The most dangerous thing about network sunsets is that they arrive on someone else’s schedule. Operators worldwide are retiring 2G and 3G networks to refarm spectrum for 4G and 5G, on timelines that vary by country and operator and shift as plans firm up. Publishing a date table here would be out of date within months, which is precisely the point: estates that plan around specific dates are planning around a moving target, while estates that plan around their own devices control the only variables they actually own.
This piece is therefore a framework for the estate side of the problem: what you have, what it should move to, and in what order, so that whenever a relevant network goes quiet, your devices stopped caring some time earlier.
Why this cannot wait for the dates to firm up
Three properties of the problem reward early planning regardless of local timelines. Migration lead times are long: hardware selection, certification, firmware, logistics and field work for a sizeable estate consume quarters, not weeks. Degradation precedes shutdown: as operators refarm spectrum, legacy network capacity and coverage can thin before any formal switch-off, so the experienced sunset arrives early and unevenly. And the field forces that perform physical swaps are a shared resource: every estate in a market migrating late competes for the same engineers at the same time, at the prices that scarcity implies.
The estates that fare worst in any sunset are not the ones with the most legacy devices; they are the ones that started counting them last.
Phase 1: Audit what you actually have
The foundation is an honest device census, and it usually surprises. For each device class, establish: the radio technology in use (genuinely 2G or 3G only, or 4G-capable hardware configured conservatively, a distinction that changes everything); the device’s function and the business cost of it going dark; physical accessibility and natural touchpoints (servicing cycles, battery replacement, planned refresh); remaining intended life; and the SIM situation, since a hardware migration is also the moment SIM strategy gets revisited.
Cross-reference the census against geography. The output is a risk-ranked register: devices by class, by market, by accessibility, by business criticality. This register, not any operator’s announcement, is the document the migration plan is built from.
Phase 2: Choose the destination technology per class, not per estate
The reflex answer to “what replaces 2G” is a single technology, and it is the wrong shape of answer. The right destination varies by device class.
LTE-M is the natural successor for most 2G-era IoT: low power consumption, mobility support, voice capability where needed, and bandwidth comfortably above legacy rates. Devices that move, devices that need over-the-air updates, and battery devices with periodic reporting mostly land here; our complete guide to LTE-M and the companion migration piece cover the deployment detail.
NB-IoT suits the static, minimal-data end of the estate: meters, sensors and monitoring points where building penetration and battery life dominate and mobility is irrelevant.
4G LTE (Cat-1 and above) is the destination for device classes whose data needs outgrew the legacy networks anyway, and for markets or use cases where the low-power technologies’ availability does not match the deployment map, which is a coverage question to validate per market rather than assume.
Technology selection per class should be validated against actual network availability in each deployment market, since LPWAN rollout varies, and against module economics, which have shifted substantially in recent years. This is also the moment to make the longevity decision consciously: selecting hardware and SIMs (eUICC-capable, multi-network) that will not need this exercise repeating when the next generation turns over.
Phase 3: Sequence by risk, not by convenience
With the register and destinations in hand, sequencing follows the same wave logic as any estate migration: start where learning is cheapest, finish where stakes are highest, and let natural touchpoints do free work throughout.
A workable default order: new deployments switch to target technology immediately (stop digging); accessible, low-criticality legacy devices form the proving waves; devices with natural touchpoints inside the planning horizon migrate at those touchpoints; high-criticality devices migrate once the playbook is proven; and the genuinely inaccessible tail gets an explicit decision per class, migrate at cost, replace at end of life, or consciously accept the risk window, rather than a default of silence.
Each wave needs the standard discipline: verification criteria, a monitoring window, and a rollback or pause decision point. Our broader guide to switching connectivity providers describes the wave mechanics in detail; a sunset migration is the same machinery with a harder deadline.
Phase 4: Use the migration to fix what the last decade taught you
A forced estate touch is expensive; touching every device twice is more so. The migration is the rational moment to consolidate the other lessons of a deployed decade: SIM strategy (multi-network access through Multi-IMSI with non-steered selection, so the estate’s resilience stops depending on any single network’s planning decisions), eUICC capability (so the next strategic change happens over the air instead of via ladder), firmware hygiene (configurable rather than hard-coded connectivity parameters), and consumption visibility per device from day one.
This is where OV’s position is relevant rather than incidental. Migrating estates onto OV means landing on a true IoT MNO with coverage across 180+ countries and 600+ networks, LTE-M and NB-IoT support, eUICC and SGP.32 alignment, and the OV ONE platform for running the migration itself, bulk provisioning, per-SIM visibility and API automation are precisely the tools a wave-based migration consumes. We work through sunset migrations with customers against their real estate registers, and a free SIM trial against your hardest sites is the sensible first wave zero.
The cost of the alternative
It is worth stating the do-nothing scenario plainly, because it is a plan too, just an unpriced one: devices that stop reporting on a date you did not choose, in markets you learn about from support tickets, with replacement hardware ordered into lead times everyone else discovered the same week. Every element of the framework above exists to convert that scenario into scheduled, budgeted, boring work. Boring is the goal.
Frequently asked questions
When will 2G and 3G networks shut down?
On different dates in different countries, set by individual operators and subject to change, which is why this framework deliberately plans from the estate side rather than a date table. The planning assumption that serves estates best is that legacy capacity degrades before formal shutdown, so readiness should precede any published date rather than coincide with it.
What should 2G IoT devices migrate to?
It depends on the device class: LTE-M for most mobile, battery-powered or update-dependent devices; NB-IoT for static minimal-data sensors; 4G LTE for classes that need more bandwidth or operate where LPWAN coverage is incomplete. Selection should be validated per market against actual network availability rather than decided once for the whole estate.
How long does a sunset migration take?
For estates of any size, quarters rather than weeks once hardware selection, certification, logistics and field work are counted, with the inaccessible tail potentially running to natural end of life. The duration argues for starting from the estate audit now, regardless of how distant local shutdown dates appear.
Can the SIM strategy change during a sunset migration?
It should, because the estate is being touched anyway. Migrating to multi-network, eUICC-capable SIMs during the hardware transition means the estate’s resilience no longer depends on any single network’s roadmap, and the next strategic change can happen over the air rather than through another field programme.
