Latency Edges, Era by Era

Ask what "fast" meant to a trader in 1900, 1990, and 2015 and you get three different answers, but the same underlying trade: know something a moment before everyone else, and act on it first. The technology changed every decade — telegraph, leased line, microwave, colocated servers feet from a matching engine — but each generation followed the same arc: a real edge, a window where only a few firms had it, then competition or regulation grinding it down to the cost of staying in the race.

The early era: information beats a wire

Before electronic markets, the edge was literally how fast news physically traveled. Nathan Rothschild is (apocryphally, but instructively) said to have profited on London bonds by learning of Wellington's victory at Waterloo before the official messenger arrived. True or not, the mechanism recurred for a century: private couriers, carrier pigeons, and later dedicated telegraph and telephone lines all monetized the same gap — knowing a price-moving fact before the public did, by minutes or hours.

Mid-century to 1990s: the wire itself is the edge

As markets electrified, latency stopped being about information content and became about the wire itself. Firms paid for dedicated leased telephone lines because a price update reaching you in three seconds instead of ten was worth trading on with zero new information — just reacting to a public price change faster. This era rewarded better infrastructure, not smarter models.

2000s: colocation and the microsecond

Electronic matching engines made latency measurable, and the game inverted: instead of a person deciding fast, a machine placed the order, and the bottleneck became physics — how far does the order have to travel? Exchanges began renting colocation space, letting firms put servers in the same data center as the matching engine for a fee. A colocated firm might react to a price change in tens of microseconds; one trading from across the country might take milliseconds — a thousand times slower.

Worked example. Light in fiber travels at roughly two-thirds the speed of light in vacuum, about 200,000 km/s. A round trip over a 1,000 km fiber route between Chicago and New Jersey takes roughly $2 \times 1000 / 200000 = 0.01$ seconds, or 10 milliseconds, from cable transmission alone, before any processing time. A firm colocated at the exchange has near-zero transmission delay for that leg. If a price-moving order arrives in Chicago and a colocated New Jersey firm reacts before a Chicago-based firm's order even finishes crossing the fiber, the colocated firm trades against stale quotes with near certainty — not by predicting anything, but by being structurally first.

2010s: microwave and the speed of light itself

Fiber is not the fastest path — microwave and, later, laser links through the air travel closer to true light speed than a fiber cable does, and firms spent hundreds of millions of dollars building microwave towers between Chicago and New Jersey to shave milliseconds, then microseconds, off transmission time. This is the era Michael Lewis's Flash Boys documented: firms racing to build the straightest line-of-sight path, because a few extra miles meant a few lost microseconds, and that was the entire trade.

What killed each generation

• Telegraph and phone edges ended when the infrastructure became commodity — once every serious firm had a leased line, none had an edge from having one.
• Colocation edges compressed as exchanges sold colocation slots to any firm willing to pay, turning a scarce advantage into a standard cost of participation.
• Microwave edges hit a physical wall: line-of-sight and the speed of light itself bound how much faster any new link can be.
• Regulation pushed back directly. Some exchanges introduced speed bumps — small deliberate delays on incoming orders — to blunt pure latency races, on the argument (Budish, Cramton and Shim) that continuous-time trading turns every price change into a winner-take-all speed race with no offsetting market benefit, and that batching trades into short discrete auctions removes the incentive to race at all.

In interviews

Trace the arc: each generation's edge came from a real, measurable speed advantage, and each decayed the same way — the technology became available to competitors, or an exchange changed market structure to remove the incentive. Speed edges are rentable (colocation, private lines) and therefore self-liquidating once enough participants can afford the rent, unlike an edge based on proprietary insight.