A column about curious stories from the worlds of science, technology, and cinema. Today—a story of how a Habr popular-science post suddenly turned out to be the key to the most expensive networking race on the planet. lvvovich’s article “Slow Light” debunks a popular myth: that copper is “faster than glass,” and that pings are explained by “the physics of optics.” In reality, electrons in a wire drift at fractions of a millimeter per second, while energy is carried by an electromagnetic wave at 0.6–0.7c—and in that sense, copper and glass are “kids from the same school.” The real delay lives elsewhere: in the crooked geography of routes, microseconds at routers, queues in buffers, and TCP handshakes. The article’s verdict: “bandwidth is about volume, latency is about path,” and CDNs with caches can deliver gains of tens of times over a medium that is only one and a half times better. A fair conclusion—for the ordinary internet. But there is one market where both levers—path and medium alike—have been pushed to the limit for fifteen years. These are the financial markets between Chicago and New Jersey, and there the myth of “slow light” wins billions.
At the start of 2014, Michael Lewis published Flash Boys, and all of America learned the word “front-running” in relation to fiber optics. The book is noisy, dramatized in places, but at its core lies a straightforward engineering fact: between Chicago, where CME futures trade, and New Jersey, home to the stock exchanges, lie about 1200 kilometers. And the winner in this setup is whoever can get their bits there faster.
Light in a vacuum travels at 299 792 kilometers per second. Remember that number—it will serve as the judge at the end of this story.
The refractive index of fused silica is n ≈ 1.47: light trudges through fiber at roughly 204 000 km/s. A radio wave in the air travels at practically c, and its route is almost straight. That gives optics a double tax. Medium: 1180 km along the geodesic at the speed of glass—that is 5.8 ms one way. Geography: you cannot lay cable in a straight line—it follows railways and rights-of-way, and any real fiber route is 10–15% longer than the geodesic. So an honest Chicago–New Jersey fiber route is ~6.6 ms one way, ~13 ms round trip. Radio: ~4.0 ms one way, ~7.9 ms round trip. The difference is about five milliseconds round trip. Sounds trivial: a blink takes twenty to thirty times longer. You will soon see how much money that is.
Dan Spivey told this story in Forbes in 2012: in 2007, a New York hedge fund hired him to devise an arbitrage strategy between Chicago futures and New York stocks. He devised the strategy—and hit a wall: to execute it, he needed quotes at two points with minimal latency, and the fastest line on the market did not exist. Spivey did not go trade. He went digging. Literally: building an optical cable along the geodesic between Chicago and Carteret, New Jersey, where Nasdaq sits. The money came from James Barksdale—yes, that one, the former CEO of Netscape. The project proceeded under strict secrecy: competitors could not be allowed to learn the route, because the route itself was the product. 827 miles of cable, 300 million dollars, launch—June 2010.
Notice the irony; it will fire again later. Spivey built virtuoso fiber—13.1 ms round trip, the benchmark for its class. But that figure was doomed from the moment it launched, because a technology without the glass tax was maturing in parallel. The irony reached its finale in 2017: Zayo bought Spread Networks for 127 million—more than twice cheaper than the construction had cost. The fiber race ended before it began: flawless fiber lost not to rival fiber, but to a radio wave.
The physics of the microwave race is embarrassingly simple. A point-to-point radio link requires line of sight: tower must see tower. Earth’s curvature strictly limits hop lengths to 50–80 kilometers, and antennas must be raised on tall masts: trees, which are 90% water, eat the signal—the same physics that heats soup in a microwave oven. The Chicago–New Jersey route is a chain of 20–30 hops, and every hop means a separate FCC frequency license, a separate tower, a separate lease for a piece of sky.
Researchers took the reconstruction of this web seriously: Bhattacherjee, Aqeel, Laughlin, Maggs, and Singla published a study at ACM IMC 2020 with the precise title “A Bird’s Eye View of the World’s Fastest Networks.” Their methodology is a beautiful example of extracting a commercial secret from public data: every radio-frequency license is filed with the FCC, and the FCC makes it all publicly available. Download the licenses, and you can reconstruct the map. Of 57 candidates, 29 serious licensees remained; as of April 2020, nine of them operated live end-to-end networks.
The table in the paper reads like a Formula 1 race report. Leader—New Line Networks: 3.9617 ms one way between CME in Aurora and Equinix NY4. Silver—Pierce Broadband: 3.9621. The gap between gold and silver is 0.4 microseconds. Four-tenths of a millionth of a second. That is what they wage wars over—for towers.
The second part of the paper is a chronicle of evolution from 2013 to 2020. The best result on the CME–NY4 route improved from 4.00 to 3.962 ms, and then came the physical floor: 3.955 ms—a signal through air, with the minimum unavoidable bends. Over eight years, the race narrowed the gap to the limit to seven microseconds—and got stuck there. And there is a second observation: networks die in this race. National Tower Company aggressively bought up licenses in 2013, finished building its route in 2014–2015... then in 2017–2018 canceled 71 licenses, vanishing from the ecosystem. In its place came Pierce Broadband, which only announced itself in 2020 and immediately reached the podium. Eight years of play, fixed endpoints, a finite limit—and the balance of power is still shifting.
Now for the main question: why? What is a microsecond worth, if people are willing to spend 300 million dollars on it?
The academic answer was calculated by Matteo Aquilina (FCA), Eric Budish (Chicago Booth), and Peter O’Neill in “Quantifying the High-Frequency Trading ‘Arms Race’” (FCA Occasional Paper 50, January 2020; later—Quarterly Journal of Economics, 2021). Their methodological insight is elegant: instead of limit-order “books,” where only the executed outcome is visible, they used exchange messages, where failed attempts are visible too. A race is when several participants simultaneously throw in an order on the same price change: the winner is visible, and so are the losers. You cannot see the losers in the book, so you cannot measure the race from the book. Results from the message logs of twelve venues across two exchange groups:
There is your answer for why they buy up hilltops in Pennsylvania. One Xeon clock cycle is ~0.3 nanoseconds; a race lives for 5–10 microseconds. A trader who is one microsecond late—those same two hundred meters of fiber—is not in the game at all: winner takes all, the first one takes the pot, everyone else paid for the infrastructure. In 2017, Bloomberg documented an absurd subgenre: an unknown company bought a strip of land right next to the CME facility in Aurora in order to run the final stretch by radio beam rather than fiber. Sniper in Mahwah, chronicler of the corridor, calculated it: 400 feet of route savings plus a faster medium yields about one microsecond—roughly 3000 instructions on a modern Xeon. A competitor without that land processes the same stream of quotes with a budget 3000 instructions smaller. That is it—you are out of the game.
By the start of the 2020s, the radio race had run into 3.955 ms, and the arena shifted to a place the early investors had never considered: inside data centers. In December 2020, the WSJ described how HFT firms were replacing ordinary fiber with hollow-core fiber—“hollow” fiber, with air rather than glass inside: dozens of parallel air-filled channels thinner than a hair. Light travels ~46% faster in air than in glass; over the short stretch from the exchange floor to a neighboring radio tower, that means several hundred nanoseconds. Flash Boys fought for milliseconds, then they fought for microseconds—now they fight for billionths.
And here a fact came to light that exploded into scandal in February 2025. McKay Brothers—the very company whose long-haul microwave wholesale service had once “leveled the playing field” for anyone interested—discovered that Nasdaq was secretly offering select clients an upgrade of their colocation fiber to hollow-core for $10 000 per month, and sent the SEC a letter using language along the lines of “a continuation of the practice of secretly providing advantages to certain participants.” The catharsis came quickly: Nasdaq told the Financial Times that, “in close consultation with the regulator and clients,” it had begun winding down the service. The story came full circle: a company that once undermined private advantages is now fighting to stop advantages from being handed out in secret. Joe Saluzzi of Themis Trading quipped that the parties brawling were “the arms dealers of the speed race” on both sides, but he captured the essence exactly: fifteen years ago, people argued over whether latency arbitrage even existed; today, they are fighting over tens of nanoseconds in a cable as long as a parking lot.
The Habr article ends with the right conclusion: “bandwidth is about volume, latency is about path”—and for the entire internet, that is true. But there is one corner of the economy where both levers have been turned all the way up. Look at the finishing chain, where every step is a separate industry:
The curve is a normal exponential; saturation has been reached: both path and medium have been conquered. The one thing that cannot be optimized is the topology of the market: the truth about price is smeared across two sides of the country, and as long as the market is arranged this way, it will be profitable for someone to pay to have their bits fly 0.4 microseconds faster than yours.
Light does not care. It travels through a vacuum at 299 792 kilometers per second, and that is the only constant that cannot be optimized. Everything else—those same eight milliseconds between Chicago and New Jersey—has already been divided up.