If a program wants to take advantage of newer CPU features it either becomes incompatible with any earlier CPU, or requires multiple code paths to handle different processors. Or you can use a Linux distribution like Gentoo and compile everything with the matching -mcpu flag.

Most publishers or binary software don't want to maintain multiple code paths, don't want to cut off potential customers with older hardware (even if the performance sill suck). At most they might do multiple code paths for specific performance sensitive tasks.

In the 90s those RISC processors had a huge performance advantage, but they came at a serious price premium and were typically only available from a single vendor. ARM is widely available, affordable, and covers the whole price/performance range.

A lot of software these days is delivered as webapps which are platform agnostic.
Mobile didn't really exist in the 90s but is huge now, mobile is already dominated by ARM.
Emulation / dynamic recompilation has improved a lot if you need to run legacy binaries.

A lot of people simply do not need windows these days, many people exclusively use android devices - especially in developing countries which tend to be heavily mobile-first.

Not really, it's not a high end machine but it's perfectly capable for most users. And with the price of memory going up people are still buying windows laptops with 8GB or even 4, which will perform much worse than the macbook.

You didn't read my comment at all. This isn't about buying burner phones and data plans, it's about getting old but still working devices (which are basically available for free) and using free public wifi networks which are all over the place these days.
You don't need much financial means for this.

There have been a steady stream of exploits and vulnerabilities for iOS too, and every other system for that matter.

These parental controls are generally easily bypassed, and kids usually know a lot more about this stuff than their parents.
Old hardware is available cheaply or even for free, and is more than adequate for accessing "forbidden" content, there are plenty of public wifi networks a kid could connect to.
Any "controls" that exist will be seen as a challenge for the kid to overcome.

And then you will find that lots of things become inaccessible if you are not running one of those mainstream systems. So instead of a free and open internet, you have "the microsoft network".

You can't prevent anyone from seeing such things unless you keep them locked in a windowless room. Sooner or later they are going to come into contact with such material. If this material is forbidden it will also be more attractive, so kids will share it amongst themselves at school for example.

So given that sooner or later kids are going to come across this material unless you take draconian measures to prevent it, surely its better that when they first encounter such things they do so under the guidance of responsible adults who can explain what it is.

The answers aren't important, the process of working out the answers is what should really matter.

Now yes, but before student loans became a thing only those with sufficiently affluent parents could afford to attend.

Millions of horses were raised and taken care of so they could pull carriages. Once they were replaced by motor cars most of these horses became superfluous, many were killed off and the population of horses declined massively. The much smaller number of horses still around are still not free, they are kept for racing or for the much smaller population of leisure riders. Horse racing in particular can be brutal, with selective breeding, serious injuries etc.

The key letter in UFO is U. Whoever saw it or took a picture of it etc cannot identify it and doesn't know what it is.
That's also why you only see blurry photos, because if the photo was clear it wouldn't be an unidentified object anymore - you would know what it was, it would be something benign and the case would be closed.

If you're going to put a backdoor in something, you'd always want deniability so of course you'd make it look like a bug.

In terms of lowest budget, they don't actually have to develop any firmware at all for a lot of devices. There is already open source firmware like OpenWRT which they could ship. This would both save them money and provide a better experience for users.

Sure it's rare in a datacenter, but i was thinking more of things like laptops which will be moved frequently and possibly subjected to shocks. Something that's in a socket is more likely to work loose.
Also someone deploying servers in a datacenter is more likely to know what they're doing, a random consumer buying a laptop is likely to only consider the price and nothing else.

You mean "they use memory which isn't rate for the clock rate you're applying to it", and yes, that can absolutely happen. It does not "drag the system down". A synchronous DRAM is either stable at a certain clock rate, or it is not. Physically, this comes down to binned silicon.

The memory will be rated for a particular clock, and will usually report that rate via SPD.
Depending on where you put that memory, it may operate the memory at the clock it supports instead of what the system would prefer (thus reduced performance), or might force the memory to run at a higher clock (thus overclocking it and potentially overheating/instability). Some systems will let you adjust memory clocks/timings in the BIOS, some will not.
You also have shady memory vendors where the SPD can be reprogrammed to report a higher clock.

But yes, with replaceable parts you have potentially infinite variables to consider. With fixed parts you take those variables away and have a single known good configuration.

You can run some random binary apps from the 90s on either platform, there are many more which won't work. There are plenty of games designed for win9x which don't work on current versions for instance.
With Linux it's less common because most software is open source and can be recompiled.

There's also two classes of problems that come up with running old binaries:

1) missing userland dependencies - where the binary is dynamically linked and depends on libraries which were present on older systems but are not present on current ones - this is what breaks most older linux binaries as there has been quite a lot of churn over the years. On the other hand it's relatively easy to fix by providing the userland libraries it wants. This happens on both platforms, but is a lot more common on linux because windows still ships with a huge set of old libraries including ones which are very rarely needed.
2) incompatibilities at the system level - where even with all the required libraries, the binary is unable to run. this actually happens more often on windows, linux has been very good at maintaining userland compatibility over the years.