they dont
ALL of those devices are based on the same RTL8196C Realtek SoC. This is one stupid/corrupted/incompetent Realtek employee/contractor putting a backdoor in SDK reference codebase (access under NDA, Realtek is famous for being hostile towards opensource just like Broadcom), and no one else bothering to even read sources before shipping devices.
On the other hand, you can easily find datasheets for many of their SoCs. Whether or not this is due to their lax security policies is a question to consider too...
Realtek might not be agreeable to opensource, but I'd say they are still a far distance away from Broadcom. They are more of a "Gongkai"[1] sort of company.
- Hanlon's Razor [1]
Hanlon's Razor, like Occam's Razor, is most useful in cases where we there isn't any other evidence that points to a specific answer.
In the case of internet security, we do have a significant amount of evidence that several national spy agencies have been corrupting security of networking hardware and software. With projects like BULLRUN and the RSA/Dual_EC_DRBG scandal, we even know they went as far as pay businesses to do exactly this kind of weakening of their product's security features. This is especially true for this kind of low-end, discount product; a business with low profit margins is probably easier to bribe.
All else is not equal, so it is reasonable assign a high probability to spy agency involvement. This isn't proof, of course, but neither is Hanlon's Razor.
This requires determined effort to ignore the level of sophistication on evidence in the same reports of nation-state level attacks which you cite. Everyone who's read them noted a consistently high level of attention paid to making sure that attacks were both targeted and concealed.
Using “super” as both a username/password for a million devices is the kind of thing you'd expect from incompetence; with the NSA I'd expect something like what they actually did with Dual_EC_DRBG, where the device would show every sign of appearing to be secure but someone with knowledge of a particular constant could save significant effort when cracking the crypto. The last thing they want is for you to notice & patch the hole because some random spammer started exploiting it.
The one remotely plausible explanation would be something like this being a way for a mole to spin a backdoor as an innocent mistake. Even that's stretching it a bit because this is so easy to find that it'd have a pretty high risk of exposure – if you're going to invest that kind of time, a more subtle bug is only a minor increase in difficulty.
If you're going to be Bayesian about it, use Hanlon's Razor as your prior, then update that. In other words, you should default to that, unless you have very good reasons.
That they were all using the same or a variation of the same firmware shows that the problem isn't maliciousness but carelessness. Someone writes a reference implementation that happens to have a hard-coded username as a debugging option. It may even have come with a note warning you that this needed to be changed before being put in the final product. But in the rush to a product to market as fast and as cheap as possible, they forego any modification to the firmware beyond branding and push it out bugs, backdoors, and all.
I think it's a stretch to say these are "major" vendors. I've only seen TrendNet in the wild and usually because someone is complaining about their internet not working well and I say, "Yeah, that's not a good router. You should buy another one." You get what you pay for.
Of course if a truly malicious actor wanted to place a backdoor into a router on purpose, I expect it'd be someplace other than the flash firmware that can be easily overwritten.