G.fast uses 106 MHz compared to only 1 MHz for ADSL. More MHz means more capacity but only works over very short loops. The actual ITU standard is 300 pages with dozens (?hundreds) of features which have relatively little impact on the speed. Traditional VDSL used 17 MHz for speeds around 100 megabits. G.fast can use 106 MHz for speeds typically 400-700 down, 100 megabits up. (The "gigabit" is pr fluff today although upgrades to actually get a gigabit are in the works.)
The distance limit: High frequencies used by G.fast don't go very far in ordinary telco twisted pair cables. It was designed for 50-150 meters and speeds drop rapidly after that. British Telecom and AT&T engineers told the ITU standards group that short reach was fine and G.fast nodes would be 8-16 homes. That set the standard.
G,halffast When British Telecom realized the cost of running fiber to 4M "distribution points/curbs" they sent the entire plan back to the drawing board. The maximum wire length was increased from 250 meters to 500 meters. Chip makers are scrambling to modify their chips to the new demand. Many English homes will get speeds of 100-400 down, not close to the promised gigabit but pretty darn good.
How many homes get 400-700 megabits down and how many don't get 100 megabits depends on how much money BT decides to spend on the network. The odds are BT will keep costs down, leaving many customers with far lower speeds than the pr is suggesting. AT&T customers will probably get the 400-700 megabits down. Most of their deployment will be fiber to the basement and G.fast to an apartment. That's often less than 100 meters and the high speeds are possible. From now to at least 2019, AT&T plans to "fiber" ~15% of their homes. Most will be apartments.
"The Brits love their gardens," I'm told. They have very few apartment buildings, so BT can't use AT&T's strategy. BT is also planning to do far more than AT&T's 15%, going quickly to "nearly all the country." They will mostly have to deal with private homes.
Yes, the gigabit is possible in a few years. Rami Verbin of Sckipio is confident a gigabit is practical. Based on testing the current generation of chips, he can increase the "constellation," loading more bits per hertz. "Non-linear precoding" should add more bandwidth as well, making the system more efficient. It increases the complexity of the chip but Moore's Law is bringing down the cost. Paul Spruyt of Alcatel has a similar expectation and these changes are being actively considered in standards.
There's lots more to know if you're building a network, but this should get the non-expert started.