[koffiezet]

Let's be clear here, everything you can do in a "cloudy" environment, you could do on big servers yourself - but at what engineering and human resource cost? Because that's something many - if not most - hardware and 'on-prem' infra focussed people seem to miss. While cloud might seem expensive, most of the times, humans will be even more expensive (unless you're in very niche markets like HPC)

[dividedbyzero]

You could also have those big servers in the cloud (I think this is what many are doing; I certainly have). That gives you a lot of the cloud services e.g. for monitoring, but you get to not have to scale horizontally or rebuild for serverless just yet. Works great for Kubernetes workloads, too – have a single super beefy node (i.e. single-node node pool) and target just your resource-heavy workload onto that node.

As far as costs are concerned, however, I've found that for medium+ sized orgs, cloud doesn't actually save money in the HR department, the HR spend just shifts to devops people, who tend to be expensive and you can't really leave those roles empty since then you'll likely get an ungovernable mess of unsecured resources that waste a huge ton of money and may expose you to GDPR fines and all sorts of nasty breaches.

If done right, you get a ton of execution speed. Engineers have a lot of flexibility in terms of the services they use (which they'd otherwise have to buy through processes that tend to be long and tedious), scale as needed when needed, shift work to the cloud provider, while the devops/governance/security people have some pretty neat tools to make sure all that's done in a safe and compliant manner. That tends to be worth it many times over for a lot of orgs, if done effectively with that aim, though it may not do much for companies with relatively stagnant or very simple products. If you want to reduce HR costs, cloud is probably not going to help much.