Trust but verify.
The use case you’re mentioning require serving static pages; publish once serve many. Strictly speaking requires parsing the request and streaming back the results from caches in ram. You might also need to serve some static visual page elements and some javascript for the browser to put together a page.
If the are small (<1MB), I’d use 10k qps per core as a rule of thumb; that would be normal for any reasonable c/c++/go perhaps even Java webserver perhaps these days if you tune the jvm a little bit.
I’d be more worried about your network setup (who provides it and at what cost).
Also I’d be worried if web developers would do something stupid like design those stupid pagination systems that return 25 results per page or do sql database queries in the critical path of those 250kQPS. If you need keyvalue serving stick the dataset into a memcached instance on server startup, count the number memcached requests per page. As a rule of thumb, if it’s more than 10 you should reorganize.
As for cloud, I’d say do not use, they’ll charge you per load balancing request and nickle and dime you to death on bandwidth; (at 250kQPS you’ll run out of money very quickly). Go to (or contact, don’t go anywhere) your local university IT dept staff that take care of the server infrastructure. Talk to them, they should be able to hook you up with some VMs or with access to a kubernetes cluster you can use. They can also recommend what makes sense to do to avoid a thundering herd effect.
Lastly, if you’re desparate and have a gig of data in something like csv files, and want to distribute it among a bunch of people, students and such, and are alone with zero budget. Pack it into a torrent file, announce the torrent on a public tracker and ask for help in reddit. …(if the other people you’re working with have only little experience with these 250kQPS web things, prepare a plan b, get ready to go this anyway)
On the other hand if this is all a theoretical exercise, lookup site reliability engineering books from Google, read up on NALSD interviews Google does. SREs at places like Google Microsoft Facebook Amazon Netflix Apple and other places that run cloud services and/or clouds ; like to interview people for SRE roles out of smaller shops like e.g. your local university it dept that has a few dozen apps, maybe 100k/200k users and sometimes does high qps stuff like you’re asking. Google in particular does this relatively methodically, which makes sense considering how large of a fraction of the world’s compute and storage and network capacity those people end up being responsible if hired.