The old excuse that the hardware was too limited to support that kind of functionality is long since over. There is really no excuse anymore.
But beneath this, one of the root issues at least in mobile devices is the system on chip nature of the design meaning you're at the mercy of the chipmaker to maintain appropriate drivers. These are invariably very very closed source on phones, as some areas like imaging and camera support were (and are now still, I guess) being bought in as super expensive IP blocks with serious NDAs and restrictions on what can be made available.
One reason Android is as fragmented as it was, was that they kept changing the system HAL, and chipset makers weren't providing upgraded drivers to support the new HAL as, from their perspective, the chip was end-of-life, old news, last year's model not making them any more money. I assume it isn't quite as bad on server, but single board computers running mainline Linux without patches, and with all peripheral and SoC component support are very rare, and noteworthy when they are found. Most still have pretty closed GPUs.
But when it's all on one chip, and you don't have any hardware discovery process, or even a standard way to boot the CPU, it's hard to see much of a generic future, as much as one would be great!
https://en.wikipedia.org/wiki/Server_Base_System_Architectur...
They were more open than an iPhone, but they were not particularly open to anyone with less than a CE/EE level of computer and electronics expertise. They certainly were not designed to be modular and open like the clone PCs that came later.
I'm worried mobile phones are reducing the expectation that the average consumer has of being able to modify, upgrade, and repair their devices, and computer manufacturers are poised to take advantage of this, rather than the other way around.
We may be about to enter an era more reminiscent of the PC clone era, which was more boring from an aesthetic point of view but far more exciting in terms of innovation and opportunity.
New platforms usually start closed and then open over time.
I think a similar transformation is slowly building in the cloud space, but it's very quiet and nowhere near critical mass. I'd keep an eye on cheap bare metal hosts like packet.net and OVH, things that make it really easy to self-host and self-manage K8S clusters, and truly RAIN (redundant array of inexpensive nodes) ready databases like CockroachDB. Pretty soon you'll be able to combine those things and host at roughly 1/10th the cost of AWS or less, with absolutely zero lock-in. At the very least it will kick off a price war in conventional cloud.
Alternately if Starlink brings much faster connectivity to the edge while simultaneously kicking terrestrial broadband in the butt to offer competitive offerings, we could see a move of some hosting back to the edge where 99.9%+ uptime is not really needed. That could include batch processing, model training, analytics, and other stuff that doesn't directly power real-time customer API or UI interactions.