What mobile network reports don't tell you about IoT

Here are six things worth checking before choosing a network on the strength of a speed-test table.

1. Low-power network footprint, not just 4G and 5G speed

Download speed matters for a dashcam or a video telematics unit. It matters much less for a smart meter, an environmental sensor, or an asset tracker sending a few kilobytes a day. Those devices typically run on NB-IoT or LTE-M (Cat-M1), low-power wide-area protocols built for multi-year battery life and better indoor and underground penetration than standard 4G. A network's consumer speed ranking says nothing about how complete its NB-IoT and LTE-M rollout is, and that rollout is entirely up to the underlying network, not the SIM. CommsCloud's CloudConnect™ SIMs give a device access to whichever radio technologies a given network supports at that location, including NB-IoT and LTE-M where the network provides it, but the coverage itself is set by the operator's own rollout, not by CommsCloud. Worth checking with any provider which specific technology is live on your device's actual route, not just which brand of SIM sits inside it.

2. What happens to the tower when the power goes

A network's average speed is measured on towers that were up and running during the test. It says nothing about what happens when they are not. South Africa's mobile operators have spent years, and by their own accounts billions of rand, building backup power at base stations: batteries, diesel generators, solar, with hub sites often engineered for 8+ hours of standby. Load shedding itself has eased sharply since mid-2025, with Eskom logging 365 days without outages by May 2026, but operators including MTN have said publicly that backup power investment continues regardless, because grid instability, severe weather and infrastructure theft at sites remain ongoing risks independent of load shedding specifically. None of this shows up in a speed-test ranking, and backup posture varies by operator and by site. Worth asking directly rather than assuming a "best network" score implies "most resilient during an outage".

3. Latency and jitter, not just raw throughput

A network offering a stable 5 Mbps at 20 ms latency beats a 100 Mbps connection with erratic jitter and dropped packets, for real-time telematics, remote monitoring, or anything sending time-sensitive data. Speed-test rankings lead with download Mbps because that is what a consumer notices loading a page. An IoT buyer should be asking about latency consistency and packet loss under load, not peak throughput.

4. Coverage on the corridor, not just the city

Consumer drive tests are weighted toward where the subscribers are, which means metros. Fleet and logistics deployments live on the N1, N2, N3 and the SADC border posts, where coverage looks nothing like a metro average. This is exactly the gap CommsCloud's corridor guides exist to cover: real-world continuity for Beitbridge, Lebombo and Ressano Garcia, Kazungula and Chirundu, rather than a national average that quietly assumes you are driving through Sandton.

5. Whether "five networks" is actually five networks

Some names in a network ranking are not independently built infrastructure. A virtual network operator riding another operator's radio access network shows up as a separate line in the table, with its own score, even though a chunk of its performance, and its failure modes, trace back to the network it is hosted on. That is a genuine consideration for redundancy planning: if a failover strategy assumes operator diversity, it is worth checking whether that diversity is real at the radio layer, not just at the billing layer. This is precisely why CommsCloud's multi-IMSI SIMs are built across multiple independent core networks, not just multiple SIM profiles. See how autonomous failover actually works.

6. What is actually routing your data

Physical connectivity is half the picture. How that traffic is secured once it leaves the device is the other half. A private APN with an IPsec VPN tunnel and static IP allocation keeps device traffic off the public internet entirely; a consumer SIM does not. None of this shows up in a speed-test comparison, and it is usually the first thing a security team asks about once a pilot moves to production. See CommsCloud's Private APN and VPN.

What this does not mean

None of this makes the network-quality reports wrong. They measure what they say they measure. It just means a "best network" headline answers a consumer's question, not an IoT deployment's. The right network for a fleet, a sensor estate, or a security camera network is the one that is live on the actual route, resilient when the power drops, secure end to end, and, if resilience really matters, backed by more than one truly independent core network.

The numbers this piece sits next to are in the Q2 2026 best and worst mobile networks refresh.