Mobile data capacity is bounded by information theory, not by engineering ambition. The Shannon entropy limit for the kind of channel a cellular radio uses sits at roughly 6 bits/second per Hz. Current 4G radios already reach about 5 bps/Hz, which leaves little technological headroom, and the allotted frequency space for wireless data caps a 4G link at around 50Mbps.
What reallocation would buy
The remaining lever is the spectrum itself: which slices are assigned to which service. If every frequency currently used for broadcast television were reassigned to wireless data, and the channel achieved roughly 20dB signal-to-noise, the theoretical throughput would be about 1.2Gbps — shared among all devices talking to that tower.
That is only marginally more than a desktop's wired connection to a home network, and real-world conditions would erode it further through multipath, interference, and zone overlap.
Geography and link direction
Frequency alone doesn't determine what a user experiences. Terrestrial TV operates over hundreds of miles, but a handheld device returning a transmission across that distance would be essentially infeasible. Shrink the coverage zones instead, and the arithmetic changes: at one square mile per zone with one in a hundred people on a phone at any moment, San Francisco would work out to roughly 7.4Mbps per user.
Shrinking further to 250x250-meter zones supports 300Mbps. An alternative, at the cost of latency, is to lean on extremely low-power mesh networks.
A limit, not a plateau
The unfamiliar part is the direction of travel. Transmission technology has improved in a long unbroken line — radio, black and white TV, color TV, HDTV — and this is the first point at which fundamental physical limits, rather than ingenuity, set the ceiling. The consolation is that 300Mbps is enough for anyone. :)
