AI Doesn't Just Need Data Centres. It Needs the Network to Reach You.

AI is often described as if it lives in an invisible cloud. In practice, an AI-enabled service still needs a physical route between the person, device or sensor using it and the system doing the work.

That route may use a mobile network, fixed broadband, a private network, local equipment, an edge service or a data centre. The choice affects speed, reliability, privacy and what happens when the connection is weak.

The AI revolution does not stop at the data-centre door. It still has to reach the user.

The current UK network question

Warnings from UK network operators, including VodafoneThree, have focused on the scale of future data demand and the infrastructure needed to support it. That is a planning and investment challenge, not proof that UK mobile networks cannot cope with AI today.

The careful framing is that operators are preparing for potentially substantial increases in future demand, including demand associated with increasingly data-rich AI services. Forecasts should remain forecasts. They should not be presented as guaranteed traffic volumes or certain network failure.

Why future AI could change traffic

A text prompt may involve a relatively modest exchange of data. Future AI-enabled work may also involve voice, images, video, cameras, smart glasses, vehicles, industrial equipment, field-service systems, sensors and continuous context.

That does not mean every future device will continuously stream huge amounts of information. It means that more business processes may depend on a reliable exchange of richer data, sometimes with tighter expectations about delay.

• A field engineer sending images to an AI assistant.
• A vehicle sharing location and sensor information.
• A mobile worker using real-time translation or transcription.
• A warehouse system combining cameras, stock data and instructions.
• A remote site synchronising documents and decisions with a cloud service.

Coverage is not capacity

Seeing a 5G symbol does not guarantee low congestion, good indoor coverage, reliable backhaul, consistent latency or strong performance at busy times. A connection that works for occasional email may not behave the same way when a new AI process depends on real-time images or audio.

What 5G Standalone changes

Many early 5G deployments rely partly on existing 4G core infrastructure. 5G Standalone uses a dedicated 5G core and can support more advanced network management and lower-latency capabilities for suitable applications.

That does not mean every SME needs 5G Standalone. It means that the network architecture underneath a service can affect what the service can reliably do. A business should start with the required outcome rather than buying a label.

Not every AI decision needs the cloud

Some processing can happen on the phone, laptop, vehicle, local equipment or an edge service close to the user. This is not a replacement for cloud AI. It is another design option.

• On-device AI can reduce the amount of information sent away.
• Edge AI can reduce the distance a time-sensitive request travels.
• Cloud AI can provide larger models, central management and shared data.
• A hybrid design can choose different locations for different tasks.

What happens to an SME with one bar of signal?

The best AI in the world is not much use to your business if your employee is standing in a field looking at one bar of signal. That may sound obvious, but technology purchases often test the product in an office rather than the place where the work actually happens.

• Engineers at customer sites.
• Construction and logistics teams.
• Retail and mobile payment systems.
• Remote locations and temporary work sites.
• Field services using cloud applications.
• Employees travelling between areas with different coverage.

Connectivity is part of business continuity

An AI-enabled process may need fixed broadband, a mobile connection, a backup path, offline operation or later synchronisation. That does not mean every SME should buy multiple redundant connections. Infrastructure should follow business risk.

• Use 4G or 5G failover where a short outage has a serious consequence.
• Allow work to continue offline where the process permits it.
• Synchronise safely when a connection returns.
• Test connectivity at real working locations, not only at headquarters.
• Use diverse providers where the business impact justifies it.

Buying an AI product does not make the infrastructure AI-ready

AI readiness can include connectivity, devices, identity, data, integration, security and resilience. A new subscription may solve the model or interface problem while leaving the network, permissions and recovery process untouched.

IT Club AI connectivity check

1. 1Which processes depend on connectivity?
2. 2Which planned AI processes increase that dependency?
3. 3Are users office-based or mobile?
4. 4Does the process require real-time access?
5. 5What happens when connectivity slows?
6. 6What happens when connectivity disappears?
7. 7Can anything run locally or offline?
8. 8Is backup connectivity proportionate to the business impact?

AI Needs Data Centres. But Who Should Pay for Them? →

Will your technology still work when the connection does not?

Ask the IT Club Advisor. Tell us what your people, devices and AI-enabled process need to do in the real world, including what happens when the connection becomes unreliable.

Sources and further reading

This is original IT Club commentary. Network demand forecasts, rollout plans and service availability change over time, so check current operator and regulator information before making an infrastructure decision.

MK Link: UK mobile networks not ready for AI traffic →

Ofcom: Connected Nations →

VodafoneThree →