The rollout of 5G was accompanied by claims about remote surgery, autonomous vehicles, transformed industry and applications that couldn't be described because they hadn't been invented.
What most users experienced was somewhat faster mobile internet. The gap has a technical explanation.
The frequency trade-off
The central physical constraint, and it determines nearly everything.
Higher frequencies carry more data and travel shorter distances and penetrate obstacles poorly. Lower frequencies travel further and through walls, and carry less.
This isn't an engineering limitation to be overcome; it's electromagnetic physics.
The dramatic speed demonstrations used millimetre wave frequencies — very high frequency, very high capacity, and range measured in hundreds of metres with poor building penetration.
Deploying that at national scale would require an enormous density of installations, which is economically prohibitive outside dense urban areas and specific venues.
So most 5G deployment uses lower and mid-band frequencies, which perform better than 4G and nothing like the demonstrations.
What was actually delivered
Being fair, the improvements are real.
Capacity. More users can be served simultaneously in a given area. This matters at crowded venues and in dense cities, and it's the least visible improvement because it manifests as things not failing.
Speed. Meaningfully faster than 4G on mid-band, though far short of headline figures.
Latency. Lower, though the improvement is smaller in practice than in specification, because latency in a real connection is dominated by factors beyond the radio link.
Efficiency. More data per unit of spectrum, which matters for operators.
These are worthwhile. They're an incremental generational improvement rather than a transformation.
Why the applications didn't materialise
Remote surgery. The frequently cited example. The problems are not primarily network latency — they're reliability guarantees, liability, regulation, the need for a surgeon present anyway, and the fact that a wireless link is a worse choice than a wired one for anything critical.
Autonomous vehicles. Vehicles must operate safely without connectivity, since connectivity cannot be guaranteed. Which means anything safety-critical must be handled on board, which means the network isn't the constraint.
Industrial automation. Genuine applications exist, mostly in private networks on dedicated spectrum within a single site. That's a real use case and it's not what consumers were sold.
The unspecified new applications. A category that appears in every generational marketing cycle and rarely materialises as described.
The latency point specifically
Worth explaining because it was central to the marketing.
Specifications describe very low latency on the radio link under ideal conditions. Real-world latency includes transit across the operator's network, across the internet, and processing at the destination.
Improving one component of a chain improves the total by that component's share. If the radio link was a modest portion of total latency, halving it produces a modest improvement.
Achieving genuinely low end-to-end latency requires edge computing — processing near the user rather than in a distant data centre — which is a separate infrastructure investment that has proceeded more slowly than the radio rollout.
The generational pattern
Worth noting because it's repeating with the next generation.
Each mobile generation has been marketed with transformative claims, delivered incremental improvement, and enabled applications that weren't predicted.
The applications that actually emerged from 4G — video streaming, ride hailing, mobile-first services — were not what the marketing described, but they were substantial.
Which suggests the useful posture: assume the specific claims won't materialise, assume something unpredicted will, and don't make purchasing decisions based on capabilities that require infrastructure nobody has committed to building.
What to expect practically
For consumers: faster mobile data in areas with good coverage, better performance in crowded places, and battery consumption that varies by implementation.
Coverage remains the determining variable. A phone connected to 5G at poor signal performs worse than one on good 4G, and the indicator showing 5G says nothing about the quality of the connection.
Fixed wireless access — using mobile networks as home broadband — is the application that has arguably delivered most, particularly where fixed infrastructure is poor. That's a genuinely useful outcome and it wasn't the headline.
The next round
Standards work on the following generation is underway, accompanied by familiar claims.
The reasonable expectation, based on the pattern: real improvements in capacity and efficiency, some latency reduction, deployment taking longer than announced, and the transformative applications arriving somewhere other than where they were predicted.
That's not cynicism. It's the historical record, applied forward.
Network slicing
One capability from the specification worth mentioning because it may yet deliver something distinctive.
Network slicing allows an operator to partition a network into logically separate networks with different characteristics — one prioritising latency, one prioritising throughput, one prioritising reliability.
The applications are largely enterprise: dedicated capacity for a specific customer or use case, with guaranteed characteristics.
Deployment has been slower than anticipated, partly because it requires substantial changes to network core infrastructure and partly because the commercial models are unclear.
It also raises questions about network neutrality, since the ability to guarantee different quality of service to different customers is precisely what neutrality rules were designed to constrain.