Long-term aim is to be able to handle 10 billion users, with 1000s or millions of ops/s and sub 100ms response time. Will take a few years to get there, but an architecture can be designed.
Discussion
I think regular web can already handle partially of that requests (1 million req/s)
https://www.techempower.com/benchmarks/#section=data-r22&test=json&hw=ph
It just we need more load balancer server with horizontal and vertical scaling applied. The most bottleneck part is probably verifying schnorr signatures multiple times for all events incoming.
Relay developers have such a big homework to achieve that. They also need to optimize many things.