What kind of infrastructure do you use for that scale? Are these computationally light ARIMA models?

We're running 150k models per day and while I'm sure I could scale this to running them every 30 minutes, it would mostly be me throwing money at the problem to parallelize the compute.