My guess is that the Apple Silicon Mac Pro (at least any half sized one) won't have any graphics upgrade capability. That's why Gurman has also said there will be likely an 8,1 Intel Mac Pro as well. That'll be the differentiator, if you want GPU upgrades or multiple GPUs you buy the Intel one. Otherwise you buy the Apple Silicon one. Basically I think the Apple Silicon Mac Pro will likely be the new trash can.
The Apple shouldn't release a new "Mac Pro". It could be a Mac DM or Mac MD ( M-series Desktop or vice vera). OR even Mac Pro M . ( which could be read as Mini or M-series. ) The last would be used to claim complete transition when still had a hold out.
But half sized leaves a huge amount of volume to fill with what? I slice in half on one dimension then that is "half". if slice on two dimensions really getting something more akin to a 1/4 sized system; not half.
If the Apple Silicon Mac Pro could take GPU upgrades, they'd likely bring MPX to Apple Silicon. No way a half sized Mac Pro could fit an MPX module, especially if there is any reduction in the depth of the case.
The major problem with doing just one MPX bay would more so constraints Apple imposed on itself. It is just a matter of effectively using the volume space.
The Mac Pro 20.8 inches high (~60cm) . half of that is 10.4 inches ( 30cm ) . Two 140mm stack up to 28cm So a two high fan stack is under the 10.4 inches.
If just "chop off" the middle fan in the current Mac pro (and associated MPX ) bay , you then end up with a CPU zone and a MPX (+ power supply ) all inside of 10.4 inches. If it is going to be a desktop targeted machine then don't need the one (or so) inch feet risers. [ Apple could keep the bottom plate above and attach some thin "non skid" feet to the bottom so doesn't scratch desktops. And still be able to upsell from multiple hundred wheels for those who wanted to upgrade. And couple hundred feet to put it on the floor ]
If then gave up on the space frame's upper handles then even easier to hit the 10" mark. Keep the dual sided board but only really need to move the NAND daughter cards up into the CPU zone ( if soldering the RAM onto the SOC that is flipping to the front. ). The box would probably be heavy enough that some handles would be very helpful.
Provision 4 TB sockets with the onboard iGPU display/TB output and just have a simple A/B Display port switch for an dual I/O card TB ports ( either iGPU or MPX module). That would simplify the backside board complexity.
Dropping the height to 10.2 inches means the height to width percentage would be 1.18. The Mac Cube actually was not an actual cube. Its height to width ration was 1.27 ( 9.8 x 7.7 x 7.7 ) That "half" Mac Pro frontal profile would be more square than the cube was. If sitting at the desk and looking at just the front of the system it would be a "like a cube".
The Mini has drifted between 6.5 inche to 7.7 inch desktop footprint. The Mac Pro 2013 was 6.6 X 6.6 footprint ( circular but inside that square). That's where the real problem kicks in. If there is some OCD Apple rule that states no desktop Mac can be have bugger than a 8 x 8 inch footprint on the desktop. That is what would "kill" the MPX slot.
Similarly if the primary concern is what the sideways profile of the system is then it would not be "cube like".
However, Physically it can be done. The more likely root cause issue is that the Apple's bigger M-series doesn't have the PCI-e I/O to provision a MPX bay and decent dual 10GbE ports and a some enhanced I/O. That would be a bigger driver to shift to a 1/4 - 1/6 sized enclosure that squeeze out the slots as a 'fig leaf' to cover up that shortcoming.
Beyond issues like the number of PCIe lanes, there is also a pretty clear reason why Apple is stalling on doing a Mac Pro tower in Apple Silicon: It would be the only Apple Silicon Mac with discrete graphics. Apple may not want to bring upgradable graphics to Apple Silicon because it would create a weird configuration that doesn't match the rest of the lineup.
it isn't a "weird configuration". It is just doing the work that desktop Operating System providers are suppose to do. As long as there is always a baseline Apple iGPU then they can just run iPhone apps on that. If the display output is on a non Apple GPU then just copy that frame buffer subset over to the other GPU. [ or just don't run iPhone apps on non Apple GPU driven displays. ]
This is more so Apple being anticompetitive .... which the longer they drag their feet the more likely they will have to sit through even more unpleasant regulatory hearings ( and possible fines ).
Nvidia being a bad partner and stop signing their drivers is one thing. But to throw Intel and AMD out under the bus also so that there is no other option ... that isn't technical difficulty... that is a business ( Scrooge Mc Duck money pit filling) plan.
There are lots of details to get right. So I understand it wasn't there in year 1. They also want developers to throughly optimize for Apple GPU specifics ( unless fully leverage all the features gets harder to complete with the upper half of the discrete market ) .... But by year 2 ...
But for Mac Apps that have been dealing with Intel/AMD/etc GPUs all along.... what is the major app hurdle here. There really should be one.
I hope Apple changes their mind on eGPUs, but it seems like there is a similar thing happening with eGPUs. Apple doesn't want to fragment graphics on Apple Silicon by having discrete GPUs be possible.
Again... besides running native iPhone apps ... this shouldn't be super hard because it already happens on the Intel side. Squeezing frame buffers over TB would be much more problematical.
That rigid homogeneity would be cutting off their MBA and MBP from heavy duty computational horsepower. dock laptop at desk and get horsepower and take more mobile system on the road.
And as long as Apple was screwing over Thunderbolt uniformity and Intel's business... I don't see see them getting a Thunderbolt label much in the long term future if stay on that path (which also coupled to not letting another other OS natively boot and run either. )
If Apple brings back eGPUs on Apple Silicon in macOS 12, that might be a sign that things are changing. But at this point I really doubt we'll ever see support for anything but integrated GPUs on Apple Silicon, software or hardware.
Apple will probably do a few more revisions to the Intel tower, like Gurman has indicated, and hope the tower users all just go away. Kind of like they hoped with the 2013.
This is the part that doesn't mean rational sense. If Apple stumbled and fell on their face with a sizable segment of the Mac Pro market with the ill-motivated hope in 2013 .... why would it work better now with that specific group?
And if going to iterate more than once on Intel ... that will run into conflict with the cost saving mode of putting macOS x86_64 into "background mode" build status.
More likely is zero iterations on the core system and perhaps tweaking the GPU support over time to keep folks who are buying the increasingly stale system happy over an extended period of time. Mac Pro will get cheaper for Apple to sell because it will get into "paid for" status and the margins will grow. ( Apple sold the MP 2013 right up till they shipped the MP 2019 models. ). Just would be a big cash cow.