Are Britain’s Mobile Networks Really That Bad? What the Headlines Mean for Consumers, Businesses and IoT

Why everyone is suddenly talking about UK mobile

Two sets of numbers have driven the recent coverage, and both trace back to the same source.

Opensignal’s Q1 2026 European analysis covered 29 markets: the EU-27 minus Malta, plus the UK, Norway and Switzerland. The UK finished 29th out of 29 on the Global Network Excellence Index, with average 5G download speeds of 84 Mbps against front-runner Denmark’s 226 Mbps, and the lowest Excellent Consistent Quality score in the region at 80 per cent versus Denmark’s 93.1 per cent. UK 5G availability came in at 57.4 per cent.

Separately, the consumer group Which? analysed Opensignal data gathered from over a million handsets between May 2025 and April 2026 and placed the UK 57th globally for overall network performance, 70th for download speeds and 55th for consistent quality. That put Britain behind every EU member state and every other G7 nation. Which? also made a structural argument that deserves more attention than the league table did: UK coverage obligations are written around whether a connection is available, not whether it is any good once you use it. That allows operators to claim 99 per cent 4G coverage and meet their obligations while large parts of the country struggle to load a web page.

National press coverage has run further, including comparisons of London’s download performance against cities most readers would not expect to be ahead of a G7 capital.

None of this should be waved away. The UK has genuine infrastructure problems: planning friction, mast height and density, spectrum costs, years of consolidation, and demand growing faster than investment. Ofcom’s predicted coverage figures have long flattered the real-world picture. The reports are measuring something real.

The question for this audience is a different one. Does any of it tell you whether your telemetry estate is going to stay up?

What these benchmarks actually measure

Opensignal’s data is crowdsourced. It comes from an app running on consumer smartphones, measuring the things smartphone users care about:

• Download speed
• Upload speed
• Latency
• Video streaming experience
• Games experience
• Consistent quality, defined as the share of tests meeting a threshold sufficient for video calling and streaming

Those are good metrics. They are also, without exception, consumer metrics. Excellent Consistent Quality is a threshold test for whether a video call would have held up. It is not a measure of whether a device stayed attached to a network for eleven months.

There is a more interesting point buried in the methodology. The measurement population is handsets running an app. An IoT SIM in a router in a plant room does not run the Opensignal app. Neither does a smart meter, an alarm panel, a traffic controller or a gateway on a reservoir. The entire M2M estate is invisible to the benchmark. Whatever the reports are telling you, they are not telling you about IoT, because IoT was never in the sample.

The industrial estate is not silent in these rankings because it performs well. It is silent because it was never asked.

Different users, different metrics

Written down, it becomes obvious why a single league table cannot serve everybody.

Notice where the “No” column begins, and what everything below that line has in common.

Slow does not mean unreliable

This is the misconception the headlines create, and it is worth taking apart properly. Speed and availability are separate engineering properties. They are not correlated in the way people assume.

Network B wins every campaign ever written. For a fleet reporting 2KB every fifteen minutes, Network A is worth more. The sensor gains nothing from the 385 Mbps it will never use. It suffers badly from the outages.

A phone and an industrial router are not the same class of equipment

The reasoning most people arrive with is simple and wrong: my phone has one bar here, therefore a connected device will not work here.

A smartphone is an extraordinary piece of engineering optimised for constraints that have nothing to do with your installation. It has to fit in a pocket, survive a day on a small battery, and put its antennas wherever there is space left once the screen, camera stack and battery have taken theirs. Then someone holds it, detuning the antenna, and puts it in a pocket, attenuating it further.

An industrial Cat 4 router is designed against completely different constraints. It sits in a cabinet on permanent power, with SMA connectors on the outside. What you put on those connectors changes the physics.

Same site, same operator, same cell, two entirely different link budgets. The router is not on a better network. It is on a better antenna, at a better height, on permanent power, watching its own connection.

The SIM difference is worth dwelling on. A managed IoT SIM is not the one in your handset. Multi-network roaming, private APNs, fixed IP addressing, fleet management platforms, session diagnostics and usage controls all change the resilience equation before the radio is even considered. The handsets in the Opensignal sample have none of them.

Where 5G fits, and why NSA versus SA matters more than the number

Most of the current news cycle is a 5G story. Most of the UK IoT estate is not.

The overwhelming majority of deployed industrial connections run on LTE, and a large share of those on Cat 4 or below. Cat 1, Cat 1 bis and LTE-M carry a great deal of telemetry perfectly happily. Opensignal itself noted that the vast majority of connectivity time in Europe is still spent on 4G. A Cat 4 router topping out around 150 Mbps down is not a compromise for a telemetry application. It is roughly a hundred times the bandwidth required, in a modem generation with a decade of field maturity behind it.

Where 5G does become relevant, the architecture distinction is the part that matters and the part the headlines skip.

5G NSA (Non-Standalone) is what almost everyone in the UK is actually using when the 5G icon appears. The 5G radio sits on an existing 4G core and the device keeps an LTE anchor. You get more download throughput. You do not get the latency floor, the uplink improvements, network slicing or the low-power device features that 5G is sold on, because those live in the core, and the core is still 4G. A bigger download number, delivered by a 4G brain.

5G SA (Standalone) is end-to-end 5G. Slicing, deterministic latency, better uplink and proper massive-IoT support become available. In Q1 2026 the UK was barely into it: EE led on time spent on a 5G SA bearer at 8.3 per cent, O2 at 5.3 per cent, VodafoneThree at 4 per cent. Opensignal’s own observation was pointed, that the operators pushing hardest on SA tend to be the ones with the most ground to make up overall.

For IoT the practical reading is this. NSA gives you almost nothing you needed. SA eventually gives you things you might genuinely want, particularly uplink for video and slicing for applications that need a guarantee rather than a hope. But it is years from being something you can design a national estate around. In the meantime RedCap and eRedCap are the more interesting development for anyone specifying hardware, because they bring 5G-generation devices down to a cost and power profile that suits IoT rather than handsets.

Buying a 5G router today for a device that sends 2KB every fifteen minutes is buying a headline. The Cat 4 unit was already overqualified.

What actually causes IoT outages

Not download speed. In practice the list looks like this:

• Planned operator maintenance, usually overnight, usually unannounced to you
• Local mast power failure
• Backhaul failure, fibre cuts, cabinet damage
• Congestion at specific places and times: stadiums, city centres, events
• SIM steering behaving differently to the way the datasheet implied
• PLMN and FPLMN state after a failed attach, leaving the device parked on a network it cannot use
• APN misconfiguration
• An antenna installed inside the steel cabinet it was meant to be mounted on
• Cable run three times longer than it needed to be
• A firmware update nobody tested against a real network

Read that list again. Most of it is not the network’s fault, and almost none of it would move a single place in a league table.

Designing for resilience instead

Professional deployments start from one assumption: networks will fail occasionally. Not might. Will. Everything follows from accepting that.

Resilience is layered rather than purchased, and the layers run roughly from cheapest to most capable.

Antenna and site engineering

The bottom layer costs a few pounds of coax and an afternoon of care, and prevents more outages than anything above it. Get the antenna outside, get it high, keep the cable short, and most of the rest becomes academic.

Multi-network roaming SIMs

Rather than depending on a single operator, a roaming IoT SIM can attach to multiple UK networks. If one operator has a local problem, another may not. It is not a guarantee of uninterrupted service, and steering behaviour matters enormously, but it removes single-operator dependency, which is the largest single point of failure in most estates.

Dual SIM, where it is justified

Two SIMs from genuinely separate providers with automatic failover. Worth it for critical applications. Not worth it for a car park sensor. The discipline is knowing which is which.

Watchdogs and automatic recovery

Ping reboot, connection monitoring, modem restart, staged escalation to a full reboot. Unglamorous, and responsible for more uptime than any other single feature. A router that fixes itself at 3am does not generate a site visit.

Store and forward

MQTT with persistent sessions, local queueing, buffered telemetry. The connection drops for forty minutes, the data does not. This turns an outage into a delay, which is a completely different class of problem.

Edge processing

Docker or Node-RED on the router, local logic, local decisions. If the control loop only closes when the cloud is reachable, the design has a fault no network can fix.

Wired or cellular failover

Cellular as backup to fibre, or fibre as backup to cellular. Multi-WAN with sensible failover rules and, importantly, tested failback.

What 99.9 per cent uptime actually buys you

This is where the headlines and the contracts collide. People read “any outage” as “failure”. SLAs do not work that way, as covered in our recent piece on what 99.9% uptime actually means.

A provider can hit 99.9 per cent, honour the contract in full, and still have dropped your estate several separate times during the year. That is not sharp practice. It is what the number means. The engineering response is not to hunt for a bigger number of nines. It is to build a system where 8.76 hours of network absence costs you nothing, because the data was buffered, the logic ran locally and the device recovered on its own.

So should businesses be worried?

About the state of UK mobile infrastructure, yes, and the reports deserve the attention they are getting. Coming last out of 29 European markets is not a rounding error, and the point Which? makes about obligations being written around availability rather than quality is a legitimate regulatory criticism. Better networks would benefit everyone, IoT included.

About whether your monitoring estate is about to fall over because London came behind cities nobody expected, no. That conclusion does not follow from the data. Well-designed IoT systems have run on these same networks, at these same speeds, for years, because they were never asking the networks for the thing the rankings measure.

The questions an experienced engineer asks are not “which network is fastest”.

• Which network performs best at this specific location, tested rather than predicted?
• Is multi-network roaming appropriate, and how does steering actually behave?
• Is dual SIM justified by the cost of an outage?
• What happens to the process when connectivity disappears for an hour?
• Does the device recover without a human?
• Can it be diagnosed remotely, or does every fault become a van?
• Is the antenna installation doing the site justice?

Answer those and the league table position becomes interesting national context rather than an operational risk.

Headlines do not tell the whole story

The recent reports are worth having. They have started a conversation about digital infrastructure the UK has needed for a decade, and the criticism of coverage obligations based on availability rather than quality is well made. Nobody serious is arguing the UK’s networks are fine.

But the leap from “UK mobile is slow” to “cellular IoT is unreliable in Britain” is not supported by the data, and it is not supported by the field either. A properly specified industrial router, on a roofline antenna, with a multi-network SIM, a watchdog, local buffering and edge logic, is operating in a different world to a handset in a coat pocket. It was never going to see the same network, and it was never going to need the same things from it.

For industrial IoT, resilience beats speed. It did before this news cycle, it will after, and it will still be true when the UK finally climbs the table.

Frequently asked questions

Are the reports about UK mobile networks accurate?

Yes, as far as they go. Opensignal’s Q1 2026 study placed the UK 29th of 29 European markets, with average 5G downloads of 84 Mbps against Denmark’s 226 Mbps and the region’s lowest Excellent Consistent Quality score at 80 per cent. Which? used Opensignal data from over a million handsets to rank the UK 57th globally. The methodology is sound. The limitation is that it measures consumer smartphone experience, which is not the same thing as network availability for machines.

Does slow mobile broadband mean IoT connectivity is unreliable?

No. Speed and availability are separate engineering properties. A network averaging 15 Mbps with 99.99 per cent availability is more useful for industrial telemetry than one averaging 400 Mbps with regular outages. Most IoT applications never approach the available bandwidth, so a reduction in headline speed has no measurable effect on them.

How much bandwidth does an IoT device actually need?

Far less than people assume. A sensor sending 2KB every fifteen minutes transmits 16 kilobits, which takes about a hundredth of a second at 1.5 Mbps. Even at the UK’s criticised 84 Mbps average, there is roughly fifty-six times more capacity available than the application will ever request. Bandwidth is almost never the constraint. Attachment stability is.

Do I need a 5G router, or is Cat 4 still enough?

For most telemetry, SCADA, metering and alarm applications, Cat 4 is not merely enough but substantially over-specified, offering around 150 Mbps down where the application may need 1.5 Mbps. Cat 1, Cat 1 bis and LTE-M cover a great deal of the estate perfectly well. 5G is worth specifying where sustained uplink matters, such as multi-camera video, or where an application will genuinely use network slicing once 5G SA is widespread.

What is the difference between 5G NSA and 5G SA?

5G NSA (Non-Standalone) puts a 5G radio on top of an existing 4G core, with the device retaining an LTE anchor. It improves download throughput and nothing else of consequence. 5G SA (Standalone) uses an end-to-end 5G core, which is where network slicing, deterministic low latency, improved uplink and RedCap support actually live. In Q1 2026, UK time on a 5G SA bearer was 8.3 per cent on EE, 5.3 per cent on O2 and 4 per cent on VodafoneThree. When the 5G icon appears on a UK phone today, it is almost always NSA.

Why does an industrial router get a connection where my phone does not?

Because it is a different class of equipment on a different RF path. The router uses external antennas, mounted at height, chosen for gain and polarisation, with properly spaced MIMO elements, on permanent power, with no hand detuning it and no battery management throttling its radio. It can also lock bands, lock cells, control PLMN selection and reboot itself when a session hangs. A phone can do none of that. The site usually has signal. The phone does not.

Does 99.9% uptime mean the network never goes down?

No. 99.9 per cent permits 8.76 hours of downtime per year, or 43.8 minutes per month. A provider can experience several separate outages, honour the SLA in full and still meet its contractual obligation. Resilient design assumes those hours will happen and ensures they cost nothing, through buffering, local logic and automatic recovery.

Will a multi-network roaming SIM fix poor coverage?

It removes dependency on a single operator, which is the largest single point of failure in most estates. If one network has a local fault or a mast outage, another may still be reachable. It will not conjure coverage where no operator has any, and steering behaviour can materially change the outcome, so it is worth understanding whether a SIM is steered or unsteered before assuming it will select the best available network.

Are IoT SIMs different from the SIM in my phone?

Substantially. Managed IoT and M2M SIMs commonly offer multi-network roaming, private APNs, fixed or public IP addressing, fleet management platforms, session-level diagnostics and usage controls. None of those exist on a consumer tariff, and all of them change the resilience equation before the radio is even considered.