Dark Fiber
Who Lights the Fiber, and Why It Matters
The defining feature of dark fiber is that responsibility for the optical layer transfers to the lessee. A lit service, by contrast, delivers a finished Ethernet or wavelength handoff: the provider owns the transponders, the channel plan, and the capacity tier, and the customer simply plugs in. With dark fiber the provider's deliverable stops at the glass. They guarantee physical continuity of the strands along a route and a maximum end-to-end optical loss, but everything that converts bits to photons belongs to the customer. This is why dark fiber is attractive to operators who want to engineer their own network and unattractive to those who just want bandwidth without owning optical hardware.
Because the lessee controls the line system, the capacity of a dark fiber pair is bounded only by the physics of the glass and the transceivers attached to it, not by a service tier. A modern DWDM system packs 80 or more channels onto the C-band at 50 GHz or finer spacing; with 400G or 800G coherent optics per channel, one fiber pair can carry tens of terabits per second. Upgrading capacity means swapping or adding transceivers on equipment the operator already owns, with no renegotiation. The same control delivers deterministic latency: because the path is a fixed run of glass with no intermediate switching, propagation delay is just the route length divided by the group velocity in fiber, about 4.9 microseconds per kilometer.
The cost-control side is equally important. Dark fiber is leased either as an indefeasible right of use (IRU), a long-term prepaid grant treated as a capital asset, or as a recurring monthly lease counted as operating expense. Carriers and data-center operators building long-haul or metro backbone routes with rising bandwidth demand usually favor an IRU because the per-bit cost falls every time they add wavelengths to fiber they already paid to use.
Span Engineering and the Loss Budget
Before lighting a leased route, the operator engineers an optical loss budget for each span. The total loss is the fiber attenuation along the route plus the loss of every fusion splice and connector. At 1550 nm, standard G.652 single-mode fiber attenuates about 0.20 to 0.25 dB/km, splices add 0.05 to 0.10 dB each, and connector pairs add 0.3 to 0.5 dB. The transceiver link budget, transmit power minus receiver sensitivity, must exceed this loss plus an aging and repair margin of roughly 3 dB. Spans longer than the transceiver reach are extended with erbium-doped fiber amplifiers or Raman amplification rather than electrical regeneration, preserving the all-optical, low-latency path.
Optical Loss Budget Equations
Ltotal = (α × D) + (Nsplice × Lsplice) + (Nconn × Lconn) dB
Link margin (must be ≥ 0):
M = (Ptx − Srx) − Ltotal − Maging dB
Fiber propagation delay:
td = D × n / c ≈ D × 4.9 μs/km (n ≈ 1.468)
Where α = attenuation (dB/km), D = distance (km), Ptx = launch power (dBm), Srx = receiver sensitivity (dBm), Maging ≈ 3 dB. Example: 80 km × 0.22 dB/km (17.6 dB) + 20 splices × 0.08 dB (1.6 dB) + 2 connector pairs × 0.4 dB (0.8 dB) ≈ 20.0 dB, well within a 30 dB coherent budget.
Dark Fiber vs. Lit Service Options
| Attribute | Dark Fiber (leased) | Lit Wavelength | Lit Ethernet (E-Line) |
|---|---|---|---|
| Who owns optics | Lessee | Provider | Provider |
| Max capacity / pair | Tens of Tbps (own DWDM) | Per purchased λ (10G to 800G) | Per tier (1G to 100G) |
| Latency control | Full, deterministic | Provider-dependent | Provider-dependent |
| Upgrade path | Swap own transceivers | Buy more λ | Order higher tier |
| Cost model | IRU (capex) or monthly | Monthly per λ | Monthly per port |
| Best fit | Carriers, hyperscale DCI, fronthaul | Mid-scale point-to-point | Enterprise WAN |
Frequently Asked Questions
How do you calculate the optical loss budget for a dark fiber span?
Sum the fiber attenuation, splice loss, and connector loss. At 1550 nm, G.652 fiber is about 0.20 to 0.25 dB/km, splices add 0.05 to 0.10 dB each, and connector pairs add 0.3 to 0.5 dB. An 80 km span with a splice every 4 km (20 splices) and a connector pair at each end totals roughly 20.0 dB (17.6 + 1.6 + 0.8). The transceiver link budget must exceed that loss plus a 3 dB aging margin, so a 30 dB coherent module covers 80 km comfortably; longer spans add EDFA amplification.
What is the difference between an IRU and a monthly dark fiber lease?
An indefeasible right of use is a 15 to 25 year prepaid grant of exclusive use of specific strands, paid as a large up-front capital sum plus an annual operations fee, and treated as a capital asset. A monthly lease is an operating expense, a recurring per-strand, per-route-mile fee on a 1 to 5 year term where the provider keeps ownership and maintenance. Both deliver the same unlit glass; only the commercial and accounting structure differs.
Why would an operator lease dark fiber instead of buying lit wavelength services?
Dark fiber gives full control of the optical layer. The lessee picks the modulation, channel plan, and DWDM line system, scaling one pair from a few 100G channels to 80 or more wavelengths at 400G or 800G without renegotiating capacity. Latency is deterministic because the path is fixed glass with no intermediate electronics, which matters for 5G fronthaul and trading links. The trade-off is that the operator must engineer the line system and amplifiers themselves.