How can I get a g6e.12xlarge instance when every European AZ reports insufficient capacity?

GPU capacity is heavily constrained across several cloud providers right now, not just in Europe. There may simply be no spare inventory for this instance type at the moment, so the realistic choices are waiting, broadening the acceptable instance families, or arranging capacity ahead of time.

For a recurring requirement, ask about an On-Demand Capacity Reservation in one of the permitted European AZs. It won’t solve the initial shortage because the reservation must be created when capacity is available, but afterward it can prevent having to repeatedly compete for capacity. Reservations can be expensive if the instances aren’t used consistently, so compare the cost with a Savings Plan or another commitment option.

If you’re requesting multiple instances at once, reduce the batch size and try smaller requests. A request for ten can fail if only nine are available, while a request for one may succeed. If even a single instance fails, there probably isn’t any usable capacity in that location right now. AWS Support or your account team may also be able to advise on expected availability if you provide the exact regions, AZ flexibility, quantity, and duration.

This is a common, usually temporary problem with highly demanded GPU instance families. If your workload allows it, support a broader set of instance types and Availability Zones rather than depending on one exact SKU. G4dn, older G5/G4 options, or newer compatible GPU families may be worth testing.

An Auto Scaling group can help by including every acceptable instance type and AZ, then retrying periodically as capacity appears. Just make sure you understand its health checks and replacement behavior so it doesn’t terminate instances unexpectedly.