Well, I've worked on two now... once was self-hosted by a large services company, which has its' own data centers and redundant connections, it also handles a significant portion of DNS for the internet (for good or bad). It worked for them, but was painful.. Getting new servers up was months of work.

I'm working at another service company expecting to clear 5M requests/day with bursts close to what you're talking about several times a year. We've had so much pain managing colocated servers, we can't justify the cost of 3 full data centers for just our load, that doesn't make sense. We're currently moving to a cloud provider to be able to scale out better when peaks really spike.

We had a customer that wanted to do several million requests in a 15 minute window, and can't currently handle that... We're restructuring/refactoring so that we can.

It may not be sustained, but getting 330k requests/second in bursts is a different way to think about problems than anything less than 1k/second, which many servers can hit without a sweat, and why I'd be more inclined to push for mid-level VPS like DO or Linode in those cases. Depends on need and expected growth.