[Clubber]

I'm sure people will disagree, but I've always viewed the main benefit of cloud is a company with no talent or time to do networking / infrastructure early on, so they use the cloud, which requires less expertise. Once they grow and can afford network / infrastructure talent, the switching costs are hard to justify, even with the 3x markup (according to the article).

I can see why it's compelling, but for me running my stuff on other people's servers in this day and age is concerning. Like many computing things, it really depends on the situation.

[0xB31B1B]

The issue is that the 3x cost doesn’t include the cost to staff the team that runs your in-house platform software on your colo setup, and the mixed incentives you get when a sizable part of your team is not shipping product but instead shipping say, Kafka queues. My previous employer has one of the largest colo deployments of the f500 companies and it was a mess technically and organizationally. At the end a few hundred infra devs were laid off and projects scrapped after ~24 months of no visible progress on platform stability and support.

[kaetemi]

Rented dedicated feels like a nice middle ground between colo or on-site and cloud hosted. It's on the cheaper end. And if you can boot into a self-configuring image, maintenance cost is quite low too.

[CodeAndCuffs]

Part of the issue I see overlooked is the cost of acquiring and losing physical assets. If you have an onsite data center and a meteor hits it, how bad off are you?

Disclaimer, I barely passed my aws associates cert and have 0 qualifications to weigh in on this subject with any authority. It's just a point I've seen glossed over before. Yeah your data center is cheaper to run, but is it cheaper to replace?

[greatgib]

Any decent IT team will backup, replicate, shard, to different locations, and not rely on a single datacenter.

As said in the blog post, with the cloud you will get the same thing that you can pay your inhouse IT for 3 times the price.

For example, lots of people are thinking that by just using 'the cloud', your data is safe / replicated to multiple geographically separated areas. But no, it is not automatic, it needs to be configured like that with the associated costs. If you lose an ec2 instance, it will be lost in the same way as a node in your personal data center.

[lumost]

How likely is it that you have a competent IT team? If they are competent, are they also fast? I may not care about having a scalable, redundant deployment if it takes 3 years to get done.

It’s tough to recruit a tech team and keep it running. Worst case scenario the team leaves and your stuck with folks who don’t know how anything works.

[flashgordon]

+1. On top of this if you need to provide slos on availability, diaster recovery, respect regionalized requirements, even with a technical playbook just acquiring physical space gets hard and expensive real quick!

[Clubber]

That's a good point. Redundancy requires more cost and more expertise, but as a counter argument, AWS has had its share of outages lately as well.