Introduction

I wanted to document some very good use cases for IaaS public cloud, in part to illustrate that the number of people needing such use cases are few and far between.

Massive elasticity

A combination of short over all time frames with massive elasticity really seems to (on paper, haven’t run any actual numbers) make a lot of sense for IaaS public cloud. Take an example of perhaps a steady state usage of a few dozen CPU cores, but then needing to spike to say, five thousand CPU cores for at most sixty hours per month, or about two hours per day. Data processing or something, of course another aspect of things is storage, how much storage is required and where is it stored. But assume for a moment the data set is small, and for whatever reason you need a lot of CPU cores+memory for a short amount of time each day. Building out physical infrastructure to run such a workload doesn’t seem like it would make sense to me, so I suspect IaaS is a good solution for this kind of workload. Whether you are using these CPU cores for 60 hours one time a month, or two hours per day, every day. The more hours you use the system the more likely the scales tip in favor of hosting it yourself.

Huge scalability testing

I can’t find the article now, but I did come across at least one article on El Reg I want to say in the past year, talking about an organization doing some kind of HPC test in a public cloud. They spun up something like 2,000 servers to run some benchmark, got the results and shut the systems down. This again seems like an ideal workload for IaaS. Maybe you end of spending $250,000 to run such a test over the span of a few days or week or something. But you don’t need to do this often, maybe once every year or two years or something. Buying a 1-2000 servers(at a cost of several million) to run such tests doesn’t make sense (unless you can find lots of other things for those servers to do in between tests, but that’s not the point here..).

Unusual hardware requirements on a short time frame

Say you got an idea, and out of nowhere you really require some data center class GPUs in order to do something, maybe something with “AI”, or other computational thing, and you don’t have any data center GPUs(and for whatever reason, desktop class GPUs won’t cut it). Maybe you don’t even know if your idea will pan out at all you just want to test stuff. Maybe your server vendor has a big backlog on GPU servers and you can’t actually buy any for two months or more. Lots of factors here. But IaaS public cloud is a good use case for this as well I believe, short term testing with unusual hardware. GPUs are becoming less unusual now but really I think the vast majority of server deployments in 2025 still don’t have any GPUs(obviously excluding the big “GPU farms”, I’m talking general purpose compute). There could be other unusual requirements such as super fast networking, or super high CPU counts on a server or memory or something. Obviously it likely makes a lot of sense to go “on prem” again with this workload if you figure something out that actually works and you plan to use for a while.

Ramping up multi region quickly

Had an email exchange with someone who suggested cloud is good for ramping up geo diverse workloads quickly, something I sort of address on the home page. But wanted to call it out here too. After thinking about it I think it very much falls into the same general category as the above “Unusual hardware requirements on a short time frame”, specifically deploying stuff short term(location is irrelevant since you are using cloud) in a super short time frame. IaaS is good for that for sure, probably nothing beats it. But realistically most organizations aren’t that short sighted to need to do something that fast (hours, days). Serious orgs would have planning regardless and that often takes weeks to months to even determine where you want to be hosted exactly and how etc. But as a stop gap measure using IaaS to bridge that time frame between now and when you have stuff ready for on prem in that general region, IaaS is a fine option. As mentioned on the home page as well, cloud companies themselves often leverage co-location facilities in similar markets in order to spin things up much faster than they otherwise would take if they were to go and build or buy a data center in that region.

Very small setups

This is somewhat less relevant to the purpose of this whole blog site, but I wanted to point it out anyway. If your IaaS bill is a few grand a month, depending on your company size, that’s not a big deal. The primary target for the topic of this site is really organizations spending at least $50-100,000/mo in IaaS. Which is an absolutely shocking number of organizations. Even more shocking are those spending many millions per month on IaaS. If you are super small and are happy that is fine, you can still probably run stuff better hosting yourself but the scale of the savings isn’t the same of course.

I’d say the number of customers that fall into this bucket far exceed the other three buckets above. But the number of customers that fall into other use cases(which IMO are not IaaS optimized) dwarf everything by comparison.

Also note of course there is an entire industry built up around “VPS”, or Virtual Private Servers, that serves this low cost market, as well as dedicated server hosting, both of which have tons of service providers in these markets. In the latter model, it’s sort of like co-location except you don’t have to supply the hardware. There are even managed dedicated server offerings from several providers, Rackspace initially popularized this approach probably around two decades ago.

Other use cases

Know of another good use case? I’d be curious to learn more, drop me a line.