If Intel manages to get a foothold in the discrete GPU add-in-card market and keeps a good enough relationship with macOS get signed drivers through then very good chance one of their 2022-24 will. DisplayPort 2.0 using the Thunderbolt protocol. That isn't necessarily TBv3 support but does mean there will be some infrastructure that supports the basic TB protocol in the GPU subsystem for displays. If Intel has to put part of it there , then they are pretty likley going to put the whole thing there. And video decode/encode is where they may have traction if grow out starting from this starting point.
Intel Server GPU Shown for Video Transcoding Applications (servethehome.com)
That is lots of execution for Intel to get right. But also indicative that Apple doesn't have to go "reinvent the wheel" if they want to enable a multiple standard video decode streamer card if just keep the good working relationship with Intel graphics drivers folks. And Apple not going draconian on GPU drivers where they are the sole solution possible.
AMD is probably walking a 1-2 year slower pace. They took small "baby steps" by putting USB Type-C ports on new RDNA2 cards. For example, see specs here:
AMD Radeon RX 6800 XT Graphics | AMD
A Type-C port isn't Thunderbolt v3 support, but it is starting process of normalizing its presence. AMD is likely going to implement DPv2 on next iteration (RDND3 or 4 ) . That still won't be thunderbolt v3. Once they get to PCI-e v5 on the cards ( 'extra' bandwidth that may not be missed ) and some standard TBv3 design libraries that AMD can license affordably , then I suspect it will make its way onto the add-in-cards. [ Even with just DPv2 would highly likely get a picture on XDR , just not working XDR USB ports. ]
Even without MPX there is a decent chance it will show up eventually. Sooner with MPX would be better for the Apple motivated display docking station ecosystem that Apple has fostered over the last 5-7 years ( XDR , LG Ultrafine iterations , legacy 27" docking station) . MPX connector is a bit ahead of the curve relative to the rest of the industry.
I'd be more concerned whether would "loose" the Infinity Fabric link between the too than with Resolve slowing to lowest common denominator. Haven't seen how much AMD is supporting multi generational Infinity Fabric (IF) generations ( may not ).
if Resolve isn't using the IF connection then it is leaving performance on the floor even with two twin cards in addition to the un-optimized, non segmented workload distribution management system. I doubt "lowest common denominator" will be a constrain for all of most high end apps for over the long haul. ( some operations my be throttled that way. But that "everything has to be 100% uniform to scale" has major disconnects with the application cores no at this point also. Both Apple's M-series and Intel's 2022 implementations. )
A hefty chunk of the price increase for the Vega II variants is for the Infinity Fabric. And if not using the link , then it is going to be much more difficult to get deep value out of the cards.
If stuck with lots of input footage that the fixed function of a W6900 MPX could chew through easily then the trade off should play out for a number of workloads ( presuming the drivers effectively leverage new fixed function logic. )
