My guess is that the iMac Pro was always intended to be Apple's answer to people clamouring for the Mac Pro. As in, Apple never had any plans to update the Mac Pro. The Mac Pro was likely decided at the last minute to placate the huge outburst over the 2016 MacBook Pro but by that time, the iMac Pro was already a done deal.
I think that's partly right, but I doubt it had anything
directly to do with the 2016 MacBook Pro - they made it pretty clear that the problem was their inability to update the nMP cylinder. Lots of businesses buy or lease their IT equipment on a 3/4 year cycle and would be starting to consider replacing their 2014 mMPs. Meanwhile, for those who do hang on to old equipment, the last proper Mac Pro - the Mid 2010 cheesegrater -
was approaching obsolescence. Sounds to me like they got a load of pushback from some big pro customers and third-party pro equipment makers.
Agree that the iMac Pro would have been in the pipeline before then, though.
It seems unlikely the modular Mac Pro will be a PCIe slot box OR use the ATX form factor. The ATX form-factor dates to the mid-1990s.
...but it gets the job done in applications where size and weight are not the prime considerations (so what if the power supply connections are a bit old-fashioned?).
Here's Apple's problem: there probably
isn't a vast market for the Mac Pro - but it is a
strategically important market if Apple want MacOS to remain a serious computing platform, because some important jobs just need a pickup truck and not a sports saloon. Microsoft can concentrate their hardware efforts on their "Surface" range for the deep-pocketed consumer, safe in the knowledge that anybody that
doesn't fit that profile can still go out and get a big box'o'slots tailored exactly to their needs and run Windows on it. That option isn't open to pro MacOS users (hackintoshes rely on ignoring Apple's software license terms, are totally unsupported and could be rendered useless overnight should Apple decide to clamp down).
The simple solution to that is to produce a fairly conventional "tower" system requiring the minimum of expensive R&D that would provide a bespoke "plan B" for any important customer for who's needs weren't served by the iMac or MBP. Now, a few years ago, that might have decimated the market for high-end laptops and all-in-one's, but today the majority
want laptops, so I'd say that the only people who
want towers are the people who need them. It doesn't have to be ATX
per se - cutting some of the legacy cruft from an ATX-like design would still be easier, quicker and cheaper than starting completely from scratch. The real mistake was made years ago when they let the Cheesegrater wither on the vine.
What Apple seem intent on doing, however, is producing something completely new, revolutionary, out-of-the-box (and, probably, with complete proprietary lock-in when it comes to GPUs, at least) that will cost a fortune to design and build, will look great in the next edition of the coffee-table book and quite possibly repeat the
real mistake of the nMP cylinder: designer lust.