Private cloud vs public cloud: the real TCO, line by line
Ten-thousand-foot 'public cloud is expensive' takes are easy. Doing the math is harder. Here's the math, with every assumption visible.
Cloud repatriation became a popular term in 2024. Most articles about it skip the actual numbers. We're going to do the math.
Consider a mid-size B2B SaaS at $4M ARR, 800 customers, 12 application containers, a Postgres primary + read replica, an Elasticsearch cluster, and roughly 8TB of egress per month.
On the hyperscale side: m6i.2xlarge-class compute × 12 (containers), db.r6g.4xlarge primary + replica (Postgres), self-managed 3-node Elasticsearch on m6i.xlarge, S3 storage of ~12TB, CloudFront in front. Add data transfer, NAT gateway, Route 53, GuardDuty, CloudWatch.
Twelve-month public-cloud bill, full reserved instance discount applied: ~$168,000. Without reservations: ~$232,000.
Same workload on Ultiblob Pro tier × 4 (sharded by service) + Scale tier × 1 (database) + dedicated FlashArray slice + included egress: ~$77,000/yr. That includes the managed service — monitoring, patching, on-call.
The savings aren't from a magic discount. They're from removing four things: shared-tenancy overhead, NAT gateway and inter-AZ data charges, managed-service surcharges, and the operational headcount public cloud implicitly assumes you have.
Ultiblob's tradeoff: you give up some hyperscaler-style elasticity (we'll provision more capacity within 4 business hours, not 4 seconds) and some global-region breadth. For workloads that don't need either, the math is decisive.
Run your own numbers in the estimator, or send us your last public-cloud bill and we'll do the line-by-line for free.
