing (threat context for coalition planning) ARM ecosystem and talent (Cambridge/Bristol clusters) Financial services cloud expertise (FCA-regulated workloads) G-Cloud framework as procurement template	Access to EU single market cloud providers post-Brexit Shared R&D costs (not bearing full development alone) Hardware supply chain via Japan/Korea partnership Collective bargaining power with vendors Political cover (not unilaterally exiting US cloud) European DC locations for DR/latency
🇪🇺 European Union	Gaia-X interoperability standards European cloud providers already at scale (OVHcloud, Scaleway, IONOS) ~€15-25B/year government cloud market (largest in coalition) Regulatory weight: GDPR as global standard-setter Manufacturing base for datacentre construction Energy grid interconnection for DC power resilience	Shared infrastructure R&D costs across more jurisdictions Potential for enhanced bilateral security cooperation with UK/CA/AU (not Five Eyes membership; subject to existing treaties) Political solidarity for US cloud exit (collective vs unilateral action) Commonwealth procurement frameworks access English-language technical talent pool (UK/CA/AU) ARM ecosystem expertise (UK Cambridge cluster)
🇨🇦 Canada	Bilingual (EN/FR) implementation capability North American timezone coverage for 24/7 ops SSC datacentre consolidation experience (lessons learned) Arctic/northern resilience experience Strong open source community (Shopify, etc.) Privacy Act compliance frameworks	Reduced US economic dependency Alternative to US cloud providers European market diversification Coalition political support for sovereignty stance Continued participation in allied intelligence frameworks (per existing Five Eyes arrangements) Technology transfer opportunities
🇦🇺 Australia	CDC Data Centres sovereign hosting model PSPF/ISM security framework expertise Asia-Pacific timezone coverage Mining/resources sector cloud experience Hosting Certification Framework template Five Eyes intelligence contribution	Reduced AUKUS/US dependency concerns European technology alternatives Northern hemisphere DR/resilience options Coalition scale for vendor negotiations Shared platform development costs Access to European cloud providers
🇯🇵 Japan POTENTIAL	Hardware manufacturing (NTT, Fujitsu, NEC) Enterprise software expertise Asia-Pacific datacentre locations DFFT (Data Free Flow with Trust) framework ~¥500-800B/year government cloud market Advanced robotics/automation for DC ops	Reduced US technology dependency Coalition support for China contingencies European market access Shared security standards with Five Eyes Alternative to US hyperscalers Diversified supply chain relationships
🇰🇷 South Korea POTENTIAL	Memory chips: Samsung + SK Hynix = ~70% global DRAM HBM (AI memory): SK Hynix = 53-62% market share 5G/networking infrastructure (Samsung) Display technology Semiconductor manufacturing capacity ~₩4-6T/year government cloud market	Diversified export markets beyond US/China Coalition support for regional security European technology partnerships Reduced dependency on US market access Alternative cloud infrastructure options Political alignment on sovereignty

Critical Supply Chain Insight

South Korea's memory chip dominance is strategically critical. Samsung and SK Hynix together control ~70% of global DRAM production. SK Hynix alone holds 53-62% of the HBM (High-Bandwidth Memory) market—essential for AI workloads. Coalition access to this supply chain provides significant leverage.

Sources: Market share figures from industry reports (TrendForce, Omdia quarterly DRAM/HBM trackers). See also: Statista, KED Global. Note: Precise figures fluctuate quarterly; verify current data from primary research firms.

---

2. Current Government Datacentre Landscape

Understanding the current state of government datacentre infrastructure is essential for planning sovereign cloud migration. Each jurisdiction has different consolidation trajectories and challenges.

🇬🇧 United Kingdom

Crown Hosting Data Centres

Structure: Joint venture (75% Ark Data Centres, 25% Government) established March 2015.

Current Locations

Site	Location	Classification	Notes
Corsham (Spring Park)	Wiltshire	Up to TOP SECRET	Former MoD underground facility
Farnborough (Cody Park)	Hampshire	Up to SECRET	Primary Crown Hosting site
North London (Meridian Park)	London	OFFICIAL	Commercial colocation

Contract: £250m framework renewed November 2022 for 7 years. First users: DWP, Home Office, Highways Agency.

Sources: Computer Weekly, Crown Hosting, GOV.UK

UKCloud Collapse Warning: The liquidation of UKCloud in October 2022 caused "real business continuity issues" for central government (Cabinet Office CCO Andrew Forzani). Government was £17.5m out of pocket. This demonstrates the risk of dependency on any single provider.

Sources: The Register, Computer Weekly

Oracle UK Government Cloud Migration

Current trajectory: Significant government migration to Oracle Cloud Infrastructure (OCI).

• Dual-region setup: London + Newport (Wales) for DR
• MoD contract: Migrating legacy systems to OCI for AI readiness
• Bank of England: Migration costs tripled from £7m to £21.5m
• Oracle UK investment: $5 billion over 5 years announced March 2025
• Classification: Oracle claims UK Sovereign Cloud designed for OFFICIAL-SENSITIVE workloads (verify current accreditation status via NCSC Cloud Security Principles)

Sovereignty concern: Oracle is a US company subject to CLOUD Act. The "sovereign" designation refers to data location and staff clearance, not legal jurisdiction.

Sources: Oracle UK Sovereign Cloud, DCD, The Register

🇨🇦 Canada

Shared Services Canada (SSC) Consolidation

Mandate: Consolidate ~720 legacy datacentres into modern Enterprise Data Centres (EDCs).

Consolidation Progress

Fiscal Year	Legacy DCs Closed	Notes
2023-24	50	Reduced GHG emissions
2024-25	25	Ongoing consolidation

Key Challenges (Auditor General 2023)

• Nearly two-thirds of GC applications in "poor health" needing modernisation
• SSC manages infrastructure; departments own applications
• Legacy systems cannot be decommissioned until applications are modernised
• 2024 GC Application Hosting Strategy: balance cloud flexibility with EDC reliability

Sources: SSC Data Centre Consolidation, SSC 2024-25 Departmental Plan

🇦🇺 Australia

CDC Data Centres (Sovereign Hosting)

Structure: Canberra Data Centres (CDC) - pre-eminent sovereign datacentre provider. Shareholders: Future Fund (34.55%), Infratil NZ (49.75%), Commonwealth Super Corporation (12.04%).

Current Footprint

• 20 datacentres across Australia and New Zealand
• First to achieve Certified Strategic status under Hosting Certification Framework
• Highest security accreditations under AU/NZ government frameworks

2025 Expansion

• Marsden Park (NSW): $3.1B hyperscale facility approved - 504MW capacity (largest in Southern Hemisphere)
• Belmont (WA): $415M, 200MW facility announced
• Data Centres Australia (DCA): New industry body launched November 2025

Sources: CDC Data Centres, ACS Information Age, NSW Government

🇪🇺 European Union

Gaia-X and National Initiatives

Gaia-X Status (2025)

• 180+ data spaces in development
• First multi-provider catalogue: 600 services from 15 providers
• Four sovereignty levels - Level 3 (highest) restricted to EU-headquartered providers only
• ~10% of use cases require Level 3 (critical infrastructure, defence)

National Sovereign Cloud Initiatives

Country	Initiative	Technology
🇫🇷 France	NUBO	OpenStack-based private cloud for sensitive data
🇩🇪 Germany	Bundescloud / OpenDesk	Sovereign digital tools consortium (DE/FR/IT/NL)
🇳🇱 Netherlands	Rijkscloud	Government private cloud

European Cloud Providers

• OVHcloud (France) - G-Cloud listed, Gaia-X Level 3 eligible
• Scaleway (France) - Iliad Group subsidiary
• IONOS (Germany) - United Internet subsidiary, G-Cloud listed
• StackIT (Germany) - Schwarz Group (Lidl parent)
• Exoscale (Switzerland) - A1 Telekom Austria subsidiary

Sources: Gaia-X, The Register, Taylor & Francis

---

3. "Cloud Native" Lock-in Risk

The push for "cloud native" architectures, while offering operational benefits, creates deep vendor lock-in that makes sovereign migration significantly more difficult and expensive.

The Lock-in Problem

Service Type	AWS Example	Lock-in Severity	Migration Complexity
Serverless Functions	AWS Lambda	SEVERE	Complete rewrite required. No portable equivalent.
Managed NoSQL	DynamoDB	SEVERE	Proprietary data model. Schema redesign needed.
Message Queues	SQS/SNS	HIGH	API changes; can migrate to RabbitMQ/Kafka.
Identity	Cognito, IAM	HIGH	Deep integration. Keycloak alternative exists.
Object Storage	S3	MEDIUM	S3 API is de facto standard. MinIO compatible.
Containers	EKS (Kubernetes)	LOW	Kubernetes is portable. Main work is cluster setup.
Relational DB	RDS PostgreSQL	LOW	Standard PostgreSQL. Export/import straightforward.

Sovereign-Ready Architecture Principles

For new systems: Design for portability from day one.

• Use Kubernetes for orchestration (not ECS/Fargate)
• Use PostgreSQL/MySQL (not DynamoDB/Aurora Serverless)
• Use S3-compatible object storage API
• Use OpenTofu for infrastructure-as-code (not Terraform - now IBM/US owned)
• Avoid proprietary managed services where open alternatives exist
• Design for multi-cloud from inception

---

4. Power & Infrastructure Requirements

Datacentre power requirements have changed dramatically with AI workloads. Understanding power density is essential for capacity planning.

Power Density by Workload Type

Workload Type	Power per Rack	Cooling	Notes
Traditional enterprise	4-6 kW	Air cooling	Typical government workloads
Cloud/virtualised	10-14 kW	Air cooling	Higher density VMs
High-performance compute	15-30 kW	Enhanced air/liquid assist	Scientific workloads
AI training/inference	40-100 kW	Liquid cooling typically required	GPU clusters (NVIDIA H100/H200); varies by configuration
High-density AI	100-200 kW	Direct liquid cooling	Advanced deployments; higher densities under evaluation

Facility-Level Power Requirements

Facility Type	Typical Power	Equivalent
Small government DC	1-5 MW	500-2,000 servers
Mid-size enterprise DC	5-20 MW	2,000-10,000 servers
Large/hyperscale DC	50-100+ MW	Small power station output
Mega DC (e.g., CDC Marsden Park)	500+ MW	Major industrial facility (per planning approval)

Power Efficiency (PUE)

• Hyperscale best-in-class: 1.09-1.20 PUE (Google reports 1.09 fleet-wide)
• Modern colocation: 1.3-1.6 PUE
• Enterprise legacy: 1.5-2.0+ PUE
• Target for sovereign facilities: ≤1.4 PUE

Capacity Planning Calculation

Example: 10,000 traditional servers at 5kW average = 50MW IT load. At 1.4 PUE = 70MW total facility power.

With AI capability: Add 500 AI-capable racks at 60kW average = 30MW additional. Total: 100MW facility (excluding N+1 redundancy).

Note: Power figures are illustrative. Actual requirements vary significantly by vendor, configuration, and cooling approach. Consult vendor specifications for planning.

Sources: Dgtl Infra, DCK, Uptime Institute

---

5. Supply Chain Leverage Analysis

Coalition partners, particularly Japan and South Korea, hold significant leverage in the global semiconductor supply chain. This creates both opportunities and strategic considerations.

Memory Chip Market Concentration

DRAM Share (Q3 2025)

HBM Share (Q3 2025)

Company	Country	DRAM Share (approx.)	HBM Share (approx.)
SK Hynix	🇰🇷 South Korea	34%	53%
Samsung	🇰🇷 South Korea	33%	35%
Micron	🇺🇸 United States	25%	11%
South Korea Total		~67%	~88%

Strategic Leverage: South Korean companies control ~88% of the HBM market (essential for AI) and ~67% of DRAM. SK Hynix is Nvidia's primary HBM supplier. Coalition partnership with South Korea provides significant supply chain leverage.

Could Japan/South Korea Limit US Hardware Supply?

SPECULATIVE SCENARIO - LOW PROBABILITY: The following analysis explores a theoretical scenario. Current geopolitical alignment, treaty obligations, and trade dependencies make this unlikely in the near term. Included for completeness, not as a planning assumption.

Factors Affecting Japan/South Korea Hardware Supply Leverage

Factor	Assessment
Current Export Controls	Japan and South Korea have aligned with US export controls targeting China (Chip4 Alliance). They participate in restricting advanced chipmaking equipment to China. Source: Congressional Research Service R48642
US-Korea Trade Deal (Dec 2025)	USTR announced implementation of US-Korea Strategic Trade and Investment Deal in December 2025, reinforcing bilateral trade cooperation. This deepens, not loosens, US-Korea economic ties.
Manufacturing Dependency	Samsung and SK Hynix have significant manufacturing in South Korea, but also: Facilities in China (subject to export restrictions) Expanding US presence (Samsung Texas fab, SK Hynix Indiana) Supply obligations to US customers under long-term contracts
Risks of Supply Restriction	Any attempt to limit US supply would face: WTO/trade retaliation: US countermeasures on Korean/Japanese exports Contract breach: Legal liability to US customers Security treaty implications: US military presence in both countries Chip4 Alliance obligations: Existing coordination commitments Investor reaction: Share price impact on Samsung/SK Hynix
Realistic Coalition Value	Rather than supply restriction, Japan/South Korea partnership value lies in: Guaranteed allocation during shortages (coalition priority) Joint R&D on sovereign hardware specifications Diversified supply chain (not single-source dependency) Reduced lead times through pre-negotiated frameworks This is achievable within existing trade structures.

Japan Hardware Manufacturing Capability

Company	Products	Coalition Value
Fujitsu	Servers, storage, ARM processors (A64FX)	Non-x86 compute alternative; Fugaku supercomputer technology
NEC	Servers, networking, vector processors	Enterprise hardware; vector computing for HPC
NTT	Networking equipment, datacentre infrastructure	Telecom-grade infrastructure; IOWN photonic networking
Renesas	Microcontrollers, embedded processors	IoT and edge compute chips
Tokyo Electron	Semiconductor manufacturing equipment	Fab equipment (controlled under export restrictions)

---

6. Alignment with GDS Discovery/Alpha/Beta

UK government digital services follow the GDS Service Standard phases. Sovereign cloud migration should align with these established delivery patterns.

Current State: "Keeping the Lights On"

Many government organisations are focused on maintaining legacy systems rather than strategic migration. The 2023 Auditor General of Canada report found nearly two-thirds of GC applications were in "poor health" and critically needed modernisation. Similar patterns exist across other jurisdictions.

Proposed Sovereign Cloud Discovery Framework

Phase	Duration	Sovereign Cloud Focus	Outputs
DISCOVERY	4-6 weeks	Complete US cloud dependency audit Map data sovereignty requirements Identify CLOUD Act exposure Assess current architecture portability Catalogue proprietary service usage	Dependency inventory Risk register Migration complexity assessment Recommended approach
ALPHA	6-8 weeks	Prototype sovereign platform deployment Test migration of non-critical workload Evaluate European provider options Validate security controls Prove data portability	Working prototype Provider comparison Technical feasibility confirmation Cost model validation
PRIVATE BETA	3-6 months	Migrate first real service Establish operational procedures Security accreditation (provisional) Limited user testing Runbook development	Production-quality service Security assessment Operational playbooks Lessons learned
PUBLIC BETA/LIVE	Ongoing	Scale migration across services Full operational capability Complete security accreditation Continuous improvement	Fully sovereign services US cloud exit complete Operational metrics

Sources: GDS Way, dxw Playbook, DfE Service Standard

---

7. Datacentre Decision Risk Framework

A framework for assessing risks associated with datacentre infrastructure decisions, including retention, closure, or expansion of existing facilities.

Risk Assessment Matrix

Risk Category	Keep Open	Close/Migrate	Expand/Modernise
Sovereignty Risk	LOW Data remains under national jurisdiction	HIGH (if to US cloud) LOW (if to sovereign alternative)	LOW Enhanced sovereign capability
Operational Risk	MEDIUM Legacy equipment failure risk	MEDIUM Migration execution risk	LOW-MED Construction/transition risk
Cost Risk	MEDIUM Rising maintenance, inefficiency (high PUE)	MEDIUM Migration costs, potential stranded assets	HIGH Significant CapEx required
Capacity Risk	HIGH Cannot scale for AI workloads	LOW Cloud offers elastic capacity	LOW Purpose-built for future needs
Skills Risk	MEDIUM Legacy skills harder to find	LOW-MED Cloud skills more available	LOW-MED Modern skills investment
Environmental Risk	HIGH Poor PUE, high emissions	LOW Modern facilities more efficient	LOW Build to modern efficiency standards

Decision Criteria Checklist

Before Closing a Government Datacentre

---