IIoT makers should treat 2G and 3G shutdowns as a product-portfolio issue, because more than 50% of cellular IoT devices in operation used those networks in the cited market data, while operators are shifting spectrum toward 4G, 5G, and purpose-built LPWAN services [S1].
The migration must cover module certification, carrier coverage, SIM and subscription dependencies, firmware, security, battery life, data volume, field replacement, and customer notification rather than simply changing a radio label [S1][S3].
Build a Device and Network Exposure Inventory
AT&T's migration white paper identifies more than 50% of cellular IoT devices in operation as 2G or 3G devices, with the installed population described as approaching 64 million [S1]. The first planning task is therefore a physical and logical inventory that records the modem model, radio technology, carrier, country, SIM identifier, firmware version, use case, data rate, power budget, enclosure, antenna configuration, and expected service life.
The inventory must extend beyond active product models. It should include devices already sold, units operating in customer facilities, devices in storage, replacement spares, and hardware embedded in machines that may remain in service for 10 years or more [S8]. Embedded 2G transmitters can remain hidden until they become inoperable, so asset discovery should include imported, privately labeled, and machine-builder equipment rather than only devices visible in the manufacturer's own BOM [S5].
Shutdown exposure differs by geography and operator. Onomondo’s 2025 article provides an up-to-date list of operator 2G and 3G sunset dates.
Match Each Workload to a Replacement Radio
The main replacement choices are LTE-M, NB-IoT, and LTE Cat 1 bis, with conventional LTE, 4G fallback, and 5G serving applications that need more throughput or wider carrier availability [S2][S3]. LTE-M is identified as Cat-M1 and operates within 4G LTE, while both LTE-M and NB-IoT are cellular LPWA technologies designed to provide greater coverage and capacity with lower power consumption than legacy networks for IoT deployments [S3].
A decision matrix should compare radio options against four controls: coverage at every deployment point, battery consumption at the required reporting interval, data transfer rate, and the availability of fallback in priority countries. LTE-M is suited to mobile assets and more demanding LPWA requirements, while NB-IoT targets constrained devices with low data rates and long battery lives.
Global deployments add a fourth dimension: roaming and certification. A module that performs well in one country can fail because a required LTE-M or NB-IoT band, carrier approval, roaming agreement, or local SIM process is missing [S8].
Full LTE or 5G should be reserved for gateways, cameras, voice equipment, or other workloads whose data and latency requirements exceed constrained LPWA capabilities. A 5G product does not automatically solve a low-power IoT problem, while replacing a working 2G terminal with a high-power cellular radio may shorten field life and invalidate the original enclosure and battery design.
Design for Reuse, Fallback, and Testability

Legacy-device reuse can reduce migration cost, but it must be proven rather than assumed. Migration can be more complex than a new IoT deployment because existing hardware, application protocols, bandwidth assumptions, and carrier integrations may not transfer directly to LTE-M, NB-IoT, Cat 1 bis, or 4G [S1]. Each retained design should pass radio, antenna, power, backend, and end-to-end tests under the intended carrier and firmware combination.
Software should contain as much network-dependent logic as practical instead of hard-coding operator behavior into the terminal. The recommended architecture keeps logic in the network, not on devices, reducing vendor lock-in and making operator, protocol, and routing changes more manageable [S2]. Firmware should also support staged remote updates, rollback, certificate renewal, and configuration changes before the carrier's final shutdown notice [S8].
For devices with long service lives, hardware abstraction can reduce replacement cost. A common application interface can sit above interchangeable network modules, while the product's data contract remains stable. This is useful where operators are turning off GSM or UMTS in stages and a SKU may need to move first to Cat 1 bis, then to LTE-M or NB-IoT without redesigning the cloud application.
Manufacturers must also test the physical layer. LTE-M and NB-IoT can operate in different frequency bands, and radio certification, antenna performance, enclosure losses, and local carrier approvals remain deployment-specific [S3]. A module datasheet is not proof that a finished device will maintain coverage, registration reliability, and acceptable battery life at the intended site.
Control Data, Power, and Lifecycle Economics
Data-transfer design should be reduced to the smallest useful payload, using compression, event thresholds, batching, acknowledgments, and application-layer rules rather than sending constant full-size records [S2]. The measurement requirement should include message size, reporting frequency, session establishment behavior, retransmissions, and expected daily data volume for each application class.
Power consumption must be recalculated for the replacement radio and reporting pattern. LTE-M and NB-IoT were designed for IoT operation with lower power use and longer battery life, but those benefits depend on the selected radio, network behavior, signal conditions, and device activity cycle [S3]. Field validation should include weak-signal operation, because repeated retries or fallback scans can materially change an original battery-life estimate.
The business case should compare the full migration cost rather than module price alone. It should include engineering hours, certification, carrier fees, SIM logistics, field technicians, shipping, packaging, customer notifications, production changeovers, and loss of service during replacement. AT&T notes that replacing a 2G or 3G IoT solution can resemble a new deployment because original installation personnel may no longer be available [S1].
Product exit rules should define the point at which a legacy SKU stops being sold. A defensible policy requires a named shutdown date from each target carrier, a verified successor radio for each deployment country, a completed field-pilot record, and an approved customer communication package before the legacy product enters an end-of-sale state.
Identify the Intended Users and Exclusions

Long-life, low-data applications require the most disciplined planning because testing and physical replacement take time. Asset tracking, environmental monitoring, smart metering, and telematics are explicitly identified as exposure areas for phased-out 2G and 3G networks [S2]. These applications should enter the migration queue before high-volume consumer products because their installed units are less likely to be refreshed through normal purchasing cycles.
The approach is suitable for manufacturers with cellular devices already in the field, operators or integrators coordinating fleet migrations, and long-life equipment providers that cannot replace every terminal at once. It is also appropriate where a product must move from 2G or 3G to LPWAN while retaining the same sensor, enclosure, installation process, and application backend.
The approach is not a substitute for redesigning devices that exceed LPWA data limits, require deterministic ultra-low latency, or need 5G performance. It is also insufficient where battery replacement is impossible and no tested low-power successor meets the application requirements. Those products need a hardware revision, a non-cellular communications option, or a negotiated service extension with the carrier.
Protect Customers Through a Documented Exit Process
Customer-facing products require more than a silent hardware swap. The manufacturer should warn customers before service interruption, explain whether hardware replacement is required, provide installation or return procedures, and define how data will transfer from the retired device [S1]. A staged program can begin with high-risk countries, continue through pilot deployments, and finish with a production cutoff and field replacement campaign.
Publicly listed UK operator schedules illustrate the need for version control rather than a global switch-over date: the 2025 table lists Vodafone 2G support through 2030, EE through 2029, O2 UK's 2G date as to be confirmed, and no 2G service for Three, while the corresponding 3G dates differ across those operators [S4]. The same operators had agreed not to support either 2G or 3G beyond 2033, which makes 2033 a planning horizon rather than permission to defer migration to that year [S4].
Manufacturers should preserve a traceable bill of materials for each radio variant, including module, antenna, SIM profile, firmware, bootloader, and cloud dependencies. Any component substitution should trigger regression testing. A product that passed with one module family cannot automatically be assumed compliant after a different LTE-M, NB-IoT, or Cat 1 bis module changes its power behavior, RF characteristics, or certification footprint.
Set Trackable Migration Gates

The first measurable gate is inventory completeness: every active cellular SKU and every installed unit should have a mapped carrier and country, with legacy exposure assigned an owner. The second gate is design qualification: each retained application must have a tested successor radio for its required market, with coverage, power, data-volume, and fallback results documented. The third gate is operational readiness, including remote provisioning, firmware rollback, customer notices, spare stock, and technician procedures. [S4]
Two trackable signals should drive escalation: an operator issuing a 2G or 3G shutdown notice for a market where a SKU remains deployed, and a firmware or module change invalidating an earlier field-pilot result. Both events should reopen the risk record and move the affected product into the field-replacement program rather than waiting for the network to go dark.