The Next Span
What’s coming to fiber and telecom, one span at a time.
Get The Next Span.
The AI and robotics changing telecom and fiber. Free, in your inbox.
Lead story
Multi-core fiber gets its first international standard
At the Global Optical Fiber and Cable Conference 2026, which opened Monday, Oct 19, in Wuhan, the organizers released what they call the world’s first international standard for multi-core optical fiber. The Asian-Pacific Optical Fiber and Cable Industry Association (APC) coordinated the standard, and FiberHome Telecommunication Technologies and Shanghai University were among the co-developers. It covers weakly coupled single-mode multi-core fibers, which put several independent cores inside one cladding (“akin to turning a one-lane road into a multi-lane highway,” in the report’s words). The target uses are links inside AI data centers, high-capacity submarine cables and other multiplexed systems. Ao Li, vice president of the China Academy of Information and Communications Technology, said multi-core fiber is moving “from application exploration to engineering deployment, with live-network verification already completed,” and that demand is expected to exceed 2.7 million core kilometers by 2030.
Quick hits
Running a fiber access network in plain language.
Korea’s ETRI says it has built “LLM-based optical access network operation technology” (announced Oct 14). An operator types a goal in natural language, and the AI plans the steps and controls the optical access equipment. ETRI says it checks AI-generated procedures in a virtual copy of the live network before applying them, so bad commands get caught first. It is also developing a local model that runs inside the network, so configuration data doesn’t leave it.
A robot that wiggles the fiber before it unplugs it.
A Nokia patent granted Sep 29 (US 12,746,685) describes a mobile robot with cameras and a gripper that connects and disconnects fibers at a patch panel. Before it pulls a fiber, it moves the fiber slightly and checks for a change in the signal’s polarization to confirm it has the right one. The patent says this guards against mistakes “caused by improper cable labelling or outdated network databases.” (This is a patent filing, not a shipping product.)
Meta’s robot technicians are slower than the people.
TechRepublic (reported Aug 31), summarizing WIRED’s reporting, says Meta is testing robots from Watney Robotics, Kinova and ABB to swap network cables, power-cycle servers and reseat hardware at its Altoona, Iowa and New Albany, Ohio campuses. The robots work under supervision, are slower than human technicians, and need people to move them between buildings. Meta has acknowledged internally that they aren’t equipped for the heavy cabling that Nvidia GB300 systems need.
Night classes for the next splicers.
Wiregrass Georgia Technical College is running a 32-hour Certified Fiber Optic Technician course Oct 20 to 25 at its Ben Hill-Irwin campus. Classes are on weekday evenings, with hands-on days over the weekend, and they cover installation, splicing, termination, testing and maintenance. The course is aimed at newcomers and at working professionals adding skills.
Field angle
The glass is getting denser, and the software around it keeps getting smarter. Multi-core fiber now has a standard, carrier labs are teaching AI to configure access gear on its own, and robots are learning to work patch panels. Look at what each story has to work around, though: wrong labels, out-of-date records, cabling too tight for a robot arm. Even a robot has to wiggle a fiber to make sure it has the right one. Every one of these machines depends on the physical plant being built and recorded right, and that work still belongs to trained people, which is why community colleges are running night and weekend classes to train more of them.
SpliceFlow is built by a working splicer for the crews doing that work. spliceflow.app
From X
@spliceflow asks: Multi-core fiber means several cores in one strand. What’s the first thing you’d want to know before splicing it? Reply on X.
The Next Span is written by the team at SpliceFlow. spliceflow.app
Lead story
A robot strings fiber on a live distribution line in Zhejiang
State Grid Zhejiang Electric Power has put what it calls China’s first wrapped optical cable on a distribution line into service, on the 10 kV Dongyun line in Jiashan, and a robot did the whole installation on its own. Ground crew send commands from a remote control system, and the robot carries the cable along the conductor at 1 to 60 centimeters per second, rotating as it goes to wrap the fiber around the line. When it reaches a crossarm or an insulator tie, built in sensors switch it into an obstacle crossing mode so it can get past without a person going up. The robot and its cable together weigh under 40 kilograms, and it uses vision, inertial and torque sensing to cross common pole types and adapt to different conductor sizes. The utility frames it as a way to add fiber to existing overhead lines where conventional aerial cable is hard and costly to build.
Source (China Energy News, May 25, 2026, in Chinese)
Quick hits
Street cabinets that patch themselves.
Orange Belgium piloted five Automated Service Cabinet robots in Ixelles, Brussels, with Netceed and Gigacom, then ordered 20 more after what Netceed calls positive results. The pitch is fewer manual interventions in street cabinets during the move from coax to full fiber.
A robot that lays sensing fiber on rail.
Researchers published a field case study in the Journal of Computing in Civil Engineering on a prototype robot that installs distributed fiber optic sensors along linear infrastructure, field tested on a rail application. They compared its speed and accuracy against an installation by human operatives, then tested both under train loading.
The AI buildout needs more hands, not fewer.
Business Insider reports that data centers built for AI need skilled fiber and low voltage technicians, citing a Fiber Broadband Association and PCCA report that nearly 200,000 additional fiber technicians will be needed to support that buildout. Meta has announced a program to recruit and train workers for the job.
The retirement wave is bigger than the hiring wave.
A Pew brief, citing an FBA and PCCA industry study, estimates about 30,000 new broadband technician jobs will need filling by 2032, plus about 64,000 more to replace technicians leaving the field over the next ten years.
Field angle
Robots are starting to take on the parts of fiber work that are slow, risky or hard to reach: crawling a live conductor, threading a duct, patching inside a cabinet. None of these stories removes the crew. Someone still sends the commands, preps the cable, works the closure and signs off on the run, and the people who know how are retiring faster than they are being replaced. The more machines join the build, the more each skilled splicer’s hours count, and the more it matters that their work is recorded right the first time.
SpliceFlow is built by a working splicer for the crews doing that work. spliceflow.app
From X
@spliceflow asks: Would you trust a robot to wrap fiber on a pole line you were responsible for? What would you check first? Reply on X.
The Next Span is written by the team at SpliceFlow. spliceflow.app
Lead story
Hollow core fiber goes live in central London
Colt, Network Planning Solutions (NPS) and Relativity Networks have completed a hollow core fiber proof of concept connecting two Colt network locations in central London, and the link is now live within Colt’s infrastructure. Instead of carrying light through glass, the fiber uses an air-filled core, which Colt says reduces latency by around 30%. NPS led the installation and validation. Its CEO said the team “successfully installed, spliced, tested and validated” the fiber inside a complex, established metro network. Colt expects hollow core to be used on the most performance-critical routes, such as links between data centers, AI clusters, cloud regions and trading locations, rather than replacing conventional fiber everywhere.
Quick hits
Orange Spain says its staff have built more than 1,000 AI agents.
In an expansion of its Google Cloud partnership (announced Oct 8), Orange Spain said employees have built more than 1,000 custom agents on Gemini Enterprise. Uses include sorting and routing technical incidents and analyzing system logs to isolate the source of network or service disruptions.
A fiber gateway that tries to diagnose itself.
At Network X, Airoha and MediaTek showed a 50G-PON gateway chip with an AI agent that runs on the gateway or ONU. A customer describes the problem in plain language, and the agent uses device data to find the cause and suggest or carry out a safe fix. Airoha says 50G-PON offers up to 5x the speed of XGS-PON (vendor claim).
Ordering interconnection by talking to an agent.
DE-CIX released a beta MCP server that lets AI agents find, order and manage its connectivity services from plain language instructions. Its CTO cited a DE-CIX survey finding that IT teams spend more than 11 hours a week troubleshooting connectivity.
Huawei pitches AI-led network upkeep, down to the fiber break.
At Network X, Huawei optical president Bob Chen outlined a plan to build networks around AI data centers and to run maintenance in three tiers. Software self-heals some client issues, engineers fix configuration faults remotely with diagnostic assistants, and crews sent out for damaged fiber use copilot software to locate the cable fault. Huawei says automated line adjustments cut optical line terminal energy use by more than 20% (vendor claim).
Field angle
Two threads run through this issue. New glass is arriving: hollow core fiber made it into a live metro network, and its own installers point out that a specialist crew installed, spliced and tested it. AI is moving deeper into network operations too, from gateways that check themselves to agents that triage incidents. Even Huawei’s model for automated maintenance ends with a person in a truck when the fiber itself is damaged; the software helps find the break, and a crew still fixes it. The more automated the network gets, the more the remaining field work is the part that can’t be skipped.
From X
@spliceflow asks: Hollow core is going in on the most latency-sensitive routes. Would you want to be on the crew that splices it? Reply on X.
The Next Span is written by the team at SpliceFlow. spliceflow.app
Lead story
Deutsche Telekom puts numbers on AI in network operations
At its first AI Investor Day (Oct 5), Deutsche Telekom said AI and automation should bring about EUR 2.5 billion in gross savings by 2030. It also said its “RAN Guardian Agent” cut response time to network strain events from several hours to about one minute.
Quick hits
T-Mobile uses AI to decide where to build and to ride out storms
T-Mobile’s CTO described AI that ranks where the next site build or upgrade matters most. He also described a self-organizing network that made more than 30,000 automatic adjustments during winter storm Fern.
Telekom Malaysia: AI agents run 65% of order-to-service, but not the install
TM says AI agents handle about 65% of its order-to-service processes, while work like sending a technician to install a fixed line “cannot be automated.” Its own network engineers built the agents, and no network staff were cut.
Openreach trial asks customers to “walk the route” before the crew arrives
In a proof of concept planned for around November, selected UK business customers would capture photos and video of the cable route and entry point, with some AI guidance, to cut repeat visits.
Nokia and MEO take AI into live network operations in Portugal
The two launched a joint program (announced Oct 14) testing AI for network management, monitoring and optimization in live conditions rather than in isolated test environments.
Robots walking data center rows to check fiber work
An inspection robot vendor reports autonomous rounds at a Houston data center since July, checking ports against the cable plan with a photo of each. It also notes that robots that plug, clean or reseat fiber are still in research and pilots. (This is the vendor’s own claim, from its blog updated Oct 14.)
Field angle
The pattern across these stories: planning, monitoring and routing decisions are getting automated fast, but the hands-on build still needs a crew. Carriers are pointing AI at where to build and how to react. The robots that exist today mostly inspect rather than build. For now, the people who plan, place and splice fiber stay at the center of the work, and the tools around them are what’s changing.
From X
@spliceflow asks: What part of your field day do you wish was automated? Reply on X.
The Next Span is written by the team at SpliceFlow. spliceflow.app
Get The Next Span.
The AI and robotics changing telecom and fiber. Free, in your inbox.