What are the overages??! Why is it not just sliding back to pay-as-you-go pricing, like reservations for say EC2 work.

I'd guess because they'd like to accurately model their compute requirements to try to get as low of a price as possible (e.g. via reserved EC2 instances).

If clients constantly and consistently shoot past what Algolia plans for, it might cause issues or at the very least cause the company to spend more on, on-demand instances than they intended to.