Cloud is not always cheaper than on-prem. This is now widely acknowledged — yet most UK mid-market organisations still under-model on-prem TCO + over-model cloud TCO when comparing. This is the honest framework Servnet runs.
The headline maths is misleading
Vendor marketing: "Cloud is 30-60% cheaper than on-prem". Sometimes true. Often not.
When cloud wins: variable workloads (peak vs average gap > 3×), greenfield with no existing infrastructure, workloads that genuinely use managed services (RDS, Cosmos DB, Lambda).
When on-prem wins: stable workloads with predictable utilisation, GPU-intensive AI training, regulatory environments where data sovereignty + control matter.
What people forget about on-prem TCO
Hardware + maintenance + power + cooling + rack + floor space + opex IT staff are all real costs.
Refresh cycles every 5-7 years — annualise the capex over depreciation period.
Idle capacity — most on-prem runs at 30-50% utilisation. Cloud charges only what you use.
What people forget about cloud TCO
Egress fees — moving data out of cloud is expensive. AWS / Azure egress to on-prem or another cloud routinely 30-50% of cloud spend at scale.
Managed services + premium support — Reserved Instances + Savings Plans negotiate but the headline pricing isn't what most pay.
Architecture overhead — VPC peering, transit gateways, identity federation, network policy. Operational complexity that didn't exist on flat on-prem networks.
The honest 5-year model
For most UK mid-market (100-2,000 employees) with stable mixed workload: cloud + on-prem hybrid is typically 15-25% MORE expensive than pure on-prem over 5 years.
The trade-off is flexibility, geographic resilience, and operational shift from capex to opex. Servnet models both sides honestly — including overlooked items like break-fix + spares stock on-prem and cloud egress + reserved-instance economics.
Modern HCI + good operational discipline narrows the on-prem gap further — and our HCI ROI threshold model shows exactly when HCI flips ahead of 3-tier. For workloads that genuinely need cloud burst capacity, see SaaS vs IaaS vs PaaS for the operating-model decision.