How much fiber does your cluster need? Run the numbers

Enter the numbers you know and the page returns fiber-km, cable-km, cable assemblies and a buy-by date. Start with GPU count and fibers per link from your bill of materials, then set tiers and oversubscription from your fabric design. Open the advanced assumptions to change run lengths and buffers. Every default is an illustrative placeholder.

Where does the fiber go in a GPU cluster?

Almost all of it sits in the compute fabric, and the number of links, not the number of racks, drives it. Each GPU has its own fabric link, and a non-blocking leaf-spine design adds roughly one more link per GPU between leaf and spine. At the defaults, 1,024 GPUs sit in 128 nodes of eight and are served by 32 leaf switches with 32 GPU-facing ports each. Each node needs 71 fibers with spares, a 72-fiber cable, and each leaf needs 282, a 288-fiber cable toward the spine. The spine side adds 1,024 links, one per GPU.

Figure 1 shows the structure. NVIDIA’s DGX SuperPOD reference architecture for DGX H100 describes a rail-optimized, full fat-tree compute fabric with eight 400 Gb/s connections per DGX H100 node, and groups nodes into scalable units of 32. Its network fabrics page lists 1,024 spine-leaf cables and 1,020 compute and UFM cables for the four-unit, 1,016-GPU configuration, so spine-leaf links roughly equal GPU count, as the 1,024 links in this model do. Storage and in-band management each use a pair of ports per node on separate fabrics. At the defaults, the leaf-to-spine layer holds 58% of the order. For conventional racks, compare fiber cabling per data center rack.

Figure 2 puts the count in context. Adding each node’s compute links, its storage and management ports and its share of the leaf-to-spine layer, a GPU node lights 136 fibers at these defaults, or 17 per GPU. A conventional server with two duplex ports lights 4. The 34 times ratio sits close to the 36 times that an STL executive was quoted as saying for AI data centers, as Tom’s Hardware reported in May 2026. The match depends on assuming 8 fibers per link, so read it as a sanity check, not proof. Links between sites are sized like a metro ring.

Fiber type and optics are the other half of the specification. NVIDIA’s documents do not state the cable type or the fibers per link, so read the optic datasheet. A 400G DR4 module uses 8 fibers, four transmit and four receive, on an MPO-12 connector, according to a transceiver vendor’s explainer, which is a secondary source. The Ethernet standard behind 400G, IEEE 802.3bs, was approved in December 2017, and single-mode fiber is defined by ITU-T G.652. Tom’s Hardware also reported that manufacturers have shifted output from G.652D toward G.657A fiber, so confirm which category your supplier can deliver.

How much extra should you order beyond the fibers you light?

Order for the lit fibers plus five adjustments, and test where your numbers sit against the standard cable sizes. In this example the adjustments take 819.2 lit fiber-km to a 1184.4 fiber-km order, and raising spares from 10% to 13% can add 68% because it crosses a cable-size step.

Re-pulls are the part of contingency teams rarely price, because every failed link test sends a technician back with a new cable, and a cluster has thousands of links. The Technician Checklist Generator turns your acceptance tests into a checklist crews actually follow, which is the cheapest way to lower that number.

Figure 3 builds the order one step at a time. The fibers actually lit come to 819.2 fiber-km; the order is 1,184.4. Slack adds 142.8 fiber-km, the most of any step, because the runs are long and every one gets the 15%. Standard sizes add only 46.1 because 71 and 282 fibers land just under the 72 and 288 steps. That is luck, not safety.

Figure 4 moves one input at a time. Fibers per link dominates: 20% fewer (6 instead of 8) cuts the order by 14.5%, while 20% more (10 instead of 8) adds 67.7%, because 32 uplinks of 10 fibers with spares no longer fit one 288-fiber cable. Spares are the trap. At 8% or 12% they change nothing, yet at 13% the leaf cable doubles to two 288-fiber cables and the order rises 68%, from 1,184 to 1,987 fiber-km.

The fabric design moves the answer as much as the optics: a third tier adds about 58% and a 2:1 oversubscribed fabric saves about 29%. Fiber itself is a small part of the cost. At the roughly $6.30 per fiber-km that Tom’s Hardware reported, the 1,184 fiber-km in this order is under $8,000 of bare fiber, so the risk is delivery, not the glass price.

When do you need to buy the fiber?

Work backwards from cluster start: buy-by date equals start date minus supplier lead time minus a shipping and QA buffer, and treat the lead time as a number to confirm, not assume. The reports below describe the market in general and some rest on a single source, so use them to set expectations and then get written quotes.

In May 2026, Tom’s Hardware reported that fiber demand from data centers grew 76% in 2025, citing CRU data, that prices rose about 70%, and that lead times reached up to 20 weeks for large-volume buyers and up to a year for smaller ones. It reported that two large Chinese suppliers, Hengtong and FiberHome, were booked into early 2027. CRU Group’s OFC 2026 review names preform as the binding constraint and says ribbon cables for AI data centers, from 1,000 to over 13,000 fibers, carry lead times over a year. Those are market reports, not quotes to you.

Figure 5 applies those lead times to an example plan. If you order on 21 October 2026 with a 20-week lead time and a four-week buffer, the earliest start is 7 Apr 2027, only 8 days ahead of a 15 April 2027 cluster start. At 52 weeks the earliest start is 17 Nov 2027, about 31 weeks late. With so little margin, any slip in a quote moves the cluster date, which is why the buy-by date matters more than the exact fiber-km.

Figure 6 shows one way to cut exposure: one call-off per 256-GPU scalable unit. NVIDIA’s table shows spine-leaf cables growing by 256 with each unit, from 256 to 1,024, which supports building and cabling in units. In this example, four call-offs cut the peak quantity on site from 1,184 to 296.1 fiber-km. The catch is commitment: phasing only protects you if the supplier agrees in writing to hold allocation for every call-off.

How do you explain the number to a board, a customer or a network team?

Write the consequence, not the parameter. Every figure on this page becomes a sentence someone else has to accept: a board approving a purchase order, a customer told a go-live date has no slack, a network team told why a spare allowance was capped. Engineers tend to write these as parameters. Readers need to know what changes for them and what you have secured. When a fabric change goes wrong, the Outage Message Generator drafts the customer update in the same plain style. The examples below use this page’s illustrative numbers.

What should you check before placing the order?

Confirm eight things, and write down who owns each.

• GPU count and phasing match the approved design, by scalable unit, not the sales forecast.
• Fibers per link come from the optic datasheet and the bill of materials, not from a reference design that does not state them.
• Fabric tiers and oversubscription are fixed, because a third tier or a changed ratio changes every leaf cable.
• Every cable sits clear of a standard-size step, or the step is a recorded decision.
• Node-to-leaf and leaf-to-spine runs are measured along the tray route, not drawn as straight lines.
• Lead time is in writing for each cable type and states whether the date is shipment or delivery.
• Call-offs are planned by scalable unit, and the supplier has committed allocation for each one.
• Link acceptance tests are agreed before the first delivery. The Technician Checklist Generator can turn them into a checklist crews use.

Frequently asked questions

How much fiber does your rollout need? Run the numbers

Enter the numbers you know and the page returns fiber-km, cable-km, reels and a buy-by date. Start with site count, sectors, radios and the fronthaul design, then set the aggregation design. Open the advanced assumptions to change route lengths, reel lengths and buffers. Every default is an illustrative placeholder.

Where does the fiber go at a 5G cell site?

A site needs fibers in three places: up the tower to the radios, out along a lateral to an aggregation point, and, shared between about ten sites, along a trunk to the core. The lateral and the trunk hold 99% of the fiber-km, and the tower run holds about 1%. At the defaults, three sectors with two radios each need 12 fronthaul fibers under a dedicated design, plus 2 for backhaul. With 10% spares the lateral needs 16, so each site gets a 24-fiber cable, and each of the 5 trunks needs 53 fibers, a 72-fiber cable.

Figure 1 shows why the layers behave differently. The tower run is the shortest and is bought as one assembly per site. The lateral is multiplied by every site and runs a few kilometres. The trunk is multiplied by only 5 aggregation points but runs the longest, so it is the biggest layer by fiber-km even though it is the smallest by count. Trunks often close into a ring; see how many fibers a metro ring needs.

Figure 2 compares dedicated and shared fronthaul. Corning’s article explains that under CPRI each sector and band gets a dedicated fiber pair, so operators could drop 24 or 36 fibers at a cell site, and that with WDM the same tower can be served by 2 fibers. It was first published in 2018 and does not cover eCPRI. Ericsson’s January 2026 white paper adds that CPRI is constant-bitrate while eCPRI is packet-based, and gives an example in which a 100 MHz, 64-antenna carrier needs 300 Gbps over CPRI but 25 Gbps over optimized eCPRI. It gives no fiber counts, so the saving here is this model’s, not Ericsson’s: 22.5% fewer fiber-km, with unchanged cable-km. Poles that carry small cells may also carry city endpoints, covered in GPON for smart city.

Fiber types matter at each layer. Standard single-mode fiber is covered by ITU-T G.652, and bend-insensitive fiber by ITU-T G.657, the usual choice for tower and cabinet runs where cable turns tight corners. If you plan WDM, the channel plan comes from the ITU-T G.694.1 frequency grid. Check that every quote names the recommendation and category you designed for, because a substitute made under shortage pressure should be a design decision, not a purchasing one.

How much extra should you order beyond the fibers you light?

Order for the lit fibers plus five adjustments, and expect standard cable sizes to be the largest of them. In this example the adjustments take 5,016 lit fiber-km to a 9,508 fiber-km order, and rounding up to standard sizes alone adds 2,370.

Site access problems are the part of contingency teams rarely price, because each failed visit means a second crew trip. The Technician Checklist Generator turns your site acceptance steps into a checklist crews actually follow, which is the cheapest way to cut re-visits.

Figure 3 builds the order one step at a time. The fibers actually lit come to 5,016 fiber-km; the order is 9,508. A site needs 14 fibers and a tower needs 12, but cables come in 12 or 24, so the rounding step adds 2,370 fiber-km against 606 for the 10% spare allowance. Sizing is the decision to review first.

Figure 4 moves one input at a time. Site count and lateral length dominate: a 20% error in lateral length changes the order by 15.0%, while a 20% error in slack changes it by 1.8%. Cable sizes make some bars lopsided. Two sectors instead of three cut the order by 23%, because the lateral drops to a 12-fiber cable, but four sectors change nothing, because 18 to 20 fibers still fit 24. The trunk is the cliff: 48 fibers plus 10% spares is 53, which forces a 72-fiber cable, and trimming the design to 43 fibers would fit a 48-fiber cable.

Design moves the answer more than any buffer: a dedicated 6-sector, 3-radio site needs a 48-fiber lateral and the order rises to 13,876 fiber-km, about 1.5 times the default, while shared fronthaul brings it down to 7,366. That is why a page that quotes “fibers per site” without naming the fronthaul design is not safe to copy into a tender. Note that adding a third radio per sector changes nothing at these defaults, because 18 fronthaul fibers plus 2 backhaul still fit 24.

When do you need to buy the fiber?

Work backwards from the first build: buy-by date equals start date minus supplier lead time minus a shipping and QA buffer, and treat the lead time as a number to confirm, not assume. The reports below describe US builds and some are anonymous, so use them to set expectations and then get written quotes from your own suppliers.

In March 2026, Light Reading reported that some ISP leaders had received revised purchase orders with fiber cost increases of 70 to 80%, and that one BEAD-winning provider was told a loose tube cable it normally buys had a 52-week lead time. Corning denied that it was behind the increases. In July, Light Reading reported that Conexon told the FCC delivery times for non-BABA cable had risen by more than 40%. CRU Group’s OFC 2026 review says BABA-compliant fibre is largely sold out for 2026 with lead times over 50 weeks, and that some manufacturers have deprioritised BEAD orders in favor of hyperscale demand. Blandin on Broadband cited an anonymous source for normal lead times of 8 to 12 weeks.

Figure 5 applies those lead times to an example plan. If you order on 21 October 2026 with a 26-week lead time and a four-week buffer, the earliest start is 19 May 2027, ahead of a 1 July 2027 first build. The buy-by date for that case is 3 Dec 2026, 43 days from today. At 52 weeks the earliest start is 17 Nov 2027, about 20 weeks late. The 26-week figure is an assumed midpoint, not a reported one.

Figure 6 shows one way to cut exposure: one call-off per build wave. In this example, waves of 20, 20 and 10 sites cut the peak quantity on site from 9,508 to 3,803 fiber-km. The catch is commitment: phasing only protects you if the supplier agrees in writing to hold allocation for every call-off, and a supplier with a full order book may ask for project proof first.

How do you explain the number to finance, operations or a site partner?

Write the consequence, not the parameter. Every figure on this page becomes a sentence someone else has to accept: finance approving a purchase order, operations asked to change a fronthaul design, a landlord told when work starts. Engineers tend to write these as parameters. Readers need to know what changes for them and what you have secured. When a backhaul cut takes a site down, the Outage Communication Audit scores how clear your updates are. The examples below use this page’s illustrative numbers.

What should you check before placing the order?

Confirm eight things, and write down who owns each.

• Site count and build waves match the approved rollout plan, not the coverage target.
• The fronthaul design is fixed, dedicated or shared, and the radios per sector are confirmed with the equipment vendor.
• Lateral and trunk lengths are measured along the duct route, not drawn as straight lines.
• Every cable has a standard size, and the unused fibers are recorded as spare, not forgotten.
• Each layer has a named fiber type and ITU-T category, and every quote matches it.
• Lead time is in writing for each cable type and states whether the date is shipment or delivery.
• Call-offs are planned by build wave, and the supplier has committed allocation for each one.
• Site acceptance steps are agreed before the first delivery. The Technician Checklist Generator can turn them into a checklist crews use.

Frequently asked questions

How much fiber does your network need? Run the numbers

Enter the numbers you know and the page returns fiber-km, cable-km, reels and a buy-by date. Start with homes passed, the density profile and the share of homes you connect at build. Open the advanced assumptions to change cabinet size, buffers and cable specifications. Every default is an illustrative placeholder.

Where does the fiber go in a 10,000-home FTTH network?

Most of the fiber sits in the distribution layer, most of the cable length sits in the drops, and the feeder is small on both counts. In a centralized-split design, one fiber leaves the central office for each PON port. At the cabinet a splitter shares that fiber between homes, and from there every home has its own fiber back to the cabinet. With 288 homes per cabinet and a 1:32 split, each cabinet needs 9 PON ports, or 10 feeder fibers once 10% spare is added. Feeder fibers then return to a backbone sized like a metro ring.

Figure 1 shows why the layers behave differently. The feeder is long but carries few fibers. Distribution is shorter but carries about one fiber for every home passed. The drop is the shortest run, but it is multiplied by every connected home. Counting these layers separately is what keeps a plan from either over-ordering the feeder or under-ordering the drops. The layers follow the residential FTTH architecture you choose.

Figure 2 counts the same network in two units. Manufacturers and shortage reports talk in fiber-km, so that is the number to compare with supplier capacity. Contractors, reels and warehouse space work in cable-km, so that is the number that sets the logistics. A plan that quotes only one of them hides the layer that matters to the other audience.

Fiber type is the other half of the specification. Standard single-mode fiber is covered by ITU-T G.652, and bend-insensitive fiber by ITU-T G.657, the usual choice for drops and in-building runs where cable turns tight corners. Both recommendations were revised in August 2024 and remain in force. Check that every quote names the recommendation and category you designed for, because a substitute made under shortage pressure should be a design decision, not a purchasing one.

How much extra fiber should you order beyond the route length?

Order for the route plus four buffers: spare fibers, slack, waste and contingency. In this example they add about 31% to the bare route, and the inputs that most change the answer are not the buffers.

Failed installations are the part of contingency most teams never price, because each one may leave a cut drop cable that cannot be reused. The ISP Failed Installation Cost Calculator turns your failure rate into a cost, which tells you whether 5% contingency is generous or thin.

Figure 3 builds the order one buffer at a time. The bare route needs 6,440 fiber-km; the purchase quantity is 8,422.7. Spare fibers and slack account for about two-thirds of the uplift, so they deserve a recorded decision and an owner, not a default copied from a previous project. Fiber is one line in the budget; see fiber cost per home passed for the rest.

Figure 4 moves one input at a time. Homes passed and cabinet-to-terminal distance dominate: a 20% error in distance changes the order by about 15.5%, while a 20% error in slack changes it by 1.8%. Spend the survey budget on measuring distance on three representative areas, not on debating a point of slack.

Layout moves the answer more than any buffer: the rural case needs about 5.5 times the fiber of the dense urban case for the same number of homes. That is why a planning page that quotes “fiber per home” without naming a density profile is not safe to copy into a tender. A split ratio also matters, but only in the feeder: moving from 1:32 to 1:64 cuts this order by about 10%.

When do you need to buy the fiber?

Work backwards from construction start: buy-by date equals start date minus supplier lead time minus a shipping and QA buffer, and treat the lead time as a number to confirm, not assume. The reports below describe US builds and some are anonymous, so use them to set expectations and then get written quotes from your own suppliers.

In March 2026, Light Reading reported that some ISP leaders had received revised purchase orders with fiber cost increases of as much as 70 to 80%, and that one BEAD-winning provider was told a loose tube cable it normally buys now had a 52-week lead time. Corning, named in the reporting, said the increases do not reflect its approach to BEAD deployments. In July, Light Reading reported that Conexon told the FCC delivery times for non-BABA fiber cable had risen by more than 40% and that it expected its deployment to slow at least until 2027. An industry source cited by Blandin on Broadband, reprinting Light Reading, put normal lead times at 8 to 12 weeks.

Figure 5 applies those lead times to an example plan. If you order on 21 October 2026 with a 52-week lead time and a four-week buffer, the earliest start is 17 Nov 2027, about 24 weeks after a 1 June 2027 start. The buy-by date for that case was 5 May 2026, which has already passed. The 12-week and 26-week lead times are shown for comparison; the 26-week figure is an assumed midpoint, not a reported one.

Figure 6 shows one way to reduce the exposure: split the order into call-offs by build wave. In this example, three waves of 12, 12 and 11 cabinets cut the peak quantity on site from 8,423 to about 2,888 fiber-km. The catch is commitment. Phasing only protects you if the supplier agrees in writing to hold allocation for every call-off, and suppliers may ask for project proof before they quote. Light Reading also reported that Corning and partners said they would align delivery lead times with deployment schedules for BEAD winners, which is the kind of commitment to request.

How do you explain the number to a board, a partner or a customer?

Write the consequence, not the parameter. Every figure on this page eventually becomes a sentence someone else has to accept: a board approving a purchase order, a partner told the start date has moved, a customer asked to wait for a drop cable. Engineers tend to write these as parameters. Readers need to know what changes for them and what you have already secured. The examples below use this page’s illustrative numbers.

What should you check before placing the order?

Confirm eight things, and write down who owns each.

• Homes passed match the approved design, not the marketing target.
• Average cabinet-to-terminal distance is measured on at least three representative areas.
• Each layer has a named fiber type and ITU-T category, and every quote matches it.
• Every buffer is written down with an owner and a reason.
• Lead time is in writing for each cable type and reel length, and states whether the date is shipment or delivery.
• Call-offs are planned by build wave, and the supplier has committed allocation for each one.
• Reel acceptance checks are agreed before the first delivery. The Technician Checklist Generator can turn them into a checklist crews actually use.
• The calculator is re-run after the final route design, not only after the concept design.

Frequently asked questions

How much fiber does your hall need? Run the numbers

Enter the numbers you know and the page returns fiber-km, cable-km, cable assemblies and a buy-by date. Start with the rack count and the switch placement, then set uplinks and fibers per uplink from your optic datasheet. Open the advanced assumptions to change run lengths and buffers. Every default is an illustrative placeholder.

Where does the fiber go between a rack and the spine?

A rack’s fiber is its uplinks, not its servers. In a top-of-rack design only the uplinks leave the rack, so the rack needs one cable sized for them, and the hall adds a second cable layer from each leaf to the spine. At the defaults, four uplinks of 8 fibers each need 32 fibers. With 10% spares that is 36, so each rack gets a 48-fiber cable. Five leaf switches serve 100 racks, and each needs 71 fibers toward the spine, a 72-fiber cable.

Figure 1 shows why the layers behave differently. The rack cable is the one you multiply by every rack. The leaf-to-spine cable is longer but multiplied by only five leafs. At the defaults the rack-to-leaf layer holds 82% of the order, which is why the rack count and the run to the leaf decide the total. Links between halls or sites are sized differently; see how many fibers a metro ring needs.

Figure 2 compares where the switch sits. With end-of-row switching, every server link leaves the rack, so in this all-fiber model each rack needs 88 fibers with spares, a 96-fiber cable. The hall order is still smaller, 232.8 against 292.5 fiber-km, because those runs are 15 m rather than 40 m. Real halls rarely wire every server link in fiber, so use this as a bound, not a forecast. GPU halls use far more fiber per rack, as our guide to fiber for an AI GPU cluster shows.

The optic sets fibers per uplink, so read its datasheet before you fix the cable size. A 400G DR4 module uses 8 fibers, four transmit and four receive, on an MPO-12 connector, and reaches 500 m on single-mode fiber, according to a transceiver vendor’s explainer; that is a secondary source, so confirm against your own module. The Ethernet standard behind 400G, IEEE 802.3bs, was approved in December 2017. Single-mode fiber is defined by ITU-T G.652, and bend-insensitive fiber by ITU-T G.657, which suits tight tray bends. For the overall facility design, the TIA-942 data center standard was revised as version C in May 2024; check it for the cabling topology your hall follows.

How much extra should you order beyond the fibers you light?

Order for the lit fibers plus five adjustments: spares, standard cable sizes, slack, waste and contingency. In this example they take 166.4 lit fiber-km to a 292.5 fiber-km order, and the largest single step is the one most plans leave out.

Re-pulls are the part of contingency that teams rarely price, because every failed link test sends a technician back with a new cable. The Technician Checklist Generator turns your acceptance tests into a checklist crews actually follow, which is the cheapest way to lower that number.

Figure 3 builds the order one step at a time. The fibers actually lit come to 166.4 fiber-km; the order is 292.5. Rounding to standard sizes adds 48.6 fiber-km and the 10% spare allowance adds 20.2, so the sizing decision is worth more than the spare percentage you debated.

Figure 4 moves one input at a time. Rack count and the rack-to-leaf run dominate: a 20% error in that run changes the order by 16.3%, while a 20% error in slack changes it by 2.6%. Two bars are flat. Three or five uplinks per rack, or 8% or 12% spares, still fit a 48-fiber cable, so they change nothing. Cable size is a step function: the rack cable stays at 48 fibers up to 10 fibers per uplink and jumps to 72 at 11.

The optic choice moves the answer more than any buffer: 16-fiber uplinks need about 1.6 times the fiber-km of the default, and duplex uplinks need about 0.27 times. That is why a page that quotes “fibers per rack” without naming the optic is not safe to copy into a design brief. Fiber itself is a small part of the cost: at the roughly $6.30 per fiber-km that Tom’s Hardware reported in May 2026, the 293 fiber-km in this order is under $2,000 of bare fiber. The risk is delivery, not the glass price.

When do you need to buy the fiber?

Work backwards from hall start: buy-by date equals start date minus supplier lead time minus a shipping and QA buffer, and treat the lead time as a number to confirm, not assume. The reports below describe the market in general and some rest on a single source, so use them to set expectations and then get written quotes.

In May 2026, Tom’s Hardware reported that fiber demand from data centers grew 76% in 2025 and that fiber prices rose about 70%, with lead times of up to 20 weeks for large-volume buyers and up to a year for smaller ones. It also reported that preform capacity takes 18 to 24 months to build. CRU Group’s OFC 2026 review likewise names preform as the binding constraint and describes ribbon cables of 1,000 to 13,000 fibers with lead times over a year. For a US reference point, Light Reading reported a 52-week quote to a BEAD winner in March 2026.

Figure 5 applies those lead times to an example plan. If you order on 21 October 2026 with a 20-week lead time and a four-week buffer, the earliest start is 7 Apr 2027, ahead of a 1 June 2027 hall start. At 52 weeks the earliest start is 17 Nov 2027, about 24 weeks late, and the buy-by date of 5 May 2026 has already passed. The 12-week case reflects an anonymous industry figure for normal markets.

Figure 6 shows one way to cut exposure: one call-off per hall. In this example, halls of 40, 40 and 20 racks cut the peak quantity on site from 292.5 to 117.0 fiber-km. The catch is commitment. Phasing only protects you if the supplier agrees in writing to hold allocation for every call-off, and a supplier with a full order book may not.

How do you explain the number to finance, a tenant or operations?

Write the consequence, not the parameter. Every figure on this page becomes a sentence someone else has to accept: finance approving a purchase order, a tenant told a handover has moved, an operations team told the cable specification changed. Engineers tend to write these as parameters. Readers need to know what changes for them and what you have secured. When a cut-over goes wrong, the Outage Message Generator drafts the customer update in the same plain style. The examples below use this page’s illustrative numbers.

What should you check before placing the order?

Confirm eight things, and write down who owns each.

• Rack count and hall phasing match the approved design, not the sales forecast.
• Switch placement is fixed, and each link type is marked as structured fiber, copper or active optical cable.
• Fibers per uplink come from the optic datasheet, with the next generation noted.
• Every cable has a standard size, and the unused fibers are recorded as spare, not forgotten.
• Rack-to-leaf and leaf-to-spine runs are measured along the tray route, not drawn as straight lines.
• Lead time is in writing for each cable type and states whether the date is shipment or delivery.
• Call-offs are planned by hall, and the supplier has committed allocation for each one.
• Link acceptance tests are agreed before the first delivery. The Technician Checklist Generator can turn them into a checklist crews use.

Frequently asked questions

How many fibers does your ring need? Run the numbers

Enter the numbers you know and the page returns route fibers, fiber-km, cable-km, reels and a buy-by date. Start with node count, ring length, fibers per ring and the dark fiber policy. Open the advanced assumptions to change reel lengths and buffers. Every default is an illustrative placeholder.

Where does the fiber go in a metro ring?

Almost all of it is the route cable. The ring is one continuous cable around the route, and each node adds only a short lateral, so the route holds 95% of the order here. At the defaults, two rings of 2 fibers each light 4 fibers. Add 20 dark fibers for leasing, growth and repair and the route needs 24, or 27 with 10% spares, so you specify a 48-fiber cable. Each of the 8 nodes then adds a 12-fiber lateral of 1.5 km and a set of short jumpers.

Figure 1 shows why the ring is cheap to extend and expensive to start. Light travels both ways around the loop, so when the fiber is cut between two nodes, traffic still reaches every node the other way around. That resilience is a property of the equipment and its protection scheme: ITU-T G.8032 covers Ethernet ring protection switching and ITU-T G.841 covers SDH protection architectures. Neither is a fiber-count rule, so take the fibers each ring needs from your equipment vendor and enter them above.

Figure 2 compares four designs on the same 60 km route. The cable is 12, 24, 48 or 72 fibers depending on the dark fiber count, and the order moves from 1,029 to 5,312 fiber-km. Cable-km is 87.3 in the three designs with two rings, because the cable route is identical. The fiber is what changes, not the trench, which is why the dark fiber decision belongs to the business case, not the civil works plan. Rings like this often carry access and mobile traffic; see how much fiber 10,000 homes need and how much fiber a 5G cell site needs.

Fiber type and wavelength planning shape the other half of the specification. Standard single-mode fiber is covered by ITU-T G.652, and bend-insensitive fiber by ITU-T G.657, which suits tight bends at nodes. If you plan DWDM, the channel plan comes from the ITU-T G.694.1 frequency grid, and a single pair can then carry many channels, so lit fibers can stay few while demand grows. Check that every quote names the recommendation and category you designed for, because a substitute made under shortage pressure should be a design decision, not a purchasing one.

How much extra should you order beyond the fibers you specify?

Order for the specified fibers plus five adjustments, and expect the standard cable size to be the largest. In this example they take 1,585 fiber-km to a 3,599 fiber-km order, and rounding up to 48 fibers adds 1,260 of that.

Splice quality is the part of contingency teams rarely price, because a bad splice on a ring can mean reopening a closure. The Technician Checklist Generator turns your splice and test steps into a checklist crews actually follow, which is the cheapest way to cut rework.

Figure 3 builds the order one step at a time. The fibers specified come to 1,585 fiber-km; the order is 3,599. The route needs 27 fibers but cables come in 24 or 48, so rounding up adds 1,260 fiber-km against 180 for the 10% spare allowance. Those 21 unused fibers are not waste if you plan to lease or grow, but the decision should be explicit. Spare fiber that can be leased is part of open-access FTTH economics.

Figure 4 moves one input at a time. Dark fibers are a cliff, not a slope: at 17 dark the route needs 24 fibers and fits a 24-fiber cable, at 18 it needs 25 and moves to 48, so 20% fewer dark fibers (16) cuts the order by 47.6% while 20% more (24) changes nothing. Route length is the only continuous driver: a 20% error moves the order by 19.0%, against 1.8% for slack and 1.2% for node count.

The dark fiber count moves the answer more than any buffer: 5.2 times between the leanest and the largest design above. A one-ring design with 2 or 4 fibers needs the same 12-fiber cable when there are no dark fibers, so the protection scheme matters less to the cable than the growth plan does. That is why a page that quotes “fibers per ring” without a dark fiber policy is not safe to copy into a tender.

When do you need to buy the fiber?

Work backwards from the build start: buy-by date equals start date minus supplier lead time minus a shipping and QA buffer, and treat the lead time as a number to confirm, not assume. The reports below describe other segments and some rest on a single source, so use them to set expectations and then get written quotes.

In March 2026, Light Reading reported that one BEAD-winning provider was told a loose tube cable it normally buys had a 52-week lead time. In July, Light Reading reported that Conexon told the FCC delivery times for non-BABA cable had risen by more than 40%. In May, Tom’s Hardware reported that data center fiber lead times reached up to 20 weeks for large-volume buyers and up to a year for smaller ones, and that new preform capacity takes 18 to 24 months to build. A metro operator competes for the same fiber, so these are the pressures to plan for, even where the reports describe another segment.

Figure 5 applies those lead times to an example plan. If you order on 21 October 2026 with a 26-week lead time and a four-week buffer, the earliest start is 19 May 2027, ahead of a 1 September 2027 ring build. The buy-by date for that case is 3 Feb 2027, 105 days from today. At 52 weeks the earliest start is 17 Nov 2027, about 11 weeks late. The 26-week figure is an assumed midpoint, not a reported one.

Figure 6 shows why phasing helps less on a ring than on other builds. The route cable is 95% of the order and it is one continuous run, so even if the laterals arrive node by node the peak quantity on site only falls from 3,599 to 3,426 fiber-km. The route cable is the item to secure early. Ask the supplier for written allocation of the full route cable, and treat the laterals as the flexible part.

How do you explain the number to a board, a customer or operations?

Write the consequence, not the parameter. Every figure on this page becomes a sentence someone else has to accept: a board approving a purchase order, a customer told the ring cannot be phased, an operations team told why the dark fiber count was set. Engineers tend to write these as parameters. Readers need to know what changes for them and what you have secured. When a ring is cut, the Outage Message Generator drafts the customer update in the same plain style. The examples below use this page’s illustrative numbers.

What should you check before placing the order?

Confirm eight things, and write down who owns each.

• Node count and ring route are fixed in the approved design, not the sales plan.
• Fibers per ring come from the equipment vendor and the protection scheme in use.
• The dark fiber policy is written down, with the commercial reason for the count.
• The route cable size sits clear of a standard-size step, or the step is a recorded decision.
• The route length is measured along the duct or pole route, not drawn as a straight line.
• Each layer has a named fiber type and ITU-T category, and every quote matches it.
• Lead time is in writing for the route cable and states whether the date is shipment or delivery.
• Splice and test steps are agreed before the first delivery. The Technician Checklist Generator can turn them into a checklist crews use.

Frequently asked questions

Why is FTTH for rural villages hard to justify?

Because the cost of each route is shared by very few homes. The Fiber Broadband Association (FBA) and Cartesian explain it plainly in their 2025 survey: cost per home passed generally rises as density falls, because total cost is spread over fewer homes, even though cost per foot falls in rural areas with fewer roads, pavements and buildings to cross.

Read the numbers together. Europe shows rural fiber is possible, and the Council credits public funding programmes and targeted deployment initiatives for reaching the most complex areas. The US survey shows costs are rising, with labor, materials, permitting delays and make-ready work named as the main drivers. Indonesia shows how much room is left in a large, spread-out market. A village project has to stand on its own sums, whoever is paying.

The three sections below take one angle each: the business case, shared infrastructure and the last mile. Cost figures come from the US, so treat them as a guide to the shape of your costs, not a price list for your country.

How do you make FTTH for rural villages commercially viable for operators?

Lower the route cost, raise the number of paying customers on it, and keep the first phase small. The business case turns on three numbers: cost per foot, feet of route per home, and who has signed up before you build. For benchmarks, see our fiber cost per home passed guide.

Two lessons follow. First, labor is the biggest line, so any choice that cuts crew time, such as using poles that already stand, saves more than a cheaper cable. Second, the per-foot saving in rural areas does not make the project cheap, because the number of feet per home is large. Figure 1 shows the effect with our own arithmetic: the rural per-foot medians multiplied by the route length per home.

The table is arithmetic, not a quote. It ignores drops, cabinets, electronics and the shared trunk to the village, and a real village has a dense core and a thin edge. Still, it shows why a project should begin where homes are close, pass the core first and treat the scattered edge as a later phase or a different technology. The same trade-offs appear in FTTH network design for low-density areas.

Stack anchor demand before you build

Households alone make a thin case. Schools, clinics and local government offices make it steadier, because they pay on contract and need the connection every day. The World Bank’s options study on open-access fiber over electricity infrastructure in Indonesia makes the point directly: aggregated procurement of FTTH services at subdistrict and village level, for schools, health facilities and local government offices, could create “anchor demand” that supports wider use and strengthens the investment case. The study was issued in July 2024 and its preliminary analysis showed potentially positive results.

A viability checklist for operators

1. Count feet per home, not just homes. Measure the route to each cluster and compute cost per home passed before you commit.
2. Prefer poles that already stand. Labor is 64% of aerial and 72% of buried cost, so using existing poles or ducts saves the most.
3. Sign anchors first. Seek written commitments from schools, clinics and the village office.
4. Build the core first. Pass the dense centre, then extend in phases as take-up proves out.
5. Plan for rising costs. The FBA survey shows costs rising for most builders, so add contingency for labor and permitting delays.

How does open access make FTTH for rural villages cheaper and attract more ISPs?

It replaces several thin, duplicated builds with one shared network that any ISP can lease. Broadband has high fixed costs. In a village there is often room for one network and no more, so open access lets that one network serve many providers. Here is how open-access FTTH improves utilization and revenue.

BroadbandNow’s 2025 explainer defines an open access network as a neutral platform where the owner does not sell retail service and offers the same terms to every provider. It says a shared network avoids duplicate builds and lowers barriers to entry, since new ISPs can lease capacity instead of building. It also warns that its figures are not independently verified and that some numbers are inconsistent, so read the examples below as indicative.

These are US examples, but the lesson carries: shared networks can attract several ISPs, yet they need patient funding because early take-up can fall short. UTOPIA’s 16% against a 35% goal is a warning to size debt for slow starts.

Reuse what exists, then open it up

The World Bank study above looks at open-access fiber deployed over passive electricity infrastructure in Indonesia, in areas that currently lack service. Using existing power poles lowers the labor-heavy part of the build. In the UK, rural provider Village Networks, which runs both full fibre and fixed wireless, has applied to Ofcom for Code Powers that could reduce street-works licences and help it reach Openreach’s ducts and poles, as ISPreview reported in February 2026. Different countries, same idea: reuse ducts and poles, and let more than one provider use the result.

Choose this if

How do you overcome long distances, rough terrain and limited infrastructure in FTTH for rural villages?

Match the method to each stretch of ground, and keep the optical budget honest. Distance alone seldom stops fiber. ITwissen’s GPON summary gives 20 km upstream reach for ITU-T G.984, with a splitter of 32 or 64 outputs. For longer routes, ITU-T G.984.6 covers GPON reach extension (base edition March 2008, amendments in 2009 and 2012). What decides a village route is cost, terrain and loss.

The FBA survey confirms terrain matters: it says median costs vary with terrain, density and construction method, and that permitting delays are increasingly hitting budgets and timelines. The table is our summary of common practice, so confirm each method with local rules and a site survey. The passive PON design helps in remote places, because splitters need no power. Only the PoP and the home ONTs do.

The last few hundred metres

Drops are where small costs add up. The FBA and Cartesian report gives median drop costs per connection, excluding the ONT and customer equipment: $338 for an aerial drop under 100 ft rising to $400 over 500 ft, and $556 buried under 100 ft rising to $675 over 500 ft. Labor is about 60% of aerial and 70% of buried drop cost. In a village that means two rules: keep drops short by placing splitter cabinets near clusters, and choose aerial drops where poles allow.

Keep the field work repeatable

1. Walk the route first. Record poles, crossings, rock and landowners before you finalise the design.
2. Count the loss. Add every splice, connector and splitter along the longest path, then confirm the budget for your split ratio.
3. Standardise the install. Remote crews have little backup. TheWriter.id’s free Technician Checklist Generator builds a pre-installation checklist that crews can follow the same way in every village.
4. Tell people what is happening. Rural outages take longer to fix, so say so early. The Outage Message Generator drafts copy-ready notices for email, SMS and status pages.
5. Keep spares and test gear near the village. A repair that needs a long trip first costs days.

What should you confirm before committing to FTTH for rural villages?

Confirm the cost per home passed, the anchors, the access model and the route, in that order. This list turns the sections above into questions for an operator, a council or an integrator.

• What is the route length per home for the village core, and what is cost per home passed by method?
• Which poles and ducts already exist, and on what terms can they be used?
• Which schools, clinics and offices will sign contracts before the build?
• Is the network open access, and who owns and finances it?
• Can the project survive a take rate well below the goal for the first years?
• What is the optical budget on the longest path, counting every splice, connector and splitter?
• Which stretches need a bore, a span or rock work, and what do they cost?
• Who writes the installation checklists and the customer notices?

FTTH for rural villages: frequently asked questions

Where does GPON fit in a smart city, and where does it not?

GPON fits fixed endpoints with steady data and a pole or cabinet to host an ONT. It does not fit endpoints that are mobile, battery-powered or in need of guaranteed real-time response. Sorting the city’s endpoints first saves you from forcing one technology onto every use.

A published case shows the fit. Tejas Networks describes a GPON network in the ELCITA industrial township in Electronic City, Bengaluru, built with OLTs at three access nodes and ONTs and switches on street poles. It covers more than 100 poles and 200 video cameras, and carries video surveillance, public announcement, smart lighting, smart waste management and Wi-Fi hotspots, with IoT devices integrated into the network operations center. The case study is from 2020, so treat it as a design example, not a current benchmark.

The low-rate row matters. An overview of LPWAN in smart cities ties LoRaWAN and NB-IoT to streetlights, parking, waste, air quality and smart meters. It describes LoRaWAN as long range, up to 15 km, and NB-IoT as running on existing LTE networks. These sensors send small readings, so a gateway on a GPON pole can backhaul many of them at once. That is a better use of the pole than a dedicated ONT per sensor.

Choose this if

How many CCTV cameras can one GPON port carry?

About 149 to 373 cameras per port at a 60% planning load, depending on resolution. Camera video flows upstream, so the upstream rate sets the limit. ITU-T G.984 GPON offers 2.488 Gbit/s downstream and 1.244 Gbit/s upstream, with a passive splitter of 32 or 64 outputs, according to ITwissen. The upstream is shared by every ONT on the port.

Step 1: pick a bitrate per camera

Bitrate depends on resolution, frame rate, codec and how busy the scene is. Video Experts Group’s recommended settings give typical H.265 ranges for 15–30 fps with moderate motion, and note that busy streets may need more. Use the top of the range for planning. When the camera count outgrows GPON, our XGS-PON network design guide shows the next step, and GPON to XGS-PON migration covers the upgrade order.

Step 2: apply the planning load

Our rule is to plan steady traffic at 60% of the upstream, which leaves headroom for motion peaks, Wi-Fi backhaul and growth. That is 1,244 Mbps × 60% = 746 Mbps usable. Divide by the planned bitrate and round down: 746 ÷ 2 = 373, 746 ÷ 3 = 248 and 746 ÷ 5 = 149. The 60% figure is a design choice, not a standard, so change it to match your own risk tolerance.

Step 3: decide when to move to XGS-PON

ITU-T G.9807.1 defines XGS-PON at a nominal 10 Gbit/s in both directions, about eight times GPON’s upstream. If the camera math fails on a district, or you plan to add 4K cameras and analytics later, specify an OLT and splitter design that can move up without re-laying fiber. Where the math passes with headroom, GPON stays the lower-cost option.

Where PoE-powered endpoints fit

The fiber does not carry power, so each pole needs its own. The usual design is mains power to a cabinet, an ONT, and a PoE switch that powers the cameras and the Wi-Fi access point over copper. IEEE 802.3 port limits are 15.4 W (802.3af), 30 W (802.3at) and up to 90 W (802.3bt). Add up the peak draw of every endpoint, including pan-tilt-zoom cameras and heaters, and keep the total under the switch budget. Read the data sheet’s peak draw, not its typical draw.

Before the crew goes out, give them a checklist. TheWriter.id’s free Technician Checklist Generator builds a pre-installation list for field technicians, which suits repeated pole installs where every ONT, switch and camera must be checked the same way.

How should a city phase a GPON rollout?

Start with one pilot district, build the shared infrastructure once, then repeat the proven pattern. The expensive parts of a smart-city network are the poles, ducts, power and trunk fiber. They are also the parts you can build once and share, so a phased plan spends most of the money where it is reused. For ducts under streets, see underground FTTH network design.

1. Pick a pilot district that shows real mixed use. Include CCTV, public Wi-Fi and a few controllers on shared poles, so the pilot tests the worst case for a port.
2. Measure before you scale. Record upstream load per PON port, PoE headroom per pole and the time to repair a fault. These are the three numbers that decide whether phase 2 is safe.
3. Build shared infrastructure once. Lay the trunk with spare fibers, size ducts for the full city and give each pole power and space for an ONT and switch.
4. Add districts by repeating the pattern. Each new district is a splitter, some ONTs and a few cameras. If a port passes its planning load, add a port or move up to XGS-PON.
5. Hand over to operations early. Write procedures, naming and escalation rules during the pilot, so the network operations center runs district two the way it ran district one.

Indonesia shows why the national backbone and the street network are separate jobs. A 2018 DCD analysis reported that the Ministry of Communications and Informatics launched its “100 smart cities” initiative in 2017, and described the Palapa Ring as a national backbone of about 12,000 km of fiber. The article is dated, but the point stands: a backbone does not give each district its own last-mile design. That is what a pilot district tests.

What should you confirm before committing to GPON for a smart city?

Confirm the endpoints, the camera math, the power and the pilot, in that order. This list turns the sections above into questions for a vendor or an integrator.

• Which endpoints are fixed with steady data, and which are battery-powered, mobile or real-time?
• What camera resolution, codec and bitrate will you plan at, and why?
• How many cameras per PON port at your planning load, and does every district pass?
• What split ratio will each port use, and what is the worst-case load on a full split?
• What is the peak power draw per pole, and does the PoE switch budget cover it?
• Can an OLT and splitter design move a district to XGS-PON without re-laying fiber?
• How will OLT sites be linked together, and by what technology?
• Is the pilot district defined, with measures for port load, PoE headroom and repair time?
• Who writes the operations procedures, and when?

GPON for smart city: frequently asked questions

What is FTTH network architecture for residential networks?

It is the layered design that carries fiber from an exchange to every home. Read it from the core outward and it has four segments.

1. Core and aggregation. The OLT sits here, with the uplinks to the wider network.
2. Feeder and distribution. Fiber runs from the OLT to splitters in cabinets, or to a fiber hand-off frame in a point-to-point design.
3. Drop. Short cables run from the last splitter or frame to each building or home.
4. In-building and home. Riser fiber and the ONT finish the path inside an apartment block or house.

The big topology choice is PON against point-to-point. In a PON, one fiber is shared by many homes through passive splitters. In a point-to-point network, each home has its own fiber back to a central frame. BEREC’s report on next-generation access notes that the two behave very differently for wholesale: in a PON only the last segment, the dedicated fiber between the end user and the last splitter, can be physically unbundled. That one fact shapes everything in the next section. On greenfield estates, see FTTH network design for new housing developments.

The numbers show why architecture matters now. Europe has passed 295 million homes but only 54% subscribe, so the next gains come from competition and service quality on existing fiber, not new ducts. Indonesia’s fixed broadband penetration of about 42% leaves wide room to grow. In both markets, the network that can host several ISPs and absorb the next speed tier wins more subscribers per kilometre of fiber.

How do you design FTTH network architecture so multiple ISPs can use one residential fiber network?

Pick the layer where each ISP plugs in, then design the network around that choice. There are three: fiber at Layer 1, bitstream at Layer 2 and IP at Layer 3. Each moves work and risk between the network owner and the ISP. For the business side of that choice, compare the open access and retail FTTH models.

BEREC describes bitstream as Layer 2 (Ethernet) and Layer 3 (IP) access, and notes that higher layers add functionality but reduce the ISP’s flexibility. A Layer 2 Ethernet bitstream typically needs VLAN allocation, quality-of-service handling, flexible interconnection and a free choice of customer equipment. BEREC also warns that a native Ethernet network is limited to about 4,000 VLANs, so large open-access networks plan their VLAN scheme early. BEREC’s report dates from 2010. The layers themselves have not changed, but check current national rules before you copy any detail.

Open access inside an apartment building

The same question appears again at the building. The Fiber Optic Association’s MDU guide describes two layouts. In the first, a splitter at the entrance facility sends a drop cable to each unit, which it calls home-run style. This lets the owner offer multiple providers, since each unit’s fiber can plug into whichever provider’s equipment the resident picks. The cost is more fiber and cable. In the second, cascaded splitters sit in the riser or on each floor. They use fewer fibers, but the FOA notes this limits the building to one service provider. Its example is a 1:4 first splitter feeding 1:8 floor splitters, a 32-way split serving 4 floors with 8 units each.

Choose this if

How do you future-proof FTTH network architecture for growing bandwidth demand and high-density housing?

Treat the fiber and splitters as a 30-year asset and the optics as a 10-year one. Demand keeps growing, and upstream is growing fastest. If the passive plant is generation-neutral, each speed upgrade becomes a change of OLT card and ONT, not new construction. Our guide to GPON to XGS-PON migration shows which customers to move first.

These figures are US data from OpenVault, reported through a news summary, so use them as a trend signal and not as a forecast for another country. The direction is clear in any market: usage grows every year, heavy users grow faster than the average, and upstream grows fastest. A residential design that sizes only for today’s downstream will be short within one upgrade cycle.

Size the split with real take-up

Splitter ratio sets the shared capacity per home. The table divides the nominal line rate of each PON by the number of homes on the split, once at the 54% take-up Europe reports and once at the worst case where every home is connected. GPON runs at 2.488 Gbit/s downstream, and XGS-PON at a nominal 10 Gbit/s symmetric under ITU-T G.9807.1.

These are our calculations (line rate divided by split size times take-up), not guarantees. Homes are rarely all busy at once, so real experience is better than the table. Still, the table shows why a 1:64 GPON split gets tight as heavy users grow, and why XGS-PON on the same fiber gives roughly four times the headroom. Note that the XGS-PON row is a PON-level division. Check the OLT and ONT datasheets for the real limits.

Let generations share one fiber

You do not have to choose one PON for the life of the network. ITU-T G.9805 defines how two or more PON systems share a common optical distribution network, using wavelength coexistence elements. It covers GPON, XG(S)-PON, NG-PON, 10G-EPON and 50G-PON. Amendment 2, approved on 29 November 2025, adds Class C+ and Class D optical path loss specifications for multi-PON module configurations and deprecates the 12.5 Gbit/s upstream option of 50G-PON. The 50G-PON physical layer itself is specified in ITU-T G.9804.3.

Design rules for a generation-neutral plant

1. Keep splitter points accessible. Put them in cabinets or MDU rooms you can reach, not buried in joint closures, so a split change is a visit and not a dig.
2. Leave a loss margin. Coexistence elements add loss. Budget the worst path with the element in place and check the optical class against ITU-T G.9805.
3. Use bend-insensitive fiber. The FOA notes it allows small drop cables inside buildings, which matters in dense blocks.
4. Plan for take-up that grows. The FOA advises against assuming 100% take-up and says to plan for future additions. Walk the building before you design it.
5. Reserve ports and duct. Spare OLT ports, spare fibers at the frame and spare innerduct cost little now and avoid civil work later.
6. Prepare migration notices. Each ONT swap is a customer event. TheWriter.id’s free Outage Message Generator drafts copy-ready notices for email, SMS and status pages.

What should you confirm before finalising a residential FTTH architecture?

Confirm the topology, the hand-off, the building wiring and the upgrade path, in that order. This list turns the sections above into questions for a designer or integrator.

• Is the network PON or point-to-point, and does regulation require physical unbundling?
• At which layer does each ISP plug in, and who runs the OLT?
• Is there a VLAN, quality-of-service and interconnection plan that stays inside the VLAN limit?
• Do apartment blocks use home-run drops, or does cascading tie the building to one provider?
• What is the shared capacity per home at realistic take-up, not 100%?
• Is the passive plant generation-neutral, with a loss budget that allows a coexistence element?
• Are splitter points accessible, and are spare ports, fibers and duct reserved?
• Who writes and sends the customer notices for each migration?

FTTH network architecture: frequently asked questions

Why does hotel Wi-Fi fail guests even when the building has fiber?

Fiber to the building does not guarantee a good signal in the room. The last few metres, a corridor AP sending radio through a door and a wall to a phone on a bed, is where most guest complaints start. Guests care about this a lot, and the radio physics work against you as the frequency goes up.

The survey is a few years old, so treat it as a direction, not a current benchmark. It was run among more than 670 guests and 200 hoteliers in the United States, and it is summarised in Travel Daily News. The wall figures are more durable because they are physics. Extreme Networks notes that brick and concrete absorb a signal heavily and drywall far less. A Cal Poly measurement study found that Wi-Fi 6E loses on average 62.3% more signal than 2.4 GHz and 43.8% more than 5 GHz through the wall assemblies it tested. Newer, faster Wi-Fi bands reach less far through walls, not further.

A real case shows the cost of the old design. Developing Telecoms (8 December 2023) reported that Wyndham Royal Lee Phuket ran VDSL at 100 Mbps with one Wi-Fi gateway for every 2–3 rooms. The result was poor coverage and actual speeds under 10 Mbps per room, with frequent guest complaints. True and ZTE then installed one main ONT and 26 room ONTs on each floor, so every room had its own ONT, and True Online offered 2 Gbps packages to give each room gigabit speeds.

Should a hotel use fiber to the room, or fiber to the lobby plus Wi-Fi?

It depends on the walls, the services each room needs and the budget per room. Neither design is always better. The table compares them on the points that decide hotel projects. It is our engineering summary, so confirm each row against your own building with a site survey.

Standards are catching up with this design, which lowers the risk of choosing it. ITU-T G.9940 defines the system architecture for fibre-based in-premises transceivers and was approved in December 2023. G.9941 and G.9942 followed in July 2024 and G.9943 in October 2025, according to an ITU-T workshop presentation. ETSI’s F5G group has worked on the FTTR use case with the Broadband Forum, CCSA and ITU-T. Ask any vendor which of these they follow and what is proprietary. If the hotel plans a 10G backbone, read our XGS-PON network design guide before you fix the room count per port.

Choose this if

Which room ONT specs matter for a hotel?

Size, noise, power, Wi-Fi standard, ports and tamper resistance decide whether a room ONT works in a hotel. A residential unit can pass a lab test and still fail a hotel, because a guest sleeps next to it, a housekeeper moves it and a bored guest unplugs it. The table lists what to ask for. The values are our recommendations, not a standard. For what an install really involves on site, see do they drill holes for fiber internet.

Power is the spec most often left to the end. Fiber carries light, not power, so a room unit needs power from somewhere. The usual options are PoE from a switch over copper, a hybrid cable carrying both fibers and copper conductors, or a local adapter. If you choose PoE, the IEEE 802.3 generations offer different budgets at the switch port: 15.4 W (802.3af), 30 W (802.3at), 60 W (802.3bt Type 3) and 90 W (802.3bt Type 4). A device on a 90 W port can receive up to 71.3 W after cable loss. Read the unit’s peak draw, not its typical draw.

A line like the second one is also what a good product page needs. If you sell hotel ONTs, write the specs a hotel buyer actually asks about. TheWriter.id’s free Technician Checklist Generator helps with the next step, turning the installation side of a room rollout into a checklist your field team can follow.

How should fiber be run inside hotel walls to reach every room?

Plan one path from the riser to each room, keep the number of joints low and use bend-insensitive fiber. In a hotel, the cable is hidden after opening day, so a mistake in the wall becomes an expensive ceiling job later. Every joint, connector or splice on the path takes a share of your optical budget, so count them before you order. Vertical risers follow the same logic as FTTH network design for high-rise buildings, floor by floor.

Pathways and patch points

1. Riser: Run the vertical backbone in a riser shaft, and restore the fire-stopping wherever cable passes through a floor or rated wall.
2. Floor box: Terminate or split at one accessible box per floor. This is where a fault is isolated first, so label every port with a room number.
3. Corridor tray: Route fibers in a tray or conduit apart from mains power, with a bend radius the cable allows.
4. Door box: Leave one accessible joint per room above the corridor ceiling or in a door-frame box. This is the last point you can reach without opening the wall.
5. Room unit: Finish with a short pre-terminated lead and keep a service loop, so the unit can be replaced without re-splicing.

Which fiber type?

Bend-insensitive single-mode fiber to ITU-T G.657 is the practical default inside hotel walls, because it tolerates the tight corners of ducts, wall boxes and door frames. The ITU-T presentation on in-premises fibre describes pre-connectorized fibre for fast installation. The bare-fiber radii below are the minimums in the G.657 family; the finished cable has its own larger limit on its datasheet.

Common install mistakes

What should you confirm before ordering room ONTs for a hotel?

Confirm the walls, the services, the power and the path, in that order. This list turns the sections above into questions you can ask a vendor or an installer.

• What are the walls between rooms made of, and what does a site survey show for coverage?
• Which services must run on a wired port in the room, and on which VLAN?
• What is the unit’s peak power draw, and how is it powered in every room?
• Is the unit fanless, with a locked mount and an offline alert?
• Which Wi-Fi standard does it support, and does it roam between rooms on a floor?
• How many joints sit between the riser and the room, and is each one accessible?
• Is the fiber G.657.A2 or better, and does the cable’s own bend radius suit the wall boxes?
• Which ITU-T or ETSI documents does the vendor follow, and what is proprietary?
• Can you pilot one floor first and measure before you roll out the building?

Fiber to the room hotel: frequently asked questions

How do you link villas, pools and beach areas across a large resort site?

Build a star from the back-of-house core, with buried duct trunks to splitter cabinets and short drops from there. The passive splitters in each cabinet need no power, so the only powered equipment sits where you can service it: the core and the endpoints. Distance is rarely the constraint. The cable route is. Where you can trench, the same rules as underground FTTH network design apply: leave spare duct for the next expansion and repair.

Aerial or buried?

Both can work. The right choice depends on soil, existing poles and how exposed the route is to storms. The table sums up what the outside-plant sources say, with our resort recommendation in the last column. For thinly spread villas, the trade-offs are close to FTTH design for low-density areas.

The ADSS facts come from ISE Magazine’s guide to aerial cable types (2019), which also notes that if a weather event exceeds the design wind and ice loadings, “this can cause a service outage.” An earlier ISE comparison (first published 2013) adds that loose-tube cable lashed to a strand is not designed for large wind and ice tensile loads, while ADSS can carry loads up to thousands of pounds depending on the design. The buried-cable facts come from the Fiber Optic Association’s outside-plant guide. One resort-specific rule of our own: never hang cable from trees or building roofs. They move, rot and get cut.

Distance limits that matter

Fiber’s reach is generous. ITwissen’s GPON summary gives 20 km for the upstream rates of ITU-T G.984 and a splitter of 32 or 64 outputs. A resort site is almost always well inside that, so reach is not the limit. Copper is. A twisted-pair Ethernet channel, including PoE, stops at 100 m, which is why a pool bar, a beach club or a far villa cluster needs fiber to its door, with an ONT or switch placed near the endpoint. Fiber’s real limit is the optical budget: every splice, connector and splitter takes loss, so count them across the whole route before you fix the split ratio.

Outdoor-rated gear

Any equipment outside needs an ingress rating under IEC 60529. As Zenitel’s explainer puts it, IP66 means dust-tight and protected against high-pressure water jets from any direction. IP67 is equally dust-tight but covers temporary immersion between 15 cm and 1 m instead. For splitter cabinets and outdoor ONTs, IP66 is the practical minimum; choose IP67 where water can pool, such as a ground-level handhole.

Choose this if

What fiber network equipment survives salt air, humidity, storms and lightning?

Glass fiber is the easy part. The enclosures, fasteners, glands and any metal in the path are what fail. A coastal resort needs equipment chosen for chlorides and condensation, a cable route sized for storms, and a lightning plan that keeps metal out of the signal path. If you need to explain the result to owners, our look at how reliable fiber internet really is covers the uptime question in plain terms.

The numbers above come from one vendor knowledge-base article on marine junction boxes. It describes itself as general information, so treat it as a guide and confirm against the datasheet of the enclosure you buy. Its points line up with common practice: specify 316L, not 304; use A4-80 stainless fasteners; use stainless or nickel-plated brass glands with EPDM seals, because nylon glands degrade quickly in salt spray; slope enclosure tops at least 15° so water does not pool; and add anti-condensation heating when the temperature swing exceeds 15°C.

For lightning, start from the cable. ADSS has no metal, so ISE notes it needs no bonding or grounding. Armoured cable has a metallic layer that, per the Fiber Optic Association, must be grounded and bonded. The ITU-T publishes guidance on protecting telecom lines against lightning: K.47 covers protection of telecommunication lines against direct lightning flashes (edition May 2012), and the ITU lists the older K.25 on optical fibre cables as withdrawn, with its requirements incorporated into K.47. Ask your designer which edition and national code they follow. The practical rule is simple: the weak point is usually copper, such as PoE runs to a pool-side camera, not the fiber.

How do you separate POS, cameras, staff and guest traffic on one fiber backbone?

Give each zone its own VLAN on the shared fiber, then control routing at a firewall that denies traffic between zones by default. One fiber backbone can carry all four lanes. What matters is where the lanes are separated and who can cross them. IEEE 802.1Q is the standard that specifies virtual LANs and VLAN bridges.

Segmentation is not only a security habit. According to Purple’s guide to PCI DSS for hotel Wi-Fi, a vendor source, guest Wi-Fi is treated as in scope until it is shown to be segmented from the cardholder data environment, and segmentation is the practical way to reduce scope. The PCI Security Standards Council’s supplement (September 2024) covers scoping and segmentation in modern network designs, and states that it is supplemental and does not replace any PCI standard. Confirm your design with your assessor.

Where to separate on a PON

1. Carry zones as VLANs end to end. Use a PON ONT or switch that supports multiple VLANs, so every zone rides the same fiber to the same outlet without being mixed.
2. Enforce rules at one firewall. Terminate the VLANs at the back-of-house core, and deny routing between zones by default. Allow only named exceptions.
3. Give POS its own path where audit demands. A dedicated ONT or port for the payment zone is a simple way to prove separation.
4. Isolate cameras. Put them on their own VLAN reaching only the recorder, and check that power over copper stays inside the building or in short, surge-protected runs.
5. Test the separation. From a guest device, try to reach a POS or camera address, and record that it fails. Repeat after every change.

What should you confirm before building a resort fiber network?

Confirm the route, the gear, the weather plan and the zones, in that order. This list turns the sections above into questions for a designer or integrator.

• Which routes can be trenched or bored, and which need aerial ADSS?
• Is the trunk in duct with spare innerducts for future villas?
• What is the optical budget for the longest route, counting every splice, connector and splitter?
• Are all outdoor boxes IP66 or better, in 316L stainless or FRP, with stainless fasteners and glands?
• Are boxes mounted at least 500 mm above grade, sloped to shed water, and heated or desiccated where needed?
• Is the cable all-dielectric where possible, with armour bonded and surge protection where copper or mains enters?
• Are outdoor copper and PoE runs short, and is the rest done in fiber?
• Are guest, staff, POS and camera traffic on separate VLANs, with a firewall denying routing between them?
• Who inspects coastal equipment, how often, and from which checklist?

Fiber network for resorts: frequently asked questions