I did not like that I am bound to their kernel on bare metal. There are setups that require kernel modules they don't provide, like DRBD. I got the modules to compile and work in the end, but it was not fun and will probably break on some future update. This is not how I expected it to be on "bare metal".
Also that prevents you from using the Xen hypervisor, which again I would not expect from "bare metal".
You only get a single IPv4-Adress per system, which not only is a burden to virtualization, but also larger container-setups.
So I found little value in the "bare metal"-part, but it's still great value if you compare it to other VM offerings.
If you treat it like a VM (or use one of their VMs), you need to be aware that disks are not redundant and their content (or at least the changes since the last boot) are gone on a crash or hardware failure. You need to mirror or replicate it yourself, or design around it on a higher level of the stack. There is no option to attach a persistent disk that would survive a hardware failure and could be attached to a replacement host/VM.
So in the end scaleway is a weird mix for me: for rented VMs I expect the platform/provider to take care of some basic reliability and redundancy stuff. For bare metal I expect to be able to setup all of that myself without trouble.