Why most cloud TCO models are wrong
The bill is one number. Real public-cloud TCO includes:
- Egress charges — often 5-15% of total spend, easy to miss
- Reservation under-utilization — 30-50% realization is common; the discount evaporates
- Idle and over-sized resources — 10-20% of compute spend
- Hyperscaler-managed-service premium — 2-4× over self-managed at scale
- Vendor lock-in switching cost — invisible until something breaks
- Platform-engineering capacity dedicated to public-cloud-specific complexity
A real TCO model captures all of these.
Destination cost — what’s missed
For private-cloud destination:
- Hardware acquisition + 5-year refresh (the refresh cliff arrives)
- Datacenter or colocation
- Network bandwidth, including egress between sites
- Storage tiering and growth
- Backup and DR infrastructure
- Identity, observability, platform tooling
- Platform-engineering capacity to operate
- Software licenses where applicable
Skip these and the TCO looks artificially good; reality bites in Year 2.
Sensitivity analysis
Honest TCO model is sensitive to occupancy assumptions:
- Steady-state at 50% utilization vs 80% changes private-cloud-economics dramatically
- Workload growth at 20%/year vs 50%/year changes hardware refresh cycles
- Egress volume changes have outsized impact
Workload-level decisions
TCO at portfolio level may be neutral. TCO at workload level usually shows:
- Top-10 workloads: 60-80% of cost case
- Tail of workloads: marginally cost-positive or neutral
Repatriating top-10 workloads, leaving tail in cloud, often best.
How to use the worksheet
Fill in your numbers, walk through with finance partner + platform engineering. Identify top-10 repatriation candidates. Validate assumptions.
For full engagement see Cloud repatriation services.
Test yourself: honest cloud repatriation TCO
5 questions · ~2 min
Question 1 / 5
What share of public-cloud spend does egress typically eat, according to the article?
Why: Egress is often 5-15% of total spend and easy to miss. Combined with reservation under-utilization (30-50% realization is common), idle/over-sized resources (10-20% of compute), hyperscaler managed-service premium (2-4× over self-managed), and lock-in switching cost — the bill number alone understates real TCO.
Question 2 / 5
What reservation realization rate does the article say is common?
Why: 30-50% reservation realization is common — meaning the discount evaporates. A reservation only saves money if you actually use what you committed to. Mismatch between commitment portfolio and actual workload mix is a typical source.
Question 3 / 5
What share of repatriation cost case do the top-10 workloads typically represent?
Why: TCO at workload level usually shows the top-10 workloads = 60-80% of cost case; tail workloads are marginally cost-positive or neutral. Repatriating top-10 and leaving tail in cloud is often best.
Question 4 / 5
What does the article say about destination cost — what gets missed in shallow models?
Why: Skipping destination items makes TCO look artificially good; reality bites in Year 2. Full destination cost includes hardware acquisition + 5-year refresh, datacenter/colocation, network bandwidth, storage tiering, backup/DR, identity/observability/tooling, platform-engineering capacity, software licenses.
Question 5 / 5
Which sensitivity factor has "outsized impact" on private-cloud economics?
Why: Honest TCO is sensitive to occupancy assumptions: steady-state at 50% vs 80% utilization changes private-cloud economics dramatically. Workload growth (20%/year vs 50%/year) changes hardware refresh cycles. Egress volume changes have outsized impact.
0 / 5