Revised Cost Model with Benchmarks

Evidence-based cost estimates benchmarked against actual UK government IT programmes including NHS NPfIT, Universal Credit, and IPA Major Projects data.

Currency Convention: This document uses £ (GBP) for UK-specific costs and benchmarks, and € (EUR) for coalition-wide totals. Local currency equivalents (CAD, AUD) shown in parentheses where applicable. Exchange rates approximate at time of writing.

Treasury Challenge: Cost Estimates

"Your revised figures are still 2x too low. Government IT programmes routinely exceed estimates by 3-5x. Why should this massive multinational coordination be any different?"

This document addresses this challenge with evidence-based estimates and explicit contingency allowances benchmarked against historic overruns.


Historical Benchmarks: UK Government IT Programmes

Before estimating sovereign cloud costs, we must understand the performance of comparable UK government programmes.

Programme Initial Estimate Final/Current Cost Overrun Status
NHS NPfIT £6.2B (2002) £12.7B (actual spend) +105% Dismantled 2011
Universal Credit £12.85B (2012) £15.8B+ (revised) +23% (ongoing) Deployed with issues
UK Borders e-Borders £750M ~£1.1B +47% Partial deployment
Police National Computer Replacement Baseline TBC +£400M overrun 5+ years delayed Ongoing issues
NAO Finding (NHS NPfIT): "The £2.7 billion spent so far on care records systems does not represent value for money... the government, and ultimately taxpayers, incurred significant costs for the program, including contract transition and exit costs which continued to accrue to a total amount of more than £10 billion."
Source: NAO Report
IPA Portfolio Statistics: "Digital and data projects are 60% more likely to be reporting 'Red' than the wider GMPP portfolio with only 9% reporting a 'Green' status... Average cost overrun for ICT was 27%."
Source: State of Digital Government Review

Derived Planning Assumptions

Metric Value Source
Average ICT cost overrun 27% IPA Major Projects Portfolio
Large programme overrun 50-100% NPfIT, Borders
Projects at high risk (Red/Amber-Red) 25% IPA Annual Report
Digital projects at high risk 60% more likely than baseline State of Digital Government

Infrastructure Costs: Datacenter Build and Hardware

Industry benchmarks for datacenter construction (2024-2025 data):

Component Cost Metric Range Source
Greenfield datacenter (Tier III) Per MW $8M - $12M Cushman & Wakefield 2025
Government/secure facility Per MW $12M - $15M (premium) Estimate (security overhead)
Per rack (Tier III) Build cost $40K - $60K Dgtl Infra
Per rack (Tier IV/Secure) Build cost $60K - $100K+ Industry benchmark
Construction per sq ft Standard market $600 - $1,100 Turner & Townsend
Construction per sq ft Prime markets ~$2,200 Major city premium

UK Sovereign Cloud Infrastructure Estimate

Assuming 40MW initial capacity across 3-4 sites (Canberra model scaled):

Item Quantity Unit Cost Total (GBP)
Datacenter capacity (40MW @ secure rate) 40 MW £10M/MW £400M
Compute hardware (2,000 racks) 2,000 racks £150K/rack avg £300M
Storage infrastructure (100PB) 100 PB £2M/PB £200M
Networking (spine/leaf, WAN) Lump sum - £100M
Security infrastructure (HSM, firewalls, SIEM) Lump sum - £75M
Site preparation, power, cooling Lump sum - £125M
Infrastructure Subtotal (UK only) £1.2B
Contingency (50% - per benchmark) - - £600M
UK Infrastructure Total (with contingency) £1.8B

Migration and Application Refactoring Costs

Migration costs vary dramatically by strategy. Industry benchmarks:

Strategy Cost Per Workload Complexity Risk
Rehosting (lift-and-shift) $200 - $500 per VM Low High post-migration costs
Replatforming $500 - $1,500 per workload Medium Moderate
Refactoring $10,000+ per complex app High Highest upfront cost
Enterprise refactoring $200,000 - $500,000+ per app Very High Requires skilled teams

Critical Correction

Previous estimates of €500M-2B for application refactoring were severely understated. Based on NHS NPfIT experience (£12.7B for a smaller scope) and enterprise refactoring benchmarks, a more realistic range is:

  • Simple rehosting (20% of apps): £500M - £1B
  • Replatforming (50% of apps): £2B - £4B
  • Full refactoring (30% of apps): £3B - £8B
  • Total application work: £5.5B - £13B (UK only)

UK Migration Cost Estimate

Category Estimate Range Midpoint
Application discovery and assessment £50M - £100M £75M
Migration tooling and automation £100M - £200M £150M
Application refactoring (estimated 5,000+ apps) £5B - £12B £8B
Data migration and validation £200M - £500M £350M
Testing and certification £300M - £600M £450M
Rollback and contingency systems £150M - £300M £225M
Migration Subtotal (UK) £5.8B - £13.7B £9.25B
Contingency (40% - per previous estimate) - £3.7B
UK Migration Total £8B - £19B £13B

Staffing and Training Costs

Previous Estimate Correction

Training estimate of €100-300M for "tens of thousands of staff" implied €2-10K per person. Government IT transformation training typically costs £40,000-£60,000 per person for full reskilling (certifications, hands-on, mentoring over 12-24 months).

Category FTE Count Cost/FTE (over 3 years) Total
Core platform team (permanent) 500 £350K (salary + benefits) £175M
Migration specialists (contract) 2,000 £450K (contractor rate) £900M
Security specialists 200 £400K £80M
Programme management 150 £400K £60M
Training and reskilling (10,000 civil servants) 10,000 £50K £500M
Staffing Subtotal (UK, 3 years) £1.715B
Premium for skills shortage (25%) £429M
UK Staffing Total (3 years) £2.1B

Total Programme Cost Summary (UK)

Category Low Estimate Mid Estimate High Estimate
Infrastructure (with 50% contingency) £1.4B £1.8B £2.4B
Migration and Refactoring (with 40% contingency) £8B £13B £19B
Staffing (3 years with premium) £1.6B £2.1B £2.8B
Platform development (36 capabilities) £0.8B £1.2B £1.8B
Programme management and governance £0.3B £0.5B £0.8B
International coordination (SCCA contribution) £0.1B £0.15B £0.2B
UK Total (Initial 3-5 Years) £12.2B £18.75B £27B

Scenario Analysis

Best Case

£12.2B

Assumes: successful reuse of existing assets, minimal refactoring needed, no major delays, full staff retention, no vendor disputes.

Probability: 10%

Most Likely

£18.75B

Assumes: typical government IT overruns (27% avg), moderate refactoring, some delays, contractor premium during skills shortage.

Probability: 60%

Worst Case

£27B+

Assumes: NPfIT-style overruns (100%+), extensive refactoring, major delays, vendor disputes, scope creep, political interference.

Probability: 30%

Four-Nation Total Estimate

Applying proportional scaling based on GDP and cloud spending:

Jurisdiction Scaling Factor Low Mid High
United Kingdom 1.0x (base) £12.2B £18.75B £27B
European Union (aggregate) 4.5x €65B €100B €145B
Canada 0.6x €8.5B (CAD 12.5B) €13B (CAD 19B) €18.5B (CAD 27B)
Australia 0.4x €5.7B (AUD 9.5B) €9B (AUD 15B) €12.5B (AUD 21B)
Four-Nation Total - €94B €145B €210B

Cost Reality Check

The "most likely" four-nation estimate of €145B (~$165B USD) represents a significant undertaking. For context:

  • NHS NPfIT: £12.7B for one healthcare system
  • EU IPCEI-CIS: €2.6B (much smaller scope)
  • This programme: entire government IT estate for 4 jurisdictions

This is 2-3x previous estimates but reflects realistic benchmarks. Previous estimates were dangerously optimistic and would have failed Treasury scrutiny.

10-Year TCO Comparison: Status Quo vs Sovereign

To justify the investment, compare against continued US cloud spending:

Scenario UK 10-Year Cost 4-Nation 10-Year Cost
Status Quo (AWS/Azure/GCP/OCI)
- Continued cloud spending (3% annual growth) £12B - £15B £85B - £110B
- Egress/lock-in penalties (exit later) £2B - £5B £15B - £35B
- Sovereignty risk exposure (unquantified) Unknown Unknown
Status Quo Total £14B - £20B £100B - £145B
Sovereign Cloud
- Initial investment (3-5 years) £18.75B £122B
- Ongoing operations (years 4-10) £4B - £6B £30B - £45B
Sovereign Total £23B - £25B £152B - £167B
Premium for Sovereignty £5B - £9B (30-50%) £20B - £50B (15-35%)
Investment Case: The 15-50% premium for sovereignty is the cost of eliminating existential dependency on US-controlled infrastructure. This is comparable to the premium nations pay for domestic defence manufacturing versus foreign procurement.

VERSION 3.0: Phased Approach (Recommended)

Following stakeholder feedback, the recommended approach is now a dual-track phased model rather than full upfront commitment. This addresses the key criticism that committing £12-27B without proven UK-scale success is unacceptable risk.

Track A: Emergency Sovereign Capability (36 months)

Achieves basic sovereign infrastructure for critical government services, providing insurance against US cloud denial while validating the approach before full commitment.

Category Low Mid High
Core infrastructure (2 UK sovereign datacenters) £400M £600M £800M
Platform development (essential services only) £300M £450M £600M
Migration of Tier 1 workloads (critical services) £800M £1.2B £1.6B
Staffing and training (36 months) £500M £700M £900M
Programme management and governance £100M £150M £200M
Contingency (25%) £525M £775M £1.025B
Track A Total (UK, 36 months) £2.6B £3.9B £5.1B
Real-World Benchmark: Schleswig-Holstein (Germany)
30,000 employees migrating from Microsoft to LibreOffice/Linux. Investment: €9M. Projected savings: €15M/year. ROI achieved in 8 months.
Source: Schleswig-Holstein Government

Track B: Full Migration (Additional 4-5 years)

Only proceeds after Track A demonstrates success at Gate 3 review. Migrates remaining government workloads including complex Oracle systems.

Category Low Mid High
Additional infrastructure capacity £800M £1.2B £1.6B
Tier 2-4 workload migration £4B £7B £11B
Oracle/complex system migration £2B £4B £7B
Extended staffing (5 years) £1.5B £2.5B £3.5B
Contingency (40%) £3.3B £5.9B £9.2B
Track B Total (UK, additional 5 years) £11.6B £20.6B £32.3B

Combined Phased Total

Track A Only

£2.6B - £5.1B

Emergency capability achieved. Full migration deferred or cancelled based on results.

Duration: 36 months

Track A + Partial B

£8B - £15B

Emergency capability plus migration of high-value workloads. Oracle systems remain hybrid.

Duration: 5-6 years

Full Migration

£14B - £37B

Complete sovereign migration including all Oracle systems. Only if Track A proves successful.

Duration: 7-8 years

Key Advantage of Phased Approach

Maximum exposure at any decision point: £5.1B (Track A high estimate).

The phased approach means the government never commits more than £5B without demonstrated success. This addresses the primary Treasury concern about NPfIT-style runaway costs.

Treasury Expected Value Framework

The investment case can be evaluated using expected value calculation, comparing the cost of mitigation against the probability-weighted cost of US cloud denial.

Variable Low Estimate High Estimate
Probability of US service denial (5 years) 1% 10%
Impact of denial (UK GDP disruption) £50B £200B
Expected loss without mitigation £0.5B £20B
Cost of mitigation (Track A) £2.6B £5.1B
Break-even probability 2-5% (if impact is £50-200B)
Investment Decision Framework: If the probability of US cloud denial exceeds 2-5% over the next 5 years, Track A investment has positive expected value even before considering non-financial benefits (sovereignty, skills development, reduced vendor lock-in).

Comparable Government Insurance Decisions

Programme Annual Cost Threat Probability Rationale
UK nuclear deterrent ~£6B/year Very low Existential threat insurance
Pandemic preparedness (pre-COVID) ~£100M/year Low (assessed) Now seen as underinvestment
Cyber security (NCSC budget) ~£200M/year High Active threat environment
Sovereign Cloud (Track A amortised) ~£1.3B/year 1-10% (assessed) Digital infrastructure insurance

Document Status

Version 3.0 incorporates stakeholder feedback recommending a phased approach with gate-controlled budgets. Previous estimates (£12-27B) assumed full upfront commitment; the revised model limits exposure to £5B before demonstrated success.

This cost model incorporates lessons from NHS NPfIT, Universal Credit, and IPA Major Projects data. All estimates include explicit contingency allowances based on historical performance.

Key sources:

Version: 3.0 (Phased Approach) | Last updated: January 2026

Back to main documentation