Shannon bound. But it's very large. I don't think we're anywhere close with current DWDM emitter/detector technology.

Actually we know that a single mode fibre (there would typically quite a lot of them in a cable) can carry around 100 Tb/s in the C band (used by most systems due to amplifier availability) over about 100km. Research systems have reached that limit and commercial systems are not very far off.

Is that right? The C band is only 4 or 5 THz wide, so that's impressive packing. (I'm way out of date, I know there is QAM and whatever.)

For the super high capacity demonstrations, 256 QAM and/or probabalistic/geometric shaping is typically used so we get to about 12 bit/s/Hz (accounting for FEC and pilot overheads). Interestingly, data rates are mainly limited by the transceivers (RF amplifiers, DAC/ADC ENOB... is not that great at 25-100GHz, which is required for the 50+Gbaud symbol rates).

Modern DWDM systems use a channel spacing of 75/100 GHz, so you easily fit more than 50 channel in a single fibre.

OS2 single mode fibre is pretty future proof. The transceivers may change, but the underlying cable should last a looong time and can be sliced and diced considerably with WDM (16+ channels AFAIK).