Glacier, Nearline or B2 are "cloud" services as well, and while they're more stable than ordinary-consumer-grade services (like Google Drive or OneDrive), the it's not like their future is set in stone (affordability, service policies, etc)
I believe the way to get the best reliability is to keep multiple replicas of the archive (git-annex's great here) over storage options from as many different vendors as possible, and hope the chance of everyone going down at nearly the same time is too low. And probably schedule some periodic fully-automated checks to minimize the chance some service had degraded and had silently lost the data.
In that respect my recommendation is to buy some of these massive Seagate archive disks. I don't know how durable they are but given you can have 8TB for less than $200, I would have one or two in independent places. But I would encrypt the data though. But then where do you store long term the key (brain is a bad choice!)?
And as technologies evolve, in 5y there will probably be cheap 30TB HDD, so it will make sense to consolidated, etc
Be careful. There's a tendency that if one drive fails, the other will follow very soon. Especially if the drives are from the same batch (I was bitten by this), but not necessarily. I'm not expert on the topic, but I believe that when planning for long-term HDD reliability, buying drives from multiple different vendors is a must.
I guess it's a good thing that Western Digital and HGST (which should now be thought of as a WD variant) have introduced their own 8TB drives too. Seagate also has "Surveillance" and "Enterprise" variants of 8TB size.
Laser-cut the key represented as a QR code, paired with classic textual hexadecimal (not even base64, to not have to worry about "o"s and zeroes) representation. Problem solved.
I believe the way to get the best reliability is to keep multiple replicas of the archive (git-annex's great here) over storage options from as many different vendors as possible, and hope the chance of everyone going down at nearly the same time is too low. And probably schedule some periodic fully-automated checks to minimize the chance some service had degraded and had silently lost the data.