EC2 VcpuLimitExceeded when launching another VM: the quota counts vCPUs across all running On-Demand standard instances

Error: VcpuLimitExceeded: You have requested more vCPU capacity than your current vCPU limit of 256 allows for the instance bucket that the specified instance type belongs to.

The On-Demand quota is per *bucket*, not per instance type. 'Running On-Demand Standard (A, C, D, H, I, M, R, T, Z) instances' covers m6i, m7a, r7a, c7i, t3 and so on, all together. It counts the vCPUs of every running instance in the region, including long-lived dev boxes and small t3s you forgot. Instances in shutting-down may still count for a short while.

Quick check before launching:

aws ec2 describe-instances --filters Name=instance-state-name,Values=running,pending \
  --query 'Reservations[].Instances[].[InstanceType,CpuOptions.CoreCount,CpuOptions.ThreadsPerCore]' --output text \
  | awk '{s+=$2*$3} END {print s" vCPU in use"}'
aws service-quotas get-service-quota --service-code ec2 --quota-code L-1216C47A   # standard bucket limit

Options: pick a smaller size in the same family (memory-bound jobs: r7a.4xlarge has 128 GB on 16 vCPU), terminate idle instances first, or request a quota increase (Service Quotas console). Increases for standard buckets are often approved in minutes to hours, but don't count on that mid-task. GPU, high-memory (X) and Spot have separate buckets.