I had this crazy theory that what Apple meant by modular for the Mac Pro was starting off with a Mac Mini-like base.
My hypothesis is similar but diverging. See, I have my doubts when Apple said a 'modular' design that they meant 'user upgradable'.
Instead, I think they were talking specifically of how the design of the current pro had different subsystems so interdependent w.r.t. heat and power usage that it wound up being very difficult to upgrade - especially when the evolution of the components and user usage wound up being so different than expected (massive GPU power usage, low use of GPU computation outside of ML/data scientist work).
So modular may mean designs that allow them to have a wide variety of products, including potentially different form factors. So its might be possible that they will scale down to non-Xeon/integrated graphics in a mini form factor, or are otherwise planning to reuse some of that development work.
Or... not. One of the things Apple (previously) prided itself on is how little power the Mac mini used compared to comparable PCs. The Mac mini is also targeted as a starter switcher machine. I could see a new mini without USB-C or TB3 at all.
I wonder if this is what that is or something else entirely. Basically you would have a base "box" which is the processor, RAM, and logic board. It would have Intel integrated graphics and an small SSD blade so it could run on it's own. Then you can stack components on top of this: GPU(s), SSDs, HDDs, capture cards and similar components for both video/audio production. It could all connect with a series of Thunderbolt 4 connectors (perhaps a variant that allows the components to stack together like lego bricks.
Possible, although this hypothetical Thunderbolt 4 controller needs to fix some of the other issues with thunderbolt. For example, The Touch Bar MacBooks cannot power two 5k monitors from the same side because that would put both on the same controller.
There's also the issue that thunderbolt does add additional latency compared to direct PCIe. This is why Apple released eGPUs to developers in advance - code that blocks waiting on reads could be WAY slower.
Finally, you wouldn't be able to do GPU switching AFAIK with this approach.
I'd assume if it is an eGPU it has its own power supply. USB-C cables have a max power of 100W, so they'd need to put a proprietary connector on one end or work with the USB-IF on an update to the PD spec (such as negotiating higher power ratings over active cables).The thing I'm not sure about is how the power supply would work, such as needing a larger one with multiple GPUs. I'm also not sure about whether Thunderbolt 4 would be fast enough for professional, highest-end GPU work. Isn't it supposed to be around 100Gbps? Perhaps the reason it has taken this long is they've been working with Intel on that standard (or building their own?).