Someone is paying for Larry Ellison's Gulfstream V. Oracle software and licensing sales last year (including non-DB revenue) were around $20B; MS SQL Server recently crossed the $1B/year revenue line.
Although, to your point, it's rather amazing that so many people do.
"(EC2 is basically a really complicated way of charging for electricity.)"
That's the takeaway. I do agree with this basic premise though. That said, it's not just charging for electricity, but electricity + profit for them. I do think it's possible to run your own operations cheaper than outsourcing. I think a lot of pure "cloud" solutions are just smoke on top of some jacked up pricing. A lot of people are billing by weird bandwith+server+cpu+disk metrics that are hard to evaluate. If it's hard to figure out, your probably getting worked. Unless they're really adding something useful at an operational level, do it yourself. Hardware is relatively cheap. Guys who can build and configure hardware are reasonably available. Having total control over all that gives you some flexibility and that's where you add your "special sauce".
EC2 really makes sense for lumpy loads where it's not efficient to invest capital in hardware for bursts of processing you want done. With a more predictable load, it makes less sense.
In my opinion that's a silly statement. You might as well say that a restaurant is a really complicated way of charging for sunlight and water. (Or if you want to go farther, a complicated way of charging for the CP violation in the laws of physics which caused a slight excess of matter over anti-matter during the big bang.) Reductionism to that level is unhelpful.
Upvoted for a well-presented opinion, although I disagree. While the author's reductionism may seem extreme, I think it's legit for his argument, which is that electricity becomes the dominant cost over time, and that realizing that fundamentally shifts how we think about the economics of hosting an app.
The next step is hardware-based key-value stores. Generic Wintel PC servers waste an incredible amount of power if all you need is a RAM-based key value store connected to the network. Contact me with funding proposal if interested .. :-)
A very good point. This is why I don't trust people telling me "fast is not important, scalable is". Actually they are going to be both very important, as you want to scale using the minimum CPU cycles required.
