Uber operates in cities where tens of thousands of drivers can be active at once, each sending an updated location roughly every four seconds.
Calculating the distance from every rider to every one of those drivers, for every single ride request, would overwhelm Uber's servers within minutes.
Uber avoids that problem by narrowing the search before any distance calculation happens at all.
Dividing the map into hexagons
The company relies on a system called H3, which divides the Earth's surface into a grid of hexagonal cells, each smaller than one square kilometre.
Every driver and every rider falls inside one of these hexagons at any given moment, based on their current location.
When someone requests a ride, Uber only searches the rider's own hexagon and the six hexagons touching it.
That narrows the pool of drivers to check from thousands down to a few hundred, before any further calculation begins.
Why hexagons rather than squares
Uber chose hexagons over square grid cells because every hexagon has six neighbours, all of them the same distance from its centre.
A square grid does not share that property: cells diagonally adjacent to a square are further from its centre than cells directly above, below, or beside it.
That inconsistency makes squares a poorer approximation of "nearby" than hexagons, which treat every neighbouring cell equally.
Distance is not the same as time
Narrowing the search to nearby hexagons only solves part of the problem, because the closest driver in a straight line is not always the fastest one to arrive.
A driver a few hundred metres away by direct distance might be separated from the rider by a river, a motorway, or a one-way system that adds several minutes to the actual journey.
Uber's matching system has to weigh real road travel time against straight-line proximity, rather than assuming the nearest car by distance is automatically the best match.
Matching in batches, not one by one
Rather than processing each ride request the instant it arrives, Uber groups requests together and matches them in batches.
Batching allows the system to assign drivers across several riders at once, reducing the chance that two nearby riders are sent competing for the same car.
The result is a matching process that looks instantaneous to a rider tapping "request", but relies on geometry, road data, and careful sequencing to work at city scale.
