Here's a stat that should keep anyone up: a single 2017 typo at Amazon S3 knocked a big chunk of the web offline for hours. That's not an argument against the cloud existing. It's an argument against treating one provider as your backup. If all your copies live inside the same ecosystem, you don't have redundancy, you have correlated failure. One billing dispute, one region outage, one ransomware breach, one policy change, and everything goes dark at once. The cloud is fine as one layer. It's fragile as the only layer. I'd say robustness means something offline, tested, and outside their control.

Imagine a firm with primary data and backups in one cloud account. A credential gets phished, ransomware deletes the live buckets and the snapshots. Multi-region replication? It copied the deletion. That's the fragility: not the disk, the shared control plane. You're right a lone external drive is also fragile. But that's my point. Robustness needs layers that fail independently. If every copy answers to the same provider, same account, same policy switch, you don't have redundancy. You have one point of failure wearing three hats.

Here's one for you: in 2017 GitLab lost production data while running five backup mechanisms, including S3 snapshots. All five failed at once. That's the whole argument. You say fix it with encryption and mirroring across providers. Who holds the keys, the accounts, the pipelines? The same control plane. A single phished credential or a policy switch still cascades. Your offline drive critique is fair, and that's why I want it as one layer, not the only one. But layers that share one failure domain aren't layers. Robustness means something that fails independently, even when the provider doesn't.