M.2 drives are indisputably a massive step backwards from the ease of serviceability of 2.5"/3.5" drives. For the latter, to swap a drive you unplug two cables from the existing drive, plug those same two cables into the new drive, and bob's your auntie. Whereas with M.2 you need to get to the motherboard (which is typically on its side in a case) and possibly remove expansion card(s) to access the M.2 slot and drive; you need to unscrew that tiny fiddly screw (make sure not to strip it or lose it!); you need to be careful that you don't slip and smash, or drop, the screwdriver into the motherboard; and you need to be careful that you don't drop or smash the tiny, light, fiddly M.2 drive.
The other advantage of 2.5" drives that I haven't seen mentioned is their superior heat dissipation properties compared to M.2 drives. The latest PCIe 4.0 M.2 controllers are approaching power consumption of 10W and I reckon that PCIe 5.0 controllers will breach that threshold, which exceeds what an entire motherboard chipset draws in some cases. Attempting to cool that on the motherboard, which is already a relatively hot and airflow-less area, is not a good idea - M.2 heatsinks are available but area also a kludge that cause more compatibility issues with Z-heights. We're even starting to see motherboards that are designed around M.2 slots and cooling them, which is just silly to me; the way to deal with heat-producing components is to move them away from each other and put them in containers that are able to absorb and dissipate that heat well, and that's exactly what 2.5" enclosures do.
So, what's really needed is a replacement for SATA, not an update to the standard. U.2 failed on the desktop precisely because it was built on top of SATA Express, using the large and clunky SFF-8639 connector, which made the drives, cables and connectors unreasonably complex and therefore expensive. What U.2 should have done was dump anything SATA and instead be a simple 4-lane PCIe connector on the motherboard, connecting to a simple 4-lane PCIe cable, that connects to a simple 4-lane PCIe connector on a 2.5" drive, all talking NVMe. Essentially, an M.2 slot designed for use with a cable as opposed to direct motherboard connection.
And we already have a working connector and cable standard that exposes and uses four PCIe lanes, while providing up to 15W of power over a 0.8m passive (therefore cheaper) cable: Thunderbolt 3. With some redesign to eliminate the unnecessary DisplayPort support, harden against the frequency emissions inside a computer chassis, and make the connectors physically dissimilar to USB-C, I don't see any reason why a TB3 derivative ("TB3 NVMe") can't be the true successor to SATA on the desktop.
The only downside of course, is that you now have 3 data connectors (SATA, M.2, and now TB3 NVMe). But I think you'd very quickly see TB3 NVMe become the dominant standard over all three because it makes drive swaps simple again; SATA would fall away because the SSD manufacturers would abandon it in favour of TB3 NVMe for 2.5" drives, making HDD manufacturers the only SATA users; and you'd see far fewer M.2 slots because TB3 NVMe allows you to have the performance of M.2 NVMe with the installation simplicity of a 2.5" drive.
Honestly, this could lead to a different form-factor for drives too. Instead of the tiny M.2 or the huge 2.5" form factors, why not one in between that is essentially a combination cooling and protective enclosure for an M.2 drive? Kinda like those external SSD-to-USB adapters that you can buy, except a little larger (1.25") so you can fit two of them in the space a current 2.5" drive goes. That would satisfy both those who like M.2 because it's small and those who like SATA because it's simple, and would neatly overcome the heat dissipation problem of high-end SSDs.
This is an interesting idea, though to me it doesn't seem like you've quite thought it through in terms of its implications. Specifically: controllers, efficiency, and complexity.
I mean, on the one hand you argue that u.2 has failed to gain any consumer adoption because it's "sing the large and clunky SFF-8639 connector, which made the drives, cables and connectors unreasonably complex and therefore expensive", while on the other you argue for basing local storage on a signalling standard that requires not one but
two new controllers (TB host and TB device) for every device, and that is overly complex in how it takes a PCIe signal (+ more), translates it into a different signal, then translates it back at the other end. While this is highly useful for TB's intended use - docks, monitors, and external single-port expansion - it will be
far more costly than any u.2 product. And, of course, this would drive motherboard costs through the roof due to the need to add 4+ internal TB controllers. That's easily another $40 in BOM cost, before taking into account how it would affect motherboard layouts and production.
The connector that you argue made u.2 expensive is, after all, just two different physical interfaces for native PCIe. There is no possible world in which it is cheaper to take that PCIe signal, convert it, convert it back, and then connect a storage controller, than to just run a decent quality PCIe riser cable. An u.2 cable kit with an m.2-to-u.2 adapter board can be had for ~$35. The price premium for literally any TB3 product over comparable non-TB products tends to be closer to $100 - and that's just for one side of the equation. TB3 also has extremely short maximum cable lengths for passive cabling, which will only be shorter inside of a case due to increased noise. And nobody wants active cabling for their storage. Plus, of course, even passive external TB3 cables are
expensive. Internal ones would be worse.
I think there's a
far superior alternative to what you're suggesting here, that already exists:
OCuLink. OCuLink transmits native PCIe x4 or 4x SATA through a single cable, has a small, user-friendly port, can be either copper or optical for longer runs/higher speeds, requires no additional chips for the copper variants, and can be broken out into whatever device-side PCIe-based connector you might want - OCuLink, PCIe, SFF-8639, whatever. It would be entirely possible to implement native OCuLink SSDs in a 2.5"-ish form factor.
Oh, and one challenge with your proposed 1.25" form factor: power delivery. That's too narrow to reasonably fit a OCuLink or USB-C cable alongside a SATA power connector, so you'd either need a new power connector, in-line power injection in the data cables, or power transmitted from the board. Definitely something that could be overcome, but it's an additional challenge.
Cooling ever more power hungry SSD controllers in the m.2 form factor is definitely a challenge, though IMO part of the solution has to be to stop the ridiculous race towards these horrendously inefficient designs that just look good in benchmarks. But beyond that, moving m.2 slots onto some sort of riser card is a simple and functional solution for this - and widely supported, though of course not universally. We need more platform PCIe lanes, and crucially more native bifurcation support, but that can be done in far simpler and cheaper ways than internal TB3 SSDs.