So, you're a 5 person company that makes a widget for some specialised industry.
over the last 4 years you've spent 30% of your turnover building an enclosure, getting the moulds made, and have orders for 3k devices.
They need to be delivered inside 3 months.
A new enclosure mould will cost 5 months turnover to produce, and god knows how many man months to prototype.
The software, sensor package and PCB will need to be re-designed, certified and tested from scratch.
bankruptcy.
Whereas if you have a complete product designed around an SBC, respinning that is expensive, and a few hundred to a few thousand units is an ugly point where that leads to really high per-unit costs.
Whereas if you make millions of a thing, the overall development cost is similar, but spreads over many more units. (EDIT: you also probably aren't using an SBC, for similar-ish reasons: a customized board is saving you money per-unit, and millions is enough to amortize the up-front cost of that and still come out ahead)
So the economic perspective IMHO makes a lot of sense.
Of course you also have large companies making low-volume products using Pis, where I guess one could argue more that those companies should be able to take the hit of temporarily not being able to make as many as they'd like. But for smaller companies without large buffers or other business, its a big problem.
I'd be curious to what degree they've looked into the exact cases when deciding who to support.
imo the failure to ramp up production is the unforgivable part
I'm inclined to say that that's true, and even that it might make sense in terms of money for the RPi folks to prioritize businesses for that reason, but it's worth remembering that it cuts both ways: Individuals also remember which vendor screwed them over.
However, I suspect that individuals are more likely to tolerate delays in hardware acquisition, and also more likely to stick with an RPi due to the ecosystem offering vastly greater levels of support and existing solutions.
(When I struggled to lay my hands on a Pi Zero a while back, I looked for other options, but found various negatives for the potential alternatives.)