We know the machines have to get smarter, the goal is to be always moving forward, adding more value, cutting costs, getting more and getting it faster. Just to discuss briefly, the first school of thought represented is that of Cisco, related by John Chambers (CEO). A person can look at things from a network hardware point of view, watching the development of network hardware and observe that as time goes on, applications move into switches and network hardware. They become network services and are then expected on the hardware, such as a firewall. This assimilation brings with it many benefits which are blood on the lips of a thirsty greenback seekers who seek to find a model and ride it until the money stops flowing. You need only see the toilet paper roll/iPod dock to know what I am talking about. From this perspective, if some applications moving to network hardware is good, more would be even better, right? Cisco's latest marketing materials and corporate direction is towards very 'smart' hardware that can do amazing things for an industry seeking virtualization. I have mentioned Cisco, but make no mistake, this is not a one horse show. Many other companies are readying for the same vision.
The idea, non-technically, is to make all of the resources in a data center flexible, so that where you need the power is where it goes. Machines that have special abilities being used for what they are most effective at performing. A most exciting concept is utilizing processor power or memory from other servers that are not running that specific application. The systems for this are often proprietary, expensive, and a massive undertaking. In addition to the high starting cost, the distinct possibility of high ongoing costs as a result of proprietary hardware/software looms over this concept to the point that I wouldn't expect the market to take to it quickly, if at all.
The other side of the house would comment that they've observed increased processing power in even the smallest 'end' devices. This exponential growth in power and subsequently power/cost leads to the conclusion that the network should remain as it is, the link between devices as opposed to replacements for them.
Where the line will be drawn is uncertain, although my wager would not be with the smart network crowd, it seems more likely we'll see a compromise than either vision capturing the majority.
The Letter J