In Engineering you have two different kinds of failures:

The first is to do with manufacturing flaws and happens in the first couple of months of use, hence how Warranties work - bad part of bad assembly so it breaks on first use or soon after.

The second kind is the device dying from decay due to use, from old age if you will.

Survivor bias, IMHO, only applies for those devices that last beyond the stage were the first kind of failure can happen as it’s kinda random (you can reduce the proportion of devices that fail, but for any one device it’s random if it will be one that fails or not)

So a 3 year old fridge dying is not from manufacturing defects but it’s dying from faster ageing, which is a flaw in the design or a choice of cheaper, lower quality components.

From what I’ve seen that’s exactly what’s been happenning: less robust designs and cheaper components with shorter lifespans, all to save on raw material costs.

Lower manufacturing quality tends to cause the first kind of failures, not the failures well past the first few months.

PS: Note that dying from the second kind of failure still has a random probability for any one device, though whilst the probability from dying from manufacturing flaws is very time dependent (starting very high and then tailing off to pretty much zero within some months), the probability of dying from age is a lot less time dependent and if that much increases slightly with increasing age (whilst the other kind decreases steeply with age, specifically decreases steeply with use). I’m mentioning this for completness, as the point still stands - if there is a high proportion of devices of a given type dying at year 3, then that design has a much higher rate of failure due to aging than devices for which a much smaller proportion dies at year 3, hence the design is not robust and/or lower quality components are being used.