On-Prem is the new Black
summary
Argues that cloud repatriation is rising because unchecked cloud spend and operational limits bite: one example shows a $40K bill from a semi‑managed Elasticsearch cluster. Recommends moving cost‑heavy workloads like GPUs on‑prem (hardware ROI over 4–5 years vs recurring cloud spend), standardizing on Kubernetes and terraform/CDK, using CI pipelines that pull L1 images, produce baked artifacts (L2) into local registries, test with Vagrant, isolate an air‑gapped secrets network, and prefer managed colo (Equinix, Hetzner) over unreliable remote‑hands.
AI-written summary · about
chapters (18)
- 0:00Introduction and background
- 1:32Cloud cost pitfalls and surprises
- 5:08Real-world cloud billing disaster
- 7:16When cloud makes sense
- 9:04Cloud costs vs hardware ROI
- 10:31Choosing between cloud and on-prem
- 12:44Hybrid and multi-cloud strategies
- 13:28Hardware knowledge gap
- 15:40Reverse migration challenges
- 16:43Cloud as an operating model
- 17:46Standardizing infrastructure tools
- 19:12SLA inversion in dev environments
- 20:14Twelve-factor app methodology
- 21:40Artifacts and deployment pipelines
- 23:49Cattle vs snowflake infrastructure
- 24:59Three key considerations
- 26:04Balancing cloud and on-prem
- 26:47Q&A on remote hands and hardware
related talks
05mOn Prem is the new Black
37mHybrid OnPrem + Cloud Architectures
30mOn-premise data centers do not need to be legacy
20mAre the days of on-premise servers over?
26mWhen the Cloud s Reign is Over
30mFrom Cloud to On-Prem - Capacity Planning and CPU Analysis
25mBreaking Free From the Cloud: Banking on Self-Hosted Kubernetes
28mScam or Savings? A Cloud vs. On-Prem Economic Slapfight