AMD instances seem to be very scarse

It seems the m7a.2x instance type is not available in sufficient quantities in the AWS region you are using. AWS Batch needs a minimum number of these instances to be available at launch in order to start the compute environment.

In order to boost AWS Batch performance, we can take a few steps. First, we should go for a versatile instance type like m7i.2x that supports both AMD and Intel options, making the most of the hardware available. Secondly, if the m7a.2x instances are more accessible in another AWS region, we might want to consider launching the compute environment there for better availability. Thirdly, we can cut down on the initial number of instances by lowering the DesiredvCPUs setting, as long as it meets the minimum requirement. And lastly, it's crucial to keep an eye on the AWS documentation regularly to stay updated on instance type availability and quotas in our region, as hardware availability can change over time.

Surely I can use (and in fact I am using) m7i.2x but m7i.2x is't more versatile and do not run on AMD.
In fact you may look at https://aws.amazon.com/ec2/instance-types/ that m7i are based on intel CPU while m7a are based on AMD Epyc CPU. Some time ago I had been able to launch m7a in my compute environment and I had verified that in my applcation m7a are faster than m7i.