I was referring to running your own DB on hardware with NVMe drives. Obviously you lose every nicety of managed services, but tooling exists to replace it, and

I was referring to running your own DB on hardware with NVMe drives. Obviously you lose every nicety of managed services, but tooling exists to replace it, and you gain stupid amounts of performance. RDS Multi-AZ Cluster gives you much of the advantages of Aurora, but with higher performance and more tuning capabilities, though you are limited to 3 nodes. Tbf 3 nodes is almost certainly enough for most companies. A few hundred thousand QPS would be easily handled by that. Re: cross-region read replicas, eh… if you’ve somehow managed to ensure that every single aspect of your app is capable of withstanding the loss of an entire region – including us-east-1, since most of the control plane functions are there – then sure, maybe. But do you need it? If an entire AWS region drops out, half of the internet goes with it, and you can just blame that. I doubt the small possibility of higher uptime is worth the literal doubling in monthly costs.