The primary reason is now performance. Don't believe me? Look at how RAM is arranged on a GPU or in a game console. The RAM chips are located as close to the CPU/GPU as physically possible, limited only by thermal constraints.
You will never see DIMM slots on an NVIDIA card because the compromises would be untenable. Longer traces means lower performance, while sockets will make the cards thicker and with poorer thermals. The reasons why NVIDIA cards are the way they are correlates with why Apple's latest comptuers are the way they are.
What Apple did with packaging RAM so close to the M1/M2 chip is a big part of the Apple Silicon performance story.
Are those ram chips on a CPU carrier in a POP configuration? or are they fanned out on the PCB allowing technician to replace individual components?
Part of the reason for Apple move to ram on SOC carrier was killing third party market. Now its impossible to repair/upgrade ram. Similar with T1/T2 proprietary pcie NAND in raid configuration.
I love how you've pushed your opinion as fact here, spinning the positives and conveniently omitting the clear anti-consumer advantage that soldering RAM achieved: non-serviceability meant shorter lifetimes, i.e. more replacement sales (and hence more e-waste, but who cares about externalized costs, right?) and more money minted during the original sale (since people couldn't buy and upgrade with cheaper components, they were forced to pay Apple prices). At the time it was implemented, there was no real technical advantage.
Performance benefits have now come along as a bonus, a happy side-effect. But the main mover has always been (and always will be, with Apple) pure and distilled greed.