The one thing that I've seen that cloud services such as AWS offer over places that offer dedicated servers or co-locating your own servers is the type of administrator you need.

Administrating a server even a dedicated server with terabytes of storage attached to it and many hundred gigabytes of RAM to either correctly run microservices or containers or VMS efficiently and redundantly is a very specialized skill set. Well there's a lot of people who call themselves and admins who can do that, there's really not that many that can do it well. And the moment they encounter problems they lack the skill set to understand why so there now embedded with service contracts to vendors who are giving them the advice on how to run it properly.

Scaling up ec2 instances and block storage and all the other cloudy widgets you completely eliminate that specialized skill set. You need people who understand something about operating systems you need people who understand the application but you don't need people who understand in detail how your OS is going to interact with your hypervisor in the metal that it's running on along with how your data storage goes down from application to spinning rust. The cloud service takes care of that specialty for you and all you do is say am I getting the correct performance metrics or not and when you're not you don't care why you just restart your instance or open a ticket with the cloud provider.

Several of my real, profitable business customers are running $200-$300/mo AWS bills. That includes multiple virtual servers, hosted RDBMS, load balancers, multi-region redundancy, etc.

If AWS is costing thousands a month the company is either big enough to need that level of infrastructure, or doesn’t know how to manage their cloud resources. As a freelance system admin I see over-provisioned services and extras added on all the time, doubling or tripling AWS bills for no reason other than ignorance and neglect. I see software stacks that are such pigs they need huge servers to put a simple web page together.

Companies should focus on their core products, and that’s even more true for startups that have less cash. Some of them I’ve worked at are using Docker and k8s and all kinds of random crap they don’t need, presumably chosen because it’s cool or some prima donna engineer had to get their way.

The reason to use services like AWS is that you can scale in the opposite direction too - provisioning cheap infra for $10s - $100s a month. That's something your hypothetical(?) startup should consider doing, depending of course on what they are doing with all that hardware.

Somewhere there's a break point between a few hundred a month on AWS, and a few tens of thousand a month.

If my startup needs, say, a pair of redundant app servers, a redundant (MultiAZ) database, a load balancer, some object storage, and the "wrappings" around that (like ssl certs, dns, vpcs for security/isolation, monitoring and alerting and dashboards, security groups for access control, decent access control and role management for platform/admin users) - that's less than $200 a month and will see better than 99.99% (assuming your app doesn't drag that down). That's a _total_ no brainer. All that stuff in the list in brackets there is free (or so cheap as to not matter). There's no way a startup with less than maybe 25 or even 100 engineers should consider standing all that up and supporting it themselves. Sure, you could do it cheaper, but the _most_ you could save is $200/month, and there's tens off thousands of engineers who already know how to do all that either from the AWS web console or via AWS APIs.

By the time you've hit $30k/month, the question is much less clear. $360k a year is a _lot_ of money, but even if you could save 90% of your hardware/hosting cost it's not enough to pay 2 senior engineers, let alone a team big enough to have 24x7x365 on-call. If you have the right team, you can probably keep a bare metal dedicated server platform running and secure and approaching 99.9% uptime, especially if you have someone (or hire someone) specifically with the skill to do that. But you still need to decide if it's worthwhile for you to have that skilled person wrangling servers instead of writing code that adds features that get you closer to product/market fit.

You're 100% right that if you've only got $200k runway, spending 30k of it a month making Jeff rich is a bad plan (except for Jeff). But if you've built a thing that actually _needs_ $30k of AWS resources a month to run but isn't bringing in enough money to help extend it's runway, your business plan is bad.

Time is one of the most important resources for any startup. Also, startups need to be very flexible since they might need to re-evaluate and move in a different direction. Long term AWS will save lots of time. You really don't know what you will need. AWS has lots of services to choose from. If you can afford it and have the expertise, start with AWS rather than individual servers. At some point you might find that you can save money by setting up your own servers but that will come after you have an understanding of your business and needs.

I think you have to chose the right tool for the right job, and you have to also understand that not all people have the same toolsets.

For most startups, they are hoping to get some money (investment, revenue, etc) in order to keep themselves going in the first bits. Buying servers is a huge capital expense, and not all startups have excellent credit and ability to do long term financing. But almost every startup will have access to a credit card and be able to pay a (reasonable) cloud bill on a monthly basis.

I've worked with a business that wanted to put TBs of data (with monthly growth) into AWS. I think the break-even point to buying a fully automated 10 tape library was something like 14 months, by which time they would have fully paid for ALL the tapes, the changer, and an extra drive to do emergency recovery offsite.

I think with startups it's about controlling costs -- developers are expensive but needed, sys-admin skills are expensive but "easily avoided" by just using AWS. Some of it is also the ease factor, where you can use a bunch of cloud services to do things that are powerful, at fractions of a penny per transaction.

I also think some biz people aren't always tech experts, and tech people tend to either want to try something shiny (to pad their resume) or stick with something they know well. Cloud and AWS particularly make that easy.

Wouldn’t your early hires’ time (and equity) be better spent doing the things that make your company unique rather than menial rack-and-stack, cabling, having to drive to the DC when a disk dies, etc?

I suppose it could be because of the ecosystem, ease of finding developers who are familiar with said ecosystem, and, as you mentioned, the reliability.

> Dedicated servers can come with terabytes of data, 60+ GB ram, and 16 cores or more while costing much much less.

What vendors do you recommend?

No. If your AWS costs matter, either your business is working, or you suck at engineering.