Comment by arcfour

An EC2 instance with lots of cores like a c6i.32xlarge should do the trick, no? You could even pay for spot instances and just checkpoint frequently and copy the progress file to S3 when you get the interruption warning.