Maybe limited compute to handle launch traffic spikes?

Still get the model out there.