2G Shutdown in Hong Kong: The AI-Ready Network Play

China Mobile Hong Kong has a date on the calendar: 23 June 2026. That’s when it plans to switch off its 2G network, after receiving regulatory approval—making it the last of Hong Kong’s four major operators to retire 2G.

On paper, this looks like routine housekeeping. Fewer than 2.3% of China Mobile HK’s customers still rely on 2G, and the operator has already begun notifying them to upgrade SIMs and, where needed, handsets. But the real story isn’t the tiny tail of legacy users.

The real story is this: every 2G shutdown is a forced modernization project, and modernization is where AI becomes practical—not theoretical. If you’re leading network operations, technology strategy, or customer migration programs, Hong Kong’s timeline is a clean example of what the next 12–24 months can look like when a market finishes the long goodbye to legacy radio.

Why 2G switch-offs matter more than they look

A 2G sunset isn’t just “turning off old kit.” It’s a structural move that changes how you plan spectrum, operate radio sites, manage customer risk, and invest in automation.

Hong Kong shows the pattern clearly:

• 3 Hong Kong ended 2G in September 2021
• SmarTone ended 2G in October 2022
• HKT received approval to end 2G in November 2024
• China Mobile HK will end 2G on 23 June 2026

Meanwhile, China Mobile HK has already proven it’s willing to be aggressive on legacy retirement: it shut down 3G in April 2025, when only 0.25% of its base was still using it.

The operational truth: legacy networks drain attention

Even when only a small fraction of customers use 2G, the network still consumes:

• Site space and power (and those costs don’t scale down neatly with traffic)
• Field maintenance and spares management
• Vendor support contracts for aging components
• Operational complexity, especially for troubleshooting and change control

Most companies get this wrong: they treat 2G retirement as a compliance task and miss the bigger prize—simplifying the network so AI can actually run it well.

AI systems perform best when the environment is observable, consistent, and instrumented. Legacy layers often aren’t.

The bigger prize: spectrum and capacity for 4G/5G growth

The cleanest business case for a 2G shutdown is not “saving money.” It’s re-allocating spectrum and engineering time toward 4G and 5G.

GSMA Intelligence data (as cited in the source article) puts China Mobile HK at 8.5 million mobile connections by end-September (largest in Hong Kong), followed by:

• 3 Hong Kong: 5.5 million
• HKT: 4.8 million
• SmarTone: 2.9 million

China Mobile HK’s base is already modernized: just over half are on 4G, with the remainder on 5G. That means the network roadmap is no longer about “whether 5G.” It’s about how to run 4G/5G efficiently, meet quality expectations, and monetise premium performance.

Why AI and 5G capacity planning are inseparable

As networks mature, the constraint shifts:

• Early stage: coverage, rollout speed
• Later stage: cost per delivered gigabyte, energy consumption, and service differentiation

That’s where AI-driven network optimization becomes a serious toolset. With 2G out of the way, operators can standardize monitoring pipelines and apply AI across:

• Traffic forecasting (cell-level and cluster-level)
• RAN parameter optimization (reducing manual tuning)
• Energy savings (sleep modes, carrier management)
• QoE prediction (seeing issues before customers complain)

Put bluntly: a simpler radio estate makes automation cheaper and more accurate.

Where AI pays off during legacy shutdowns (not after)

The mistake is waiting until the old network is gone to “start the AI journey.” The migration itself is where AI can remove cost and risk.

1) AI-assisted customer migration: prevent churn, reduce calls

During a 2G shutdown, the highest-risk customers are often the least engaged:

• users with old feature phones
• elderly subscribers
• M2M/IoT devices installed years ago (alarms, lifts, payment terminals)

AI can help you treat migration as a targeted operations program instead of a generic SMS blast.

Practical plays I’ve seen work:

• Propensity-to-upgrade models to prioritize outreach
• Churn-risk scoring to trigger retention offers early
• Next-best-action recommendations for contact center agents
• Natural-language routing that detects “my phone stopped working” patterns and fast-tracks resolution

If you’re trying to generate leads for AI in telecom, this is one of the strongest entry points because the ROI is measurable fast: fewer inbound calls, fewer truck rolls, fewer failed migrations.

2) Predictive maintenance: keep the old network stable until the day it ends

Here’s the tension: you want to spend as little as possible on 2G, but you can’t let it fall apart while customers are still on it.

AI-based predictive maintenance helps you run that tightrope.

It typically involves:

• collecting historical alarms, performance counters, and field repair logs
• training models to predict site degradation and component failure
• prioritizing interventions where customer impact is highest

Even with a small 2G user base (like China Mobile HK’s <2.3%), the reputational damage from a messy shutdown period is real. Predictive maintenance keeps the legacy layer “good enough” without over-investing.

3) Decommissioning analytics: the overlooked cost sink

A shutdown creates thousands of micro-decisions:

• What hardware is reusable?
• What needs certified disposal?
• Which sites can be consolidated?
• Which backhaul links can be resized?

AI can speed up decisions using optimization models over inventory, site attributes, and demand forecasts.

A “snippet-worthy” way to think about it:

Decommissioning is supply chain plus network planning plus customer care. AI is one of the few tools that touches all three.

A practical 2G shutdown blueprint for telcos (12–18 months)

If you’re planning a 2G sunset—or finishing one—this is the operating model that tends to reduce surprises.

Phase 1: Get brutally clear on who still needs 2G

Start with a single source of truth that merges:

• subscriber/device capability (handset and SIM)
• last-seen RAT (2G/3G/4G/5G)
• roaming dependency
• enterprise/M2M contract details

The goal is not a dashboard. It’s a list you can act on.

Deliverable: a live “2G dependency register” refreshed weekly.

Phase 2: Segment migrations (don’t treat everyone the same)

Create at least four migration tracks:

1. Consumer feature-phone users (handset replacement + simple plan)
2. Smartphone users stuck on old SIMs (SIM swap)
3. M2M/IoT endpoints (field upgrade windows + coordination)
4. Edge cases (roaming, special devices, accessibility needs)

AI helps by ranking which customers belong where, and predicting who will fail migration without human intervention.

Deliverable: migration playbooks per segment with clear KPIs.

Phase 3: Manage the network like a runway, not a cliff

Operators often under-communicate the network plan internally. Then NOC teams get surprised when parameters change.

Run 2G retirement as a controlled runway:

• freeze unnecessary change
• tighten monitoring thresholds
• use AI to identify weak cells and top complaint geographies
• gradually re-farm spectrum only when a cluster is “clean”

Deliverable: cluster-based shutdown schedule tied to adoption metrics.

Phase 4: Prove the post-shutdown gains (finance will ask)

You’ll be asked what you got for the effort. Be ready with metrics that matter:

• spectrum re-farmed (MHz) and where it was re-used
• reduction in legacy trouble tickets
• reduced energy use at sites (kWh)
• reduction in vendor/support costs
• QoE improvement in the re-farmed footprint

This is also where AI-driven network optimization can show momentum: better forecast accuracy, fewer incidents, faster recovery.

People also ask: Will 2G shutdowns hurt IoT?

Yes, if you don’t manage the device estate early. A lot of “quiet” IoT is 2G by default because it was cheap and ubiquitous.

Operators and enterprises typically have three options:

• move devices to LTE-M / NB-IoT where available
• move to 4G modules for higher bandwidth needs
• redesign connectivity using multi-bearer modules (more expensive, less migration risk)

AI helps most in the discovery phase: identifying which endpoints are still active, where they live, and which ones are likely to fail when 2G disappears.

What Hong Kong signals for the next wave of network modernization

Hong Kong’s operators are finishing 2G because the economics no longer justify the complexity. That logic is spreading across markets: fewer legacy users, higher opportunity cost for spectrum, and more pressure to run networks efficiently.

For the AI in Telecommunications series, this is a useful anchor story because it shows where AI becomes a practical operations layer:

• AI isn’t “a lab project” when you’re migrating real customers
• AI isn’t “nice to have” when you’re predicting failures on aging equipment
• AI isn’t “future tech” when your 5G experience depends on how well you re-farm and optimize

If you’re responsible for network transformation, don’t frame your 2G switch-off as the end of something. Treat it as the moment you finally get to simplify the estate—and build an AI-ready operating model that scales.

The next question worth asking on your side isn’t “When do we shut down 2G?” It’s: Are we using the shutdown to build the data, processes, and automation discipline we’ll need for 5G optimization?