
Running Octopus SaaS and the On-Prem Dilemma

CI/CD with Robert Erez
Topic 9 of 10
Published Jun 17, 2026 · Clip 59:39–1:09:25 (9:46)
Rob recounts Octopus SaaS history: the first VM-per-customer experiment lost money (~$100/customer/month cost vs ~$20 revenue), then a rebuild on Kubernetes with cell-based 'reef' architecture where each customer runs in a pod. He describes current efforts to make the deployment process more resilient and reduce upgrade downtime. He discusses the challenges of supporting both SaaS and on-prem: on-prem upgrades take ~200 days for 50% adoption and some customers run 5-7 year old versions, yet the majority of customers (banks, governments) still demand self-hosting for control and compliance, which also reduces competition.