What do you mean why is nobody doing that? Lots of providers are doing that. Almost the only providers in the same price league as AWS are Google Cloud, Azure and Rackspace. All of them appear to assume (rightly so, judging by peoples reaction) that they can keep their margins high on the basis of brand name recognition and specialist features, and none of them seem interested in rocking the boat.
They will drop their publicly listed prices once smaller customers gets comfortable enough with cloud services to start seriously shopping around, not before. Larger customers will already negotiate (e.g. I know of one AWS customer with ~75% discount over the listed prices)
> The only place I have a gripe on AWS pricing is bandwidth. I keep waiting for the announcement that bandwidth is being cut in half next month and it hasn't happened.
AWS bandwidth pricing is 5x-50x higher than the providers I usually use. E.g. for the cost of 1TB of transfer at AWS you can rent a server for a month + multiple TB of transfer elsewhere. Cutting it in half would still leave them ridiculously expensive for bandwidth.
> I've done the datacenter thing many times and there are so many pieces to the costs of running your own that most people don't include. The biggest one is just plain agility. Predictable workloads can benefit from self-hosted, but then again you can get much better prices at AWS when your loads are predictable as well if you're willing to commit.
I do the datacenter thing every day, and I also do AWS and other cloud providers every day, and when I move customers to dedicated hosting they typically cut their total infrastructure costs to 1/2 or 1/3 vs AWS costs after considering reserved instances and maintenance costs. For many of them, if they were large enough to get serious attention from Amazon they might have been able to negotiate deals that'd make it cost effective to stay, but I've also had more extreme examples, like a customer that cut their hosting cost to 1/10th.
The agility argument doesn't fly with me - I manage multiple systems that are hybrid mixes of dedicated predictable loads on self-owned or leased hardware, coupled with cloud resources that are "tied in" via VPNs as needed.
With well designed systems, the very availability of cloud makes dedicated hosting cheaper even if you never or rarely use it: Where we before often had to plan for 50% to 100% overcapacity on dedicated hardware to handle spikes (and might still have to just accept that some spikes would slow things down), we can now plan on almost no overcapacity and spin up on-demand instances to take spikes (we need only sufficient overcapacity to react and for additional capacity to come online), or even to shave off the top peaks of daily traffic.