#THREE MILLISECONDS, THREE HUNDRED MILLION DOLLARS
In 2009 and 2010, a company called Spread Networks spent roughly $300 million laying an unusually straight fiber-optic line between Chicago, home of the CME, and Carteret, New Jersey, home of Nasdaq's data center. The payoff was a round-trip latency cut from about 16 milliseconds to about 13: three milliseconds, purchased at a rate of roughly $100 million per millisecond. It worked, for a while.
Within about two years the advantage was gone, not because the cable degraded but because someone built microwave relay towers along the same corridor. Microwaves travel through air faster than light travels through glass, and the new route beat Spread Networks' by another 4.5 milliseconds, a shift Michael Lewis chronicled in Flash Boys. Spread Networks sold the line to Zayo Group in 2017 for $127 million, about 40 cents on the dollar. That isn't a story about one firm making a bad bet. It's a story about what a market does to capital when speed is the only lever available: it burns it, on schedule, indefinitely.
#THE ARBITRAGE WAS NEVER GOING AWAY
In 2015, economists Eric Budish, Peter Cramton, and John Shim published "The High-Frequency Trading Arms Race" in the Quarterly Journal of Economics, and it's the paper that actually explains why the Spread Networks story had to repeat itself. Using millisecond-level exchange data, they showed that continuous limit order books generate what they call mechanical arbitrage opportunities: price discrepancies that exist even when the underlying information is entirely public and symmetric, purely because prices update at different venues at slightly different instants. The interesting finding wasn't that these opportunities existed; the industry already knew that. It's that competition among traders never shrank them. It only raised the speed required to capture them. Firms weren't racing each other down toward zero profit; they were racing toward a fixed prize that the exchange's own matching rule kept regenerating, millisecond after millisecond, no matter who got faster.
#AN ORDER TYPE THAT MAKES SPEED WORTHLESS
Budish, Cramton, and Shim's proposed fix is frequent batch auctions: instead of matching orders continuously and granting priority to whoever arrives first, collect every order that lands within a short discrete window (they model one-tenth of a second) and clear them all at a single uniform price. Arriving a microsecond ahead of a competitor buys you nothing if everyone in your window settles at the same price. Competition shifts from who's fastest to who's offering the best price, which is the kind of competition markets are actually supposed to reward.
IEX built a narrower version of the same idea rather than wait for a full redesign of market structure. It routes every incoming order through 38 miles of coiled fiber-optic cable, adding a fixed 350-microsecond delay that gives IEX's own system time to update its quoted NBBO before the order reaches the matching engine, engineered specifically to blunt the mechanical arbitrage Budish's paper describes. IEX has been candid that the delay alone isn't sufficient against newer, predictive latency-arbitrage strategies that anticipate a price move before the last quote is exhausted, which is why it layered a prediction model on top of the delay. But the principle is identical to the batch auction: change the rule that lets a nanosecond convert directly into money, and the race for that nanosecond stops paying.
The Speed Advantage That Never Lasts
Each latency-cutting investment on the Chicago-New York route (the $300M Spread Networks fiber line, then microwave relays 4.5ms faster) was overtaken within about two years, and Spread sold in 2017 for just 40% of its build cost.
#THE NUMBERS, LINED UP
Put next to each other, the sequence of numbers tells its own story better than any argument could.
- ▸Spread Networks: ~$300 million spent (2009-2010) to cut round-trip latency from ~16ms to ~13ms.
- ▸Microwave relay towers beat that same route by a further ~4.5ms within about two years, per Flash Boys.
- ▸Spread Networks sold to Zayo Group in 2017 for $127 million, roughly 40% of the original build cost.
- ▸IEX's fix: a 350-microsecond delay via 38 miles of coiled cable, approved as a national securities exchange in 2016.
- ▸Hudson River Trading's 2024 revenue: roughly $8 billion, nearly double 2023's, per Navnoor Bawa's analysis of the firm's financials.
#WHAT HRT'S OWN NUMBERS ADMIT
The clearest evidence that the arms race was never actually load-bearing to profitability comes from the firms that won it. Hudson River Trading, built on precisely this kind of speed infrastructure, pulled in roughly $8 billion in revenue in 2024. In a single quarter, HRT and Jane Street together reported about $22.5 billion in trading revenue, with HRT's profit up roughly 175% to about $4.2 billion. Those numbers don't come from anyone winning a microwave-tower fight. They come from modeling order flow and pricing risk better than whoever is on the other side of the trade.
The lesson isn't that speed stopped mattering to trading; it's that speed was never the actual asset. It was a symptom of one specific, fixable design choice: continuous matching with no order type that lets a trader commit to when, exactly, execution happens. Batch auctions remove that choice's leverage entirely; IEX's speed bump blunts it under the current architecture. Either way, the firms that used to win by being fastest now make their money the way good trading was always supposed to work: by being right about prices, not first through the door. The arms race was never proof that markets have to reward speed. It was proof that one badly designed rule can look like a law of nature for a decade, right up until someone bothers to change the rule.
Speed Race vs. Modeling Race, by Revenue
illustrativeHRT's revenue roughly doubled from an estimated $4 billion in 2023 to $8 billion in 2024, then HRT and Jane Street together posted about $22.5 billion in a single 2025 quarter, showing profit now scaling with modeling power rather than microsecond speed.