Quick Answer: Private 5G network operations are the ongoing work that keeps a factory network running after the integrator’s project team moves on: spectrum and grant records, SIM and device onboarding, software upgrades, fault and performance management, coordination with operational technology (OT) teams, and security monitoring. Annual costs can match or exceed the original build, and vendor exits and sales can leave a live network on a retired or newly owned platform. Integrators that price this work as a managed service can turn it into recurring revenue. A plant keeping the work in-house needs a named owner for each workstream. AI tools already help with fault diagnosis and anomaly detection, while autonomous radio tuning remains mostly lab work.
In September 2026, Simon Bryant, who leads the research team for the Global mobile Suppliers Association (GSA), cited independent research putting typical capital spending on a dedicated private network at around $150,000 and annual operating costs at $150,000 to $200,000, according to Light Reading. He said the figures vary widely by sector and scale. Even at the low end of his range, one year of operations costs as much as the build.
Eight months after go-live, a factory network collects jobs no one scoped. The core vendor publishes a release that carries a security fix. Operations orders a second fleet of autonomous mobile robots (AMRs) whose modules need SIM profiles and quality-of-service settings before they leave the dock. Overnight, the Spectrum Access System (SAS) suspends a grant on one radio, and the plant’s IT team hears about it from a robot that stopped mid-route.
By then, the integrator’s project engineers are on another customer’s build. They left a handover binder on how the network was put together, with little on who handles a week like this one, or what that week costs.
Private 5G Network Operations Run on Six Workstreams
A build follows a plan with an end date. Operations work arrives on triggers such as a vendor release, a new device model or a production layout change, and it draws on skills that rarely sit in one team. AI tools now target parts of this work, one more reason to split it into workstreams before assigning owners.
On a factory network, the work falls into six workstreams:
• Spectrum and grant management. In the US, a radio on Citizens Broadband Radio Service (CBRS) spectrum transmits under grants from a SAS, which can suspend them when federal incumbents need protection. In Europe, local licenses bring their own terms and renewal dates to track instead.
• SIM, eSIM and device lifecycle. Profiles for each device type, module testing before a new model joins the network, and credentials retired when assets leave.
• Radio and core software lifecycle. Upgrades and security patches for the radio access network (RAN) and the mobile core, tracked against each vendor’s support dates.
• Fault and performance management. Alarms, key performance indicators (KPIs) and root-cause work when a robot, scanner or camera drops off.
• Change management with OT. Maintenance windows set by production, and radio retuning when a moved line or new steel racking changes coverage.
• Security monitoring. An inventory of the devices behind each cellular router, and segmentation that keeps IT and OT traffic apart.
Radio retuning recurs on floors that get reconfigured. NTT East, which ran a multi-vendor private 5G test program with 26 companies, wrote in March 2026 that parameter tuning lets private 5G fit many use cases, but the operational burden of that tuning remains a challenge. Each workstream has its own triggers and skills:
| Workstream | Recurring tasks | Triggers | Skills |
| Spectrum and grant management | Grant monitoring, registration records, license renewals | Incumbent activity, new or moved radios | Spectrum rules, SAS administration |
| SIM, eSIM and device lifecycle | Profile provisioning, device testing, credential retirement | New device models, departing assets | Subscriber management, device knowledge |
| Radio and core software lifecycle | Upgrades, patches, support-date tracking | Vendor releases, security advisories | RAN and core administration |
| Fault and performance management | Alarm triage, KPI review, root-cause analysis | Dropped devices, coverage complaints | RAN and core troubleshooting |
| Change management with OT | Maintenance scheduling, retuning after layout changes | Line moves, new equipment, shutdowns | Radio planning, OT coordination |
| Security monitoring | Device inventory, segmentation review, anomaly checks | New devices, policy changes, alerts | IT and OT security |
An integrator or plant team inheriting a network it did not build gets the most from week one with an inventory in this order:
• Spectrum records: CBRS registrations and installer details, or local license documents, plus whose name each sits in.
• Subscriber records: a database export from the core and a SIM inventory matched to physical assets.
• Software versions: radio and core releases checked against each vendor’s published support dates.
• Configuration backups: current copies the plant can reach without the previous operator.
• Support contracts: open tickets, contract terms and who is allowed to open a case with each vendor.
| PRO TIP: Spectrum Records Come First in a TakeoverRegistrations and license documents are hard to reconstruct once the people who filed them have left, and CBRS adds a timing constraint on new grants. Google is one of the commercial SAS administrators. In its SAS documentation, Google says the daily coordination window between the SAS platforms runs from 7 AM to 10 AM UTC, that no new grants are issued during it, and that maintenance needing new grants should finish before it begins. A radio swap scheduled inside that window can leave the new radio dark for up to three hours. |
Who Should Own Each Workstream
Handover assigns responsibilities once, and each trigger in the table can test them again. Six decisions, one per workstream, hold up better than a contract line saying IT owns the network.
Factory networks run today under four broad operating models: in-house cellular teams, integrator managed services, vendor or cloud-managed platforms, and inherited operations, where a vendor sale or provider exit hands a running network to someone new.
Asad Khan, 5G research director at SNS Telecom & IT, told Fierce Network in June 2025 that Airbus, John Deere and Tesla had invested in hiring engineers to manage multi-site deployments, and that managed services and network-as-a-service (NaaS) offerings can ease the burden for smaller enterprises. For a single plant, that burden starts with the cellular skills gap between Wi-Fi experience and cellular operations.
Managed services are where an integrator’s expanding role turns into recurring revenue. NTT DATA announced a fully managed private 5G service at two Celanese plants in Texas in September 2025. Acting as integrator, NTT DATA also scaled Cargill’s private 5G deployment, built on a Celona system, from 20 test sites to 50 commercial deployments in the US and Europe within six months, RCR Wireless News reported in February 2026. Cargill’s Robert Greiner said at the UPTIME event that the first site took four years.
Cloud-managed platforms carry an exit risk of their own. Amazon Web Services listed Private 5G as having reached end of support in a May 2025 AWS service update.
The inherited model is about to play out at scale. On October 13, 2026, Nokia agreed to sell its Enterprise Campus Edge (ECE) business, with nearly 900 customers, to the private equity firm AURELIUS. According to RCR’s follow-up report, the Digital Automation Cloud platform moves with ECE, the deal is expected to close in the first half of 2027, and some contracts will likely need re-signing.
Whichever model a site runs, a workable split comes from five steps in this order:
• Workstream map: the six workstreams listed against the site’s actual vendors and contracts.
• Cloud dependency check: a mark on each workstream that relies on a vendor’s cloud or subscription.
• Named owners: IT, OT or the integrator for each workstream, recorded as a team and a person.
• Escalation paths: who calls which vendor, and who can approve a change during production hours.
• Review triggers: a vendor sale, a new site or a new device class reopens the split.
A sensible default gives IT the core and SIM records and leaves OT to decide priorities, maintenance timing and what counts as an outage, since production carries the cost when a machine stops.
| PRO TIP: A Managed Service Contract Needs a Handback ClauseExit terms work best when they list what returns to the plant at contract end: a subscriber database export, current configuration backups, and spectrum registrations or licenses in the enterprise’s name. Without those items, the next operator starts by rebuilding them. |
| Follow How Private Networks Run After Go-LiveGet deployment coverage and analysis of how plants and integrators run private networks after the build. Subscribe to PrivateLTEand5G |
AI-Assisted Operations: What Ships, What Is Announced, What Is Still in the Lab
AI for network operations now shows up in vendor platforms, SAS software and research programs, and product announcements arrive faster than production evidence. Buyers plan better when the two are kept apart.
Across the six workstreams, the evidence falls into three groups:
• Shipping: root-cause analysis, anomaly detection and interference modeling in the SAS.
• Announced: AI troubleshooting that recommends remediation, bundled with a platform rolling out from late 2026.
• Lab or grant-funded: closed-loop radio optimization and agent-based operations.
OneLayer says its AI does the most work today in root-cause analysis and anomaly detection, explaining connectivity failures in plain language and flagging devices that drift from normal behavior. OneLayer sells security software for private cellular networks.
For spectrum, Federated Wireless introduced Spectrum AI in June 2026, an AI-driven SAS upgrade that models propagation, interference and spectrum assignments, with general availability planned for mid-July. The company reports up to 5 times more network capacity across live commercial CBRS deployments and aims the first release at large and mid-sized operators. Federated Wireless is also a SAS administrator.
Brain, Celona’s AI troubleshooting layer, works from live network data and returns a verified root cause and recommended remediation to the engineer, according to Celona. It comes with Orion, the converged private 5G and Wi-Fi platform that Celona announced in August 2026 for availability from the fourth quarter, with key capabilities introduced in phases. Celona is a private wireless equipment vendor.
NTT East has tested autonomous radio tuning across vendors with a RAN Intelligent Controller (RIC). During the tests, the controller read traffic levels and adjusted the RAN, cutting power consumption by about 65% under certain conditions. The Aerix-led MOSAIC-5G consortium in the UK plans to add AI-agent network operations to a unified management platform under a grant-funded program running from August 2026 to January 2028.
Mapped to the workstreams, the evidence looks like this:
| Workstream | Where AI shows up | Evidence status |
| Spectrum and grant management | SAS interference and assignment modeling (Federated Wireless Spectrum AI) | Shipping, aimed at operators |
| SIM, eSIM and device lifecycle | No AI product found | Not found |
| Radio and core software lifecycle | No AI product found | Not found |
| Fault and performance management | Root-cause analysis, anomaly detection (OneLayer); troubleshooting with recommended fixes (Celona Brain) | Shipping (OneLayer); rolling out from Q4 2026 (Celona) |
| Change management with OT | Traffic-based radio adjustment through a RIC (NTT East); none found for layout retuning | Lab |
| Security monitoring | Device behavior baselines and anomaly flags (OneLayer) | Shipping |
Testing an AI operations pitch against a plant’s own data takes five steps, in order:
• Scope: the workstream the tool claims to cover, from the table above.
• Incident set: last quarter’s tickets for that workstream, with their known root causes.
• Replay: the vendor running the tool against those incidents on site data.
• Action mode: whether the tool recommends a change or makes it, and who approves.
• Coverage: confirmation that the tool reads data from every supplier’s equipment on site.
| PRO TIP: Last Quarter’s Tickets Beat a Vendor Demo as an AI BenchmarkA plant’s own incident history comes with known root causes, so a replay shows whether the tool would have found them. Tools that start in read-only mode, explaining faults and suggesting fixes, build a record the operations team can judge before any agent receives permission to change a live network. |
Where Does This Leave You?
If you’re a systems integrator or vendor, the years after go-live can hold more of a factory account’s value than the build, but only when operations are priced the way the work arrives. Quote private 5G network operations per workstream, each with a number the plant can track: restore time when a robot or scanner drops, days from a vendor’s security release to an installed patch, turnaround for onboarding a new device model, and recovery time after a grant suspension. When a customer asks about AI, answer with the evidence status for each workstream, since plants that watched AWS retire its private 5G service have reason to weigh roadmaps carefully. Integrators that run factory networks also hold operating data few buyers see, and the Executive Voice Program and Partner With Us put that experience in front of the plant IT and OT teams who read this platform.
If you run IT or OT at a plant with a private network, name an owner for each of the six workstreams before the next software notice arrives, and write down who signs off on changes during production shifts. Track the same four numbers you would ask an integrator to report, whoever runs the network. Check every vendor’s support dates and contract assignment terms, especially if your platform is changing hands, and ask any AI operations vendor which workstream its tool covers and whether it recommends or acts. The 10-question wireless self-audit gives a scored read on your facility’s current wireless setup before any sales conversation. None of this is as visible as the build, but it decides whether the network still works in its third year.
FAQ
Can a plant’s Wi-Fi team handle private 5G network operations?
Switching, firewall rules, segmentation and device inventory carry over from Wi-Fi work. Cellular radio planning, mobile core upgrades, SAS grants and SIM provisioning are new, and a Wi-Fi team taking them on needs training or vendor backup through at least its first upgrade cycle.
What happens to a running network if its vendor is sold or exits?
A network with an on-premises core usually keeps running through a sale, while patches, support cases and cloud-hosted functions carry the risk. A full exit forces a move to a new core, and SIMs whose authentication keys can’t be exported to it have to be replaced or reissued.
Is a managed service cheaper than hiring cellular engineers?
It depends on what the contract covers. A fee that leaves out after-hours response, spectrum administration or new device onboarding pushes that work back to the plant, so pricing both options per workstream gives a fairer comparison.
Should AI agents be allowed to change a live factory network?
Small, reversible changes are fair candidates once the tool has a track record on that site. Work on the core, cells serving AMR fleets and anything during production hours should keep human sign-off, and every automated change needs a rollback the OT team can trigger.
