I remember observing more than twenty years ago, "The answer to any interesting problem in computer science is 'it depends'." Or, as the Right Pondians put it, "horses for courses".
Our entire IT org is less than 15 people. Most are the programmers that maintain the applications that our entire company runs on. It's <10qps. Our entire cloud spend last month was about $5k. For that, we get all the love that AWS & Heroku can give a small company.
We are getting a GREAT deal.
You can save a lot of money if you don't carry insurance. You'll probably lose a lot of business as well, because clients understand the importance of limiting risk. A substantial value of the cloud is in mitigating risks--which is why I'm so strongly opposed to using Azure.
I learned SRE at Google. I know what it takes to deliver 5 nines at scale. If your business is going to take a reputational hit if your miss 4 nines, you better understand as well before taking it on yourself. What would a black swan event do on that front? Can you handle it?
If you're not in three different data centers in three different parts of the US, with onsite staff, you're not 5 nines. (I pity those in a small country with PII or PHI.) That requires tripling your hardware AND maintaining failovers, by the way. Or you could be in four DCs in four different parts of the US and only double the hardware. With staff at each location.
Of course, economics change. It should be expected that AWS & the like want to go the route of Oracle. (And by the way, just how IS Oracle's cloud business doing?) A rapidly growing company should probably evaluate it's on-prem / in cloud position every year, a stable one every two.