cloud-provider-tradeoffs

Identify the two or three managed services your architecture depends on and compare those in depth, since the commodity layer (compute, storage, load balancers) is equivalent across providers. Then weight team skill, enterprise agreements, data location, and pricing model fit for your workload shape.

BigQuery is serverless and bills per query scanned, suiting sporadic analytics with zero cluster management. Redshift is a provisioned cluster that is cheaper under steady, predictable load but requires sizing and tuning. Model your actual query pattern before deciding.

Data gravity means large datasets attract compute because egress is billed per gigabyte and cross-cloud transfer is slow and expensive. Decide where your data lives first and place compute beside it, since moving later becomes a major project.

Yes, for container-heavy teams the managed Kubernetes maturity matters more than underlying VM prices. GKE is generally considered the most mature, with stronger autoscaling and upgrade automation, which reduces operational burden over time.

Multi-cloud is genuinely driven by sovereignty requirements, regional availability gaps, or a truly unique service, not by a default desire to stay flexible. Splitting compute and data across clouds without a forcing requirement creates egress costs and operational complexity.

Published rates ignore pricing model shape: serverless billing favors spiky workloads while provisioned capacity favors steady load. The same architecture can be cheapest on different clouds depending only on its usage curve, so model your actual workload.