Light travels through a vacuum at 299,792 kilometers per second. That number sounds like it should solve every latency problem ever. It does not. In fact, the speed of light is less a solution than a hard ceiling, and the data center industry has spent billions of dollars learning to live with it.

The Glass Slows It Down

Here is the part that surprises most people. Fiber optic cable does not carry light at the full speed of light. It carries light through glass, and glass has a refractive index of roughly 1.47. That means signal inside a fiber strand moves at about two-thirds the speed it would in open air, closer to 200,000 kilometers per second.

The practical math: one kilometer of fiber adds approximately 5 microseconds of one-way delay. A 1,000-mile route carries a minimum one-way propagation delay of about 8 milliseconds, so roughly 16 milliseconds for a round trip, and nothing changes that. No software update. No hardware upgrade. No overclocking. Distance times physics equals latency, and that equation is closed.

For most applications, 8 milliseconds is irrelevant. For financial trading systems, AI inference pipelines, and synchronization between geographically distributed data centers, it is not. Every millisecond of latency in a high-frequency trading system translates to whether your order executes before or after the other firm's order. At scale, that is the difference between profit and loss, and it is why some of the most creative infrastructure builds in history were done by people chasing fractions of a millisecond.

How Fiber Got Where It Is

The modern long-haul fiber network follows a simple economic principle: dig once, use forever, and never dig where you do not have permission. Permission is expensive and hard to get, which is why America's fiber backbone largely runs where other infrastructure already cleared the path.

Railroads were the first. Starting in 1983, MCI built the first major single-mode fiber network in the country along the Amtrak Northeast Corridor, with the roughly 240-mile Washington-to-New York system coming into full service in 1984. The right of way was already cleared, the ground was already disturbed, and the access was already granted. The pattern stuck. Southern Pacific built its own construction train in the late 1980s capable of burying up to twelve conduits five feet underground at walking speed, running alongside its tracks. That infrastructure eventually became Sprint. The logic was the same everywhere: the railroad already owns the corridor, the fiber crew just needs to follow it.

Highways came next, but slower. Federal policy did not formally invite fiber onto Interstate right-of-way until 2005, when Congress directed the Department of Transportation to study three multistate corridors for feasibility. States followed with their own accommodation policies. The result is that a large portion of the national fiber backbone runs inside green highway medians and railroad berms, invisible to everyone driving or riding above it.

The routing consequences are significant for data center placement. If your facility sits along a major rail corridor or Interstate, you probably have access to well-established fiber paths. If you sit in a secondary market, you may be two or three expensive lateral builds away from the same capacity. Location is not just about power and land. It is about where the fiber already went.

The Fiber Boys and the Secret Trench

No story captures the latency obsession better than Spread Networks.

In 2008, a trader named Dan Spivey had a simple observation: the standard fiber route from Chicago to New York took 17 milliseconds. The route was not straight, because the cable followed railroads and highways that were not straight. If someone dug a straighter line, the latency would drop. Whoever controlled that straighter line would have an advantage over every other firm using the crooked one.

Spivey raised money, recruited James Barksdale (the former CEO of Netscape) as a backer, and started digging. Workers were told to keep quiet about the project. The route cut through farmland in the dead of night, bored under rivers, and blasted through miles of solid rock in the Allegheny Mountains. The total run was 825 miles, Chicago to Carteret, New Jersey. Total cost: roughly $300 million.

By Michael Lewis's account in Flash Boys, the finished cable cut Chicago-to-New York latency from about 17 milliseconds to 13. That 17-to-13 figure comes from the book and gets repeated everywhere; independent tick-data measurements of the period found the real-world round-trip gains more modest. Either way, Spread Networks launched in 2010 and sold access to high-frequency trading firms at a significant premium, on the understood condition that clients not discuss what they were paying or why they needed it.

By the time Flash Boys came out in 2014, Spread's advantage was already gone.

The Microwave Supersedes the Trench

The reason Spread's $300 million edge evaporated is the same reason fiber runs at two-thirds the speed of light instead of the full thing: glass slows signals down, and air does not.

Microwave transmission through open air moves at roughly 99 percent the speed of light. The moment someone realized that a chain of microwave towers between Chicago and New York would beat even a perfectly straight fiber line, the arms race shifted. Researchers who measured the corridor's latency traced the shift to line-of-sight microwave networks licensed in 2011 and 2012, and estimated the combined infrastructure and five-year operating cost of the low-latency buildout at more than $500 million. Each microwave path beats fiber by several milliseconds. Spread Networks itself was eventually sold to Zayo Group for $127 million, less than half what it cost to build.

The microwave buildout has since pushed into Atlantic crossings, with firms experimenting with shortwave radio links to European markets, and into low-earth orbit, where latency over long distances can beat both fiber and microwave by routing through space rather than along the surface of the earth.

When Someone Cuts the Cable

Most people who think about fiber infrastructure think about the things running through it. Less attention goes to the infrastructure itself, which is buried underground, unmanned, and in most cases unmonitored at the physical level.

The attacks in California are the clearest illustration of the vulnerability. In April 2009, four underground fiber cuts in a single night knocked out landline, cellular, and internet service for tens of thousands of people across Santa Clara, Santa Cruz, and San Benito counties. Deliberate cuts and vandalism have caused major outages repeatedly in the years since. The scale shows up clearly in recent industry data: between June and December 2024, U.S. operators reported 5,770 incidents of theft and vandalism against communications infrastructure, an average of 27 a day, with California and Texas alone accounting for about half. That count, compiled by a coalition that includes NCTA, USTelecom, CTIA, and WIA, covers only incidents reported by participating providers, so the true national total is higher.

The most common culprit is not a sophisticated actor. It is a copper thief who did not know the conduit contained fiber instead. Fiber has no scrap value, so the theft fails, but the cut still happens and the outage is real. Emergency calls drop. Financial systems lose connectivity. Data centers that rely on those paths either fail over to backup routes or go dark until repairs are made, which in the case of buried cable can take hours to days.

Nation-state interest in submarine cables adds a different layer. The same physics that make long fiber routes valuable make them hard to defend. A cable at the bottom of the ocean has no guards. The race to protect that infrastructure, both physically and through redundant routing, is one of the quieter but more serious problems the industry is working through right now.

The Bottom Line for Data Center Siting

The speed of light matters to data centers for two reasons that compound each other. First, proximity to end users and to other data centers directly determines the latency floor your tenants and customers can achieve. No amount of engineering gets below what the distance dictates. Second, the fiber that carries traffic between facilities follows paths shaped by railroad rights-of-way, highway corridors, and a handful of very expensive private builds that were laid by people chasing milliseconds hard enough to move mountains, literally in the case of the Alleghenies.

Where the fiber runs shapes where the latency is acceptable. Where the latency is acceptable shapes where workloads land. Arizona sits at an interesting intersection of several major fiber corridors connecting the Pacific coast to the interior and the Southwest to the major Midwest and East Coast markets. That geography is not accidental. It is the outcome of every railroad right-of-way decision, every highway utility accommodation policy, and every time someone with a backhoe followed the path of least resistance toward the nearest existing trench.

Sources

  • The Washington Post, "MCI, Amtrak to Build Communications Cable" (Dec 1982): https://www.washingtonpost.com/archive/business/1982/12/02/mci-amtrak-to-build-communications-cable/30cbb09e-c020-49a4-b3eb-054d0a5bb624/
  • Laughlin, Aguirre, and Grundfest, "Information Transmission Between Financial Markets in Chicago and New York" (arXiv 1302.5966; published in The Financial Review, 2014): https://arxiv.org/abs/1302.5966
  • Business Wire, "Zayo to Acquire Spread Networks" ($127 million, 2017): https://www.businesswire.com/news/home/20171127005288/en/Zayo-Acquire-Spread-Networks
  • NCTA, USTelecom, CTIA, and WIA, "Protecting the Nation's Critical Communications Infrastructure" (5,770 incidents, June to December 2024): https://www.ncta.com/wp-content/uploads/2025/04/Vandalism_2025_Report.pdf
  • Michael Lewis, Flash Boys (W. W. Norton, 2014), for the Spread Networks account and the 17-to-13 millisecond figure.