Perhaps the ASICs should be designed so that they are a little more general-purpose then. Perhaps capable of mining for the next cryptocoin on the block?
ASICs only offer a benefit through significant specialisation. Specifically with bitcion the double-SHA256 hash lends itself to an extremely efficient hardware operation which beats any general-purpose system substantially.
In general what you want to do is design the algorithms differently. There's a few reasons to do this. Firstly your new coin probably doesn't want to run best on hardware which is already used by a bigger coin, especially bitcoin, because then there's a massive attack vector just looming over you as the existing miners could just turn a fraction of their processing power towards your network and destroy it easily. This means you generally want to come up with a novel proof of work function which is not easily computed by existing miners. The best form of this is using a completely different hardware resource.
Secondly (and apart from the few large coins which generally got to this idea first, competing with the first point), if you care about keeping your miners distributed in terms of ownership and control, it helps if said algorithm is tuned for some general purpose bit of hardware your users are already likely to have for some other purpose. Ethereum's PoW was expliticly designed to be optimal for GPUs (because computing it is memory-bandwidth limited), and Monero's was designed to be optimal for CPUs. Chia's system was optimised for storage space instead of compute power. Lower-effort clones (e.g. dogecoin) tend to at least change the hashing algorithm used at little to provide some protection, but to a lesser degree (e.g. Ethereum is probably safe from attacks by bitcoin miners, Monero is probably safe from ethereum miners, but while dogecoin is probably safe from attacks by bitcoin miners, it's not safe from the other larger coins, and especially bitcoin forks have actually been attacked by bitcoin miners).
No amount of "ASIC unfriendlyness" will bring back the cute old days of distributed desktop mining. Concentration to a few specialists is the natural end game of any PoW scheme, no matter how much effort you put into inconveniencing them. If a highly concentrated cryptocoin network is not good enough, cryptocoin is not good enough.
(your last point however is valid nonetheless, the scenario of large blocks of capacity jumping between networks and how deliberate algorithm design tweaks might offer some protection)
FWIW, Dogecoin is a fork of Litecoin which was originally designed to be GPU and ASIC proof, but that claim didn't last long and there are ASICs that can mine them.
This is sort of what Nvidia did with the cards they sold directly to miners. At the end of the day they're still GPUs (albeit without display outputs) so they could in theory be repurposed for work that actually benefits society.
In practice that would never happen because by the time miners are done with the cards they'd be too out of date for machine learning or other general compute workloads, and useless for budget gamers because of the lack of display outputs. It's really a shame how much power and silicon is being wasted on cryptomining.
But pure ASICs get rid of even more of the GPU than just the output stage.
A pure ASIC to cover the fundamental calculations in hardware with just a CPU to feed them data and pull results is the most efficient.
Think of something as direct as a cell phone modem. The batteries last as long as they do because a ton of the low level modem behavior is done in accelerator blocks direct in hardware, and the DSPs just route data between the blocks and manage the higher level protocols.
I don't disagree, I'm just saying that Nvidia literally shipped normal Ampere cards directly to miners. Ethereum is tricky to build purpose built ASICs for.
I guess that explains why GPUs are so cheap and plentiful.
Or perhaps they're mining another crypto currency.