The Anatomy of a Route: Lit, Dark, Diverse, and Fragile
To treat fiber as a first-class variable you need the vocabulary of the people who build and sell it. This lesson lays down the structural terms the rest of the course assumes, because the analysis is only as sharp as the distinctions underneath it.
Begin with the physical asset. A long-haul fiber route is a bundle of glass strands buried in conduit, typically following a right-of-way granted by a railroad, a pipeline, a highway authority, or an electric utility. The right-of-way is the scarce, durable thing. Glass is cheap; the legal permission to bury it across hundreds of kilometers of other people's land, negotiated once and held for decades, is not. This is why the carriers that matter are, at root, owners of rights-of-way and conduit, and why their networks trace the old railroad and pipeline corridors of the twentieth century. You cannot easily build a second route beside an existing one, because the permits and construction cost are prohibitive, and that non-repeatability is the source of every scarcity in this market.
Within a route, distinguish two products. Lit fiber is capacity: the carrier owns the glass and the optical equipment that pushes light through it, and sells you a service measured in bits per second, say a 400-gigabit-per-second wavelength between two cities. You buy transport and the carrier operates it. Dark fiber is the strand itself: the carrier leases you unlit glass, and you supply and operate your own optical equipment on each end. Lit is a service; dark is infrastructure. The next module examines the strategic trade-off between them in depth. For now, hold the distinction, because it recurs everywhere: buying capacity is not the same as controlling a physical path, and control is the currency of resilience.
The second structural concept is diversity, and it is the one analysts most often get wrong. Two fiber paths are diverse only if they share no common point of failure: no common conduit, no common bridge crossing, no common building entrance, no common amplifier hut. Carriers sell redundancy that is not truly diverse all the time. Two circuits sold as separate can ride the same physical conduit for a stretch, so a single backhoe severs both. The industry term for the audit that verifies genuine separation is a route diversity or path-separation study, and for an AI campus it is as load-bearing as a power redundancy analysis. A site with two carriers whose fibers converge at one river crossing has, for that crossing, no diversity at all, only the appearance of it on a marketing map.
Third, understand the failure statistics, because they justify the paranoia. Fiber cuts are not rare tail events; they are routine operational facts. Across large networks, long-haul routes experience accidental cuts on a recurring basis, most commonly from construction and excavation, the proverbial backhoe, along with agriculture, storms, and vehicle strikes on aerial spans. A cut typically takes hours to locate and splice. For a campus running distributed training, a multi-hour interruption on the only path between sites is not a degraded experience; it is a halted job and idle capital measured in the millions. This is why single-trunk exposure is a genuine underwriting flaw and not a theoretical one, and why the certainty of cuts, not their rarity, should drive the design.
Fourth, the meet-me point. Fiber routes do not connect to campuses by magic; they terminate at carrier hotels and interconnection facilities where networks physically cross-connect. A site's usefulness depends not only on the long-haul routes that pass near it but on its access to a rich interconnection point where many networks meet. A campus adjacent to a major interconnection hub inherits optionality: many carriers, many routes, competitive pricing, easy capacity adds. A campus far from any such hub may have power in abundance and still be, in network terms, remote, forced to pay a terrestrial toll on every packet before it reaches anywhere that matters.
Assemble these into an evaluation frame you will use for the rest of the course. For any site, ask five questions. How many physically diverse long-haul routes serve it? Who owns each, and are they independent parties or the same carrier wearing two hats? Are the routes truly separate, verified by a path-separation study, or do they converge at a chokepoint? What is the fiber-kilometer distance, and resulting latency, to the target user population and to any partner campuses? And how far is the nearest rich interconnection point? These five questions convert the vague instinct that fiber matters into a structured, comparable assessment that survives contact with a real site.
My read: most site scorecards in circulation collapse all of this into a single yes-or-no field labeled fiber available. That field is nearly worthless. Fiber is available almost everywhere in some degraded sense; the questions that determine deliverability are diversity, ownership independence, distance, and interconnection richness. An analyst who replaces the binary with the five-question frame is pricing a risk the rest of the market is rounding to zero. The vocabulary in this lesson, lit versus dark, right-of-way, true diversity, path-separation study, meet-me point, is the toolkit for that repricing, and every later module builds directly on it.