On the surface I get your point, but I still don't really buy it. Is economies of scale really an issue at this point? Their supply chain has been /mastered/ at this point,
"mastered" isn't economies of scale. Just because operating at economies of scale in year n doesn't mean you get a 'free ride' in year n+2 . They have to sell just as much , if not more, each year to keep it. You can't 'bank' , 'store' economies of scale and then trot it out in the future when you need it.
i don't see any reason to not update it in the mac pro unless they arent planning to or its getting a newer variant.
If they don't have it , then hard to ship what you don't have. The M3 Max die used to not have the connector. Now it does. In the "used to not have" state the amount they had to ship was zero. They can't ship zero.
First, early on TMSC N3B had yield growth issues. Deferring a larger chip unitil later when yields are higher would be a very good reason.
Second, By using separate chips for laptops versus desktops you
LOOSE some economies of scale since all the laptop contribution is
gone. The laptops don't count for dies shipped with a UltraFusion connector because they don't have it. They some shared expenses in the overlap ( CPU clusters , GPU clusters , NPU clusters , A/V en/decoders ). But that shared is only just getting the Ultra M3 into existence.
Similarly the more the laptop Max configurations outsell the Studio/MP then the more silicon wafers just wasting with zero value , dead silicon . 500K Max 'full size' in laptops some waste. 1M 'full size' in laptops more waste . 2M 'full size' in laptop even more waste. Economies of scale has impact in context of waste also. As the number of laptop vs desktop deployments diverge it gets worse.
The Studio skipping the M3 Max means they don't get scale there either. Apple very likely juiced production of the laptop variant of M3 Max to just fit the end of the production run of The M3 Max MBP 14/16". When that stopped there is very good chance they turned on the stuff heading for Mac Studio. ( Remember also that the top volume selling MBA M3 is soak up most of the TSMC N3B wafers. If that demand doesn't collapse there are likely not lots of extra wafers lying around. Intel has ramped fulling on N3B at this point. ).
With the M3 generation they are also missing overlaps in reused design M3 Pro since its layout/floorplan is different also. M1/M2 the Pro was 99% the same CPU/NPU/Thunderbolt/ "x1 PCI-e' design as the Max. The Max was add more GPU clusters , AV en/decode , and UltraFusion connection. Again a backslide in scaled sharing of expenses. So the 'pro' is retreating now also from helping to offset the Ultra.
Additionally, if Apple has pushed the size of the M3 Max die so big that Apple had to move from InFo-LSI to CoWoS-LSi then there is no CoWoS-LSI capacity there either.
Good news: CoWoS capacity is set to be expanded.
www.tomshardware.com
Nvidia and other AI data center players have bought up all of that capacity. Even if Apple had the raw Max die there may not be production capacity to join them together.
The Mac Pro sells at a low enough run rate , it would probably help to get several M3 Ultra studio's out there. Folks can benchmark the SoC to see if useful. If it works and folks need higher I/O then they just wait for a Mac Pro version. If it shows up in June and these evals were all completed until May ... there is really no material long term impact.
If there is some software 'value add' for the Mac Pro then wouldn't be much long term material impact if have to wait for macOS 16. (or a far more stable version of 15 ).