If you run your 24/7 steady state load using on-demand hourly charged instances, and don't factor in the cost of managing your own hardware and data centres then yes, EC2 is hideously expensive.
If you plan your spend (much like you would in buying your own hardware and DC space) it really isn't - reserved instances considerably reduce prices, and come with bulk discounts when you buy enough of them.
It's been a while, but when costing out our migration to EC2 it came out cheaper than any other VM provider, and was in the same ballpark as leasing our own hardware.
Cloud providers bring you not just localized infrastructure, but distributed infrastructure, with clear mechanisms for scaling as needed, plus support, etc. With AWS, you have a predictable cost, predictable up-time, and ability to scale at peak, and your costs are spread over time, rather than all up front.
With your own servers you have to build out to handle peak load (and run mostly dormant during non-peak hours), across multiple locations, to achieve the same availability.
I'm not sure the costs for the convenience of AWS are so obscene in such a case. They may still be higher, I honestly don't know, but the convenience benefit is huge. They key bit is that AWS' costs are largely -known-, whereas the costs of DIYing aren't.
Per Netflix's blog post about why they use AWS for serving everything except video content - "We could have chosen to build out new data centers, build our own redundancy and failover, data synchronization systems, etc. Or, we could opt to write a check to someone else to do that instead."
Oftentimes for enterprise, having all your infrastructure in place, done right the first time, with someone else eating any costs associated with the unexpected, and ready to start building your application on, is worth the cost. For enterprise, paying a known amount to avoid risk is oftentimes worth it.
"If you’re used to designing and deploying applications in your own data centers, you need to be prepared to unlearn a lot of what you know."
Which is it? Did it save them money because they didn't have to "build their own redundancy and failover" or did they have to build a bunch of custom tools like Chaos Monkey because "I knew to expect higher rates of individual instance failure in AWS, but I hadn’t thought through some of these sorts of implications." ???
From reading Netflix' blog, it's clear that it was a huge learning and engineering effort. That cost has to be taken into account and added to the fact that across many AWS offerings, you're looking at as much as a 10x price / performance penalty versus dedicated or collocation. I'm unconvinced that they couldn't have done it cheaper and better through more transitional approaches, and moreso that it's a meaningful indicator for anyone else.
AWS has been massively innovative. But it's much more expensive, and a huge part of their business is sales and getting CXOs onboard, not necessarily providing good value.
This.
The pricing is especially egrerious in bandwidth charges which for AWS are almost pure profit.
Amazon is only cheap if you don't use it.
We've run the numbers, and we can't host it ourselves anywhere close to what we pay Amazon if we want the same amount of redundancy that AWS gives us "for free" -- we'd have to duplicate most of our infrastructure across 4 separate datacenters to get the kind of multi-AZ, Multi-region redundancy we have now.
Eh, I think the bar is a lot higher than that, especially if you have to meet external compliance standards with yearly audits, or have to have good disaster recovery with multiple datacenters. Those kind of things get very expensive if you are under a certain scale.
Because they are corrupt? Could be? Large companies are basically centrally controlled economies, and we know how that turns out..
It seems to me that large companies live in a special bubble where they are willing to pay extra for everything. On the other hand, they also charge extra, so externally, it will mostly cancel out.
The winners of this scheme are probably the people involved in running (either directly or as shareholders) these large companies, because their salaries (and other bonuses) are artificially inflated compared to the rest of the society. The losers are probably the government and the taxpayer (i.e. the public), these can be easily made to pay extra but it's difficult for them to charge extra.