I wish we could observe a parallel universe where Netflix went a different route and stayed with dedicated or went with colocation. I'd love to compare the outcome with their decision to go with AWS (minus CDN).
They came from a datacenter world before they moved to AWS. The problem was that they had a traditional enterprise IT organization with big process and relying on Oracle, SANs, and all the same old enterprise IT solutions. It was built to run a company that was focused on operating warehouses and logistics, and not an agile web company.
Moving to the cloud was as much an organizational transformation as it was a technical one. And the move to AWS has paid them dividends in organizational and technical agility, as well as execution speed. I will never believe that it saved them money when compared to a well run physical infrastructure that was more purpose built for what they do, but they did not demonstrate the ability to run a physical infrastructure well so it's kind of moot.
It worked out great for them and gave Adrian some great stories to share. Congrats to him on the move to Battery, that's a fantastic team.
Not only Netflix they get some better price (I have no info but it will not be surprising) but I have even heard that Netflix do work with the AWS team to test first some new features before AWS release it to a larger public.
I have even heard that a couple AWS service were in fact created because Netflix was asking for it.
Netflix being probably the most visible company on AWS, I am sure they do have some special deal with Amazon.
Anyway concerning the news, it will be interesting to see Adrian in his new role! He is a great guy!
One thing to note about a lot of "infrastructure" businesses is it can make business sense for them to offer deep discounts to large customers to cover their overhead - they then make their profit by selling whatever excess capacity is left over to smaller customers.
I.e. imagine you want to setup a new courier business. You convince Amazon to move all their shipping to your new courier company, by giving them a price which is almost unprofitable for you. You then make your profit on the excess capacity you have, by selling it to individuals who need shipping at more profitable prices.
Actually, many many companies (including a startup I once worked for) had negotiated rates w/ Amazon. Netflix is certainly paying far less than sticker price.
Not to mention, Netflix is a flagship customer for AWS and directly results in AWS getting more business, so I'd expect the discounts to be even deeper for Netflix.
I remember reading something about how AWS is pretty much running very close to at cost (super razor-thin margins, just like the Amazon shops themselves),but seeing these numbers seems to imply otherwise (20% is a big chunk of change).
Isn't the opposite true? It's fine for under 10 instances when you're first starting up and still flexible, but over that you're better off going colo?