Migrating workloads from on-prem to AWS in 2026 - what's changed?

Hey folks, been architecting a pretty large migration for a client (roughly 200 VMs across different business units). Last time I did something like this was 2021, and I'm sure a lot has shifted.

Currently looking at:

• Graviton3 instances vs traditional x86 for cost optimization
• Database migration strategies (mix of Oracle, PostgreSQL, MySQL)
• Storage tier decisions with S3 Intelligent-Tiering
• Whether Savings Plans are still the move or if on-demand flexibility makes more sense now

Also curious about networking costs. Are folks still seeing surprising egress charges, or has AWS gotten more predictable? And anyone actively using AWS DataSync vs just rolling with DMS?

Would love to hear what's working for you guys in 2026. Any gotchas I should watch out for?

Graviton3 is solid for cost savings if your workloads aren't tied to specific x86 libs, but honestly test it first—we found ~15% cheaper per instance but spent weeks chasing ARM compatibility issues with some legacy Oracle stuff. Ended up hybrid.

For Savings Plans vs on-demand: with 200 VMs you're looking at major commitment. We locked 1-year RIs on the stable workloads and kept on-demand for the ones still being optimized/migrated. Gives you flexibility without leaving money on the table.

One thing that's changed since 2021: AWS DataSync is way better now for large DB migrations. DMS still has its quirks. Check the docs if you haven't lately—https://docs.aws.amazon.com/

Yeah good point on the ARM compatibility—that's actually what I was worried about. Sounds like I should do a proper POC with Graviton3 first before committing the whole fleet. Did you end up sticking with it after sorting those issues, or did you revert to x86 for some workloads?