We used to have an internal (and external) cloudfoundry instance. That was pretty nice as far as on-prem deployment options.

It’s just a shame they were permanently out of database servers with SSD storage, and for some reason couldn’t provision more for over a year.

Yes, this is the awkward phase of on-prem. Some companies stay there forever. Good companies will continue innovating and treat the time to resolution of your request as a KPI to reduce down to days, minutes, or even seconds.

On-prem and AWS/GCS/Azure aren't the only options.

There are smaller cloud providers, rented VMs, rented dedicated servers and rented colocation space.

I believe renting dedicated servers is often overlooked. You pay someone else to install hardware, ensure network connections and be on-site for hardware swaps etc but still have the maximum degree of flexibility.

Even larger companies can work well with that model, traffic also tends to be cheap enough that you can spread across different vendors to avoid lock-in. And in that case, your sysadmins can sit wherever they want, no need to be physically close to the servers.

Also, as there's much less knowledge to be a dedicated server provider, competition is strong and prices are comparably low.

> I think the case is for big companies that have a hard time attracting IT talent.

Companies that have made a name for themselves by outsourcing to the cheapest IT contractor that will promise them the moon and fill the seats with barely warm bodies? I was one of those bodies so I know exactly why they can't attract talent - they don't bother, and don't reward it. They treat IT as a cost center and are surprised when they get disrupted. The only good options in those companies are to work on the business side or worst case as a project/product/program manager interfacing with the warm fungible contractor bodies.

Many Enterprises are only alive because of inertia and goodwill from earlier decades.

Once you scale you'll need DBAs anyway to do things AWS won't do for you,or developers with the same skills, like figuring out why your developers are writing queries that kills the production database because they didn't test with data sufficiently similar etc.. I used to manage about 100 Postgres instances alongside ca ~1000 VMs total spread across colocated servers and managed hosting in several countries. The time I spent spent on the type of DBA tasks that e.g. AWS RDS automates away from you was measured in minutes per month after I'd spent a few days automating backups and log shipping and failover.

I kept being asked to price out a migration to AWS, and we kept coming up with 2x-3x the cost. Part of the reason was that we could pick and choose servers that fit our workload in a way we couldn't with AWS, and partly the absolutely insane bandwidth prices AWS offered.

I use AWS. I like AWS for the convenience. But it's a luxury that is ok when you're either small or really high margin, and you're paying massively over the odds for that luxury.

The reason these services get away with being so expensive is that people massively overestimate the complexity and don't bother actually getting quotes from people or companies to manage these services for them. When I was doing consulting my biggest challenge in offering up alternatives to AWS was that people were so convinced AWS was cheap that even when presenting them with hard data they often didn't believe it. For me it was a mixed bag - I tended to make more money off the clients who stayed on AWS as they usually needed more help to keep an AWS setup running than those I migrated to managed hosting setups, despite paying more for the hosting too.

On prem rarely makes sense any more other than at that kind of huge scale or for privacy, sure. But that's because dedicated hosting operates on really thin margins and has become really cheap. You have to get to massive scale before cloud providers will give you big enough discount to start approaching the kind of costs you can get that way with a decently engineered system. Not least because cloud providers themselves provides a weapon: Set up your system so it can scale up using a cloud provider to handle traffic spikes and you can load those dedicated servers much more heavily than you could otherwise risk.

The biggest issue, though, is how few people are aware they can negotiate with their cloud provider. I've seen so many places just pay the sticker price without even trying to get discounts.

(Conversely, I once got a contract to do zero-downtime migrations first from AWS to Google Cloud and then to Hetzner so a startup could launch on AWS and spend the huge amount of free credits they'd been given there, then migrate to Google Cloud to do the same, and then finally move to Hetzner once they had to actually start paying; relative to what they'd have to start paying if they'd stayed on either AWS or Google after their credits ran out the cost of having me do the extra setup to handle that was covered with ~2-3 months of their savings)