How Fast News Travels Is How Far Power Reaches

Tang China ran two parallel signalling systems. Beacon towers were astonishingly fast—one every thirty li along the frontier, smoke by day and fire by night, over a thousand li in a day—but they could say only one thing: the enemy is here. The postal relay was slower: the Tang Liudian records 1,639 stations and more than 20,000 couriers, with express runs of 500 li a day, but it could carry any content at all. An Lushan (703–757) held three military governorships at once—Fanyang, Pinglu and Hedong—commanding over 100,000 men, close to a third of the empire’s frontier army.

On 16 December 755, An Lushan raised his army at Fanyang. Emperor Xuanzong, then at the Huaqing Palace below Mount Li, learned of it roughly six days later—over two thousand li, which is precisely the express-relay rate. The system had not failed; it had run at its design limit. The failure lay elsewhere: it took the emperor six days to learn of the revolt, and it had taken An Lushan ten years to build it. Court knowledge of daily frontier affairs came from low-frequency quarterly memorials; the governors were permitted to raise their own troops, fund their own supplies and appoint their own officials precisely to cover the lag central authority could not close. The rebellion was not the result of a communications failure. It was what happened once that latency had been institutionalised.

Every distributed organisation sits on the same curve: the slower the round trip to the front line, the more decision rights must be devolved—and once devolved, monitoring degrades from live state to a monthly report. The CAP trade-off in distributed systems is the same thing.

Samuel Morse (1791–1872) was a portrait painter by trade. In 1843 he obtained $30,000 from Congress and built a forty-mile experimental line from Washington to Baltimore. The first message, sent on 24 May 1844, came from the Book of Numbers: What hath God wrought. The technology was proven. What remained to be decided was ownership.

In 1844–45 Morse and his partners offered to sell the patent to the federal government for $100,000 and fold the telegraph into the Post Office. The Postmaster General reviewed it and judged that the telegraph could not pay for itself. Congress did not take it. Every line thereafter was laid privately: from roughly 2,000 miles in 1846 to 23,000 miles by 1852, across dozens of mutually unconnected companies; in 1866 Western Union completed its consolidation and became the first nationwide monopoly in American business. Britain went the other way: the Telegraph Act of 1868 authorised purchase, the wires passed to the Post Office in February 1870, and a flat low tariff followed. The same technology handed the two countries two different societies. British service was cheaper and more widespread; America got Western Union, then AT&T—and the long-distance backbone and Bell Labs were products of that private path.

Broadband, cloud computing, payment rails—every country is replaying the same question. Declining to intervene because a technology “cannot pay for itself” carries a cost that surfaces decades later, by which time the structure is locked in.

On 5 August 1858 the two ships of Cyrus Field (1819–1892) joined their cable in mid-Atlantic. The chief electrician, Wildman Whitehouse, was a surgeon by training and believed a weak signal called for more voltage. The consultant William Thomson (1824–1907, later Lord Kelvin) argued the opposite: a long cable smears and flattens the pulse, so the answer is a more sensitive receiver, not a stronger transmitter.

On 16 August, Queen Victoria’s ninety-eight-word greeting to President Buchanan took about sixteen hours to send and receive. To speed things up Whitehouse pushed the voltage to some 2,000 volts and punched through the insulation; by 1 September the cable was dead, after barely three weeks of service. The 1861 joint inquiry by Parliament and the Board of Trade laid the blame on high voltage and rushed testing—and produced something more consequential: the British Association began defining reproducible units of resistance and voltage. The ohm as we know it came out of the clean-up after this failure. On 27 July 1866 the Great Eastern succeeded on the first attempt with an improved cable and Thomson’s mirror galvanometer, which deflected a tiny mirror with a faint current and threw a spot of light onto a scale. A message from London to New York fell from ten days to a few minutes.

Over ninety per cent of intercontinental data still travels by submarine cable, but ownership is shifting from telecom consortia to tech firms laying their own. Control moves from states to companies; the rule that whoever holds the physical link holds the traffic does not.

Claude Shannon (1916–2001) had already wired Boolean algebra to relay circuits in his 1937 master’s thesis, and spent the war on cryptography at Bell Labs. In July and October 1948 the Bell System Technical Journal published, in two parts, “A Mathematical Theory of Communication”.

The paper opens with a cut: the semantic aspects of communication are irrelevant to the engineering problem. Engineering has only to reproduce a string of symbols exactly at the other end; whether it says love letter or share price is not asked. With meaning deleted, information became measurable for the first time: the unit is the bit, the measure is entropy. He then gave the channel capacity formula C = B·log₂(1 + S/N)—below C, error can be driven arbitrarily small; above it, no amount of effort helps. That line turned “can this still be improved?” into a decidable question. By contrast, Paul Otlet (1868–1944), whose Mundaneum had accumulated some fifteen million index cards by 1934, tried to organise meaning directly, and never managed to make it an engineering discipline. Shannon’s detour worked precisely because it went around that wall.

Large language models are now working back over the very layer Shannon stripped away, while standing entirely on his foundation: training a model to predict the next token is compression, and the optimal compression rate is exactly Shannon entropy. The incision that refused to ask about meaning has become the main instrument for approaching it.