For the last few years, 'virtualization' has been all the rage for the corporate IT set, and hence for many venture funders. The idea of being able to shuffle applications and operating system images among physical servers, rather than let those with low loads stand idle, burning capital and energy, makes a lot of sense. But even a good thing can be carried to extremes, and awake a counter-trend. Enter the antithesis: physicalization, a proposal to break the hardware processing elements themselves down into smaller bits.
To be sure, the meme has been in part hatched by an interested vendor whose products misfit the current trend, but they are raising some valid points. Compaction of processing into ever-more-dense multicore units might have overshot its economic sweet spot, at least for some users. With commercial real estate in free fall, optimizing every square foot of data center floor space may not be so important, nor does spending tight capital on rebuilding HVAC to deal with higher energy densities seem such a great idea. More to the point of industry economics, the high end servers used for both virtualization and 'scale up' approaches are one of the few places where the silicon merchants still command high margins. The physicalization approach relies on more commoditized silicon, the processors cranked out en masse for consumer-grade machines, potentially putting some of that profit back in the hands of the IT managers. If this approaches sticks, look for some margin pressure on Intel et. al.
The third stage is synthesis, and indeed the two approaches are not that dissimilar, if regarded from the software point of view. Both have the issue of managing a plethora of deployments over a dynamic computing base. Adopting a physicalization approach won't get out of that problem, since their smaller power implies the need to be able to rapidly stand-up new images on further mini-servers when loads increase. Figuratively speaking, virtualization creates the problem of managing many small logical boxes (VMs and app images) within large physical boxes. Physicalization will pose the dual problem of managing large logical process boxes across many small physical 'boxes'. We've been there before. (Now where did I put my old 'grid' slides?)