> It wasn't me that came up with the backup example - which I do think is the best example of the point they were making because you wouldn't expect that partic

> It wasn't me that came up with the backup example - which I do think is the best example of the point they were making because you wouldn't expect that particular service to fail nor someone to get fired. And if it does, it's more likely yourselves to blame for setting it up wrong than it is the cloud vendor at fault. Backups are notorious for going untested because too often nobody cares if they work until you discover the hard way that they don't. That's the "setting it up wrong" problem and cloud doesn't really change that. Cloud vendors then add a different problem, which is that they operate at a scale that brings multiple redundancy failures into the realm of plausibility. Then, because the same infrastructure is shared by many customers, you end up with greater exposure to those kinds of previously negligible probability multiple systems failures when something happens that affects thousands of customers at once. Which is still not very common, but it can be very bad when it does. > Their general point was more that if you self-host and the infra goes down due to hardware failure (eg switch goes pop) then it's up to you to fix. With cloud services that responsibility now becomes someone else's problem. There are two components to this. One is whether it's "your problem" in a CYA sense, which I understood to be the original argument, i.e. if the system fails and your company loses ten million dollars then you lose your job. But that hasn't really changed. If the backups don't work, pointing at the vendor is not very effective cover when you're the one who chose them. The point you're making is that when something fails you don't have to spend time fixing it because that's somebody else's job now. Which is fine, but then you're paying a premium for that and the question becomes whether such hardware failures are common enough that you come out ahead that way or not. > You pay a premium for that (interest and insurance) and are still stuck with the repayments even if your project crash and burns before the instalment plan finishes. You pay even more of a premium for third party hosting, and server hardware is a general purpose commodity. If your project ends then it can be used for the next one. And even something bought on credit can be resold and the money used to pay off most/all of the debt, especially now that Moore's Law is in decline and hardware doesn't depreciate as fast as it used to. Moreover, if you know ahead of time that your project has a high short-term failure probability then it can make sense to not want a longer-term commitment, but not everything is so uncertain. Many projects are known to have a >95% probability of still existing in five years. > If you're a business and looking to buy hardware you are much better off buying it outright - you have a stronger position to negotiate a discount (so you're not paying advertised price) and you have capital you can resell if the worst happens. That's true. But what if for some reason you prefer "effectively paying for your hardware on a subscription basis rather than paying tens or hundreds of thousands of pounds / dollars up front"? :) > If you're just looking for a VMWare equivalent then AWS et al isn't going to be that impressive. Where public clouds have overtaken self-hosted solutions is with their SaaS solutions. This is why in my other post I make the distinction between "lift and shift" type deployments on the cloud, and using their SaaS offerings. I don't think we're really disagreeing all that much. There are circumstances where cloud hosting makes sense, particularly when your needs vary unpredictably over time. My point is that people overuse it. You see companies with entirely static and predictable future needs moving their entire operations into the cloud anyway. And I suspect people are going to regret the long-term experience of SaaS. Once you have a decade of your business data on someone else's proprietary service with no export function, what does your negotiating position look like?