In response to my Azure reaction, James Bender pointed out that Azure, in its current form, really wouldn't make sense as a self-hosted platform. Rather than burying this point in a comment, I'm going to talk about it here, because this positioning (by Microsoft) is pivotal, and it's crucial for Microsoft's future.
First, I think James is certainly right about Azure in the form you see today. Step back from the forest a little, though, and imagine an Azure that's designed specifically for enterprises to deploy into their own data centers. Certainly, you wouldn't have the (nearly) unlimited scaling that Azure promises, and certainly, you would have a much higher start up cost, which mitigates Azure's appeal for startups, but there's no question at all that Microsoft could have made a self-hosted platform that would create a compute fabric in your own data center.
You would be able to virtualize computing resources across your back-office, so as one application reached peak demand, it grabbed cycles from a pool of available computing resources in your fabric. You might have been able to spin up and spin down servers on-demand - perhaps as VM's. I'd bet it wouldn't be much of a leap to have a management interface actually power-up and power-down physical servers on demand. This wouldn't save anything on the acquisition costs of these peak-usage servers, but you'd save a ton on power to run them and cool them, and in any decent-sized data center, those costs hit the bottom line in a big way.
This model would have one more huge benefit relative to the Azure we see today: a fast network backplane. There's a huge blind spot in everyone's cloud computing visions, and it's network bandwidth. If you're counting on moving a bunch of data up to the cloud, you'd better get out your pocket calculator and figure out how long it's going to take you to get your data back and forth from your back office to the cloud and back. Start with backups, and let me know if your router is smoking yet.
So why the hosted-only model for Azure? After all, Amazon's already pushing that model of cloud computing - why not sell a competing vision where your data center becomes the cloud? This model, in fact, is the way you'd have expected Microsoft to roll out Azure, based on the way they've built product in the past - build your infrastructure any way you want, using Microsoft software to build it.
But Azure is different. Azure is hosted by Microsoft - exclusively. If you use Azure, you're going to write Microsoft a check every month (although nobody knows the terms), and that's the difference. Microsoft is deliberately going after the recurring revenue model that only hosted service can bring.
You see, Microsoft has pretty much saturated the market for its operating systems and Office software, and that's where most of its revenue comes from. Selling software products is an awesome way to drive the valuation of a software company - I'll try to do a post on the math that drives this at some point, but the key driver is that once Microsoft has written Office, it costs them effectively nothing to print another copy. This product valuation is what made Bill Gates' fortune.
Wall Street demands growth, though, and the Zune just isn't getting the job done. Microsoft, therefore, is making the big shift to a services company, or at least a hybrid. This is exactly the same transformation that Lou Gerstner managed at IBM, and it's driven from the same factors. I don't expect the transformation to be as dramatic it was at IBM, because Microsoft is starting before the ship is on the rocks. In fact, since Microsoft's software is still selling pretty well (unlike IBM's software when they made their shift), I expect Microsoft to be a hybrid company for a very long time.
There's no question at all, though, that Azure's recurring revenue model is the prize that Microsoft is fighting for here.