There are a couple of problems with your theory, though it could play out that way. Right now the smaller vendors are often more efficient than the larger ones. Smaller players can be more nimble.

Second the larger players all have vastly different models. Just to pick a few examples of the bigger players
AWS -- Generic low quality server experience offered cheaply. Walmart
Sungard -- Highly custom environments quality management lots of value added labor
Verizon (was Terremark) -- Moderately custom environments, mix of high performance cloud IaaS and colo. Some value added services with strong partner service model.
Oracle -- Unified cloud stack offering IaaS plus advanced management especially knowledge of Oracle applications
Azure -- IaaS with Microsoft based PaaS. Good pricing on SQL Server.

How do those consolidate? I think we are looking at a situation more like clothing where stores are genuinely different fulfilling niches for various customers.

The crossover between tricky to move is much lower than the crossover for better to run your own datacenter. Not necessarily the case though for better to run your own cloud out of someone else's colo or better to jointly administer a cloud with a colo provider. So this can happen. The cost of multiple good quality data centers is very very high compared to the cost of getting data out of one.

As far as GP's post. He's wrong. First off clouds are designed to scale so adding another copy of parts of the data for replication is not hard. Second you can do crossover networking from one data center to another if you need to move and the cost of using the cloud provider's bandwidth is too high. Also there are devices that can be physically connected to the racks and then trucked (think backup drive moving physically but 20-100x scale). Mainly a data move is the sort of thing an agent can coordinate easily.

This is demand driven investment. How could that be a bubble? The bubble can happen later when cloud investment starts to outpace customer demand or even starts to generate its own cross over demand. But right now this is not a bubble this is a success.

That's not what's happening, they are buying more servers. The types of computational workloads are also shifting during the cloud migrations. The savings are coming from: staffing efficiencies, reduced real estate costs, reduced power costs, reductions in physical security, savings in the procurement process...

Excellent point, and good example.

Yes that's a fair characterization. For companies below the line, i.e. those that would lose a lawsuit easily this is helpful.

A server not connected to a network in a physically secure location was the situation for the computer that Bradley Manning stole from.

Centcom is interested in starting to build a government infrastructure for defense. They agree this needs more collective action and government assistance. Right now the public is pulling in the opposite direction however.

I agree there are terrible auditors that don't understand what they are doing. But in most companies you can push back against that, it is just that then the burden switches to you. You have to verify and certify that alternative approach X is better than industry standard approach Y.

As far as the rest, the purpose of an insurance company is to pool risk. The person being insured should likely not want to have to file a claim because that means something bad happened. The company doesn't want to give nothing in return because then there is no need for their product.

Industry handles this in other areas and for that matter security as well by having auditing firms and engaging in a "best practices" audit. "Best practices" doesn't actually mean best practice but rather not doing stupid or dangerous stuff. The audit is how that gets determined.

They ran into financial trouble as Mandrake. The distribution was moderately profitable but they lost their shirt on an educational software venture that failed.

I liked them late 1990s (Mandrake) they were my favorite distribution because so many things "just worked" and their configurations were often more sensible. You started off far closer to a working system.

Didn't try the server product much though did use it once for a RAID product and it did a great job on defaulting the RAID.

Linux driver support definitely is a bit crappier, but it's a lot better than it was even say 5 years ago.

My experience is that it has gotten worse. 5 years ago I could pretty much run an arbitrary Linux distribution on an arbitrary 1 year old laptop and have say an 80% chance of few if any problems. Today most interesting laptops have whole swaths of features not covered and many drivers not included. I think hardware got more interesting and the Linux community has gotten less focused on desktop (understandably) and the result has been a huge downgrade in terms of compatibility.

LAN Manager was multiuser. The client wasn't but that doesn't make much difference as the non-multiuser smartphones phones using apps and websites today proves quite well.

I see turbidostato below made the same point.