Not even AI can justify mobile edge computing

One of the key potential applications of 5G was edge computing, sometimes known as mobile edge computing (MEC). The idea was that mobile network operators (MNOs) could deploy computing resources closer to their base stations, delivering ultra-low latency and enabling MNOs to charge more for premium services.

 

Fast forward five years and edge computing deployments by MNOs remain virtually non-existent. Only a handful of operators have trialled partnerships with search engine companies, applying their own computer resources. But the idea is still alive, and there is renewed discussion as to whether 6G could bring about mobile edge, especially with the potential deployment of AI-related resources deep within the network. Indeed, Nvidia’s investment in Nokia is partly based on this concept.

 

The edge paradox: why the physics and economics don’t work

 

The core premise of mobile edge computing is that moving compute closer to the end-user drastically reduces latency. However, a closer look at network physics and market realities reveals why the concept has stalled.

 

The fundamental physics of data transmission poses a significant hurdle for edge computing, primarily due to the speed-of-light limitation in fibre-optic cables. Light travels through fibre at approximately 200 kilometres per millisecond (ms). Compared with a hyperscaler data centre located 100 kilometres away, a negligible single millisecond is added to a signal’s round-trip response time.

 

This single-millisecond reduction becomes even less significant when set against existing network latency bottlenecks. Standard 5G networks currently operate with latencies of 30 to 40 milliseconds, while even high-end private networks struggle to get below 10 milliseconds. When combined with the several milliseconds required to process data, such as video compression, shaving off a single millisecond by moving compute resources directly to a local base station yields virtually no practical operational benefit.

 

Furthermore, geography and economics strongly favour centralised facilities over local edge infrastructure. Most geographic regions already house a major urban hub within that one-millisecond range, hosting large-scale hyperscaler data centres. These sprawling facilities enjoy massive economies of scale, and hyperscalers possess far superior commercial capabilities to package and sell compute services compared with traditional telecoms operators.

 

Current architectural realities

 

Instead of processing data locally at base stations, modern network traffic continues to rely on traditional internet routing pathways. In the standard operational model, mobile network operators transport user data directly to internet exchange points (IXPs), or peering points, where various networks intersect. From these exchange hubs, traffic travels to centralised hyperscaler data centres for processing before returning to the user along the same path.

 

The primary exception to this flow occurs in private 5G deployments, where network traffic is routed directly into the owner’s internal IT system. However, although industry marketing often labels this architecture as “edge computing”, it is functionally just a localised form of standard network peering into a private corporate network.

 

Far from expanding edge compute, MNOs are actively centralising their operations – consolidating baseband processing from multiple base stations into centralised units rather than pushing processing to the network perimeter.

 

Will 6G rescue edge computing?

 

At this stage, what 6G will be remains uncertain. MNOs are pushing for “software-only” updates to minimise operational expenditure, while equipment manufacturers promote “hyper-5G”, featuring higher speeds and lower latencies. However, for 6G to make edge computing viable, four strict criteria must be satisfied simultaneously:

• A new application requiring latencies below 5ms
• A commercial willingness to pay substantially more for such ultra-low-latency services
• Additional spectrum allocation to support a low-latency air interface
• A sufficiently widespread 6G deployment to make edge computing viable across entire regions

Presently, all four conditions appear highly improbable. Proposed 6G applications are largely recycled 5G promises that failed to materialise. Neither consumers nor enterprises have shown a willingness to pay extra for 5G, and securing new spectrum is becoming increasingly difficult. Given that mid-band 5G (3.5 GHz) reaches only around 20 per cent of most countries, 6G coverage will likely be even more restricted.

 

The new front-runners: sensing and AI

 

Industry advocates point to two emerging technologies as potential catalysts for 6G edge compute, but both fall short under scrutiny.

 

Integrated sensing is often highlighted as a major advancement for 6G, but its commercial viability remains highly uncertain because of weak market demand. There is currently no clear customer base or proven commercial use case willing to pay a premium for embedded network-based sensing capabilities.

 

Beyond the lack of a clear market, the technology faces fundamental technical trade-offs. Implementing effective radar-like sensing within a mobile network requires operating in extremely high-frequency spectrum bands. However, these elevated frequencies are ill-suited to standard, wide-area mobile communications, making it difficult to unify both functions within a single operational network.

 

The argument that AI will drive demand for edge computing overlooks how AI workloads operate in practice. Most AI tasks fall into two extremes: lightweight models that run locally on end-user devices – such as mobile AI assistants or computer vision tasks – and heavy-duty processing that relies on high-performance hyperscaler data centres. Very few AI applications demand sub-millisecond response times, and there is little commercial appetite to pay a premium simply to shave off a single millisecond of latency.

 

Similarly, proposals for mobile network operators to monetise their “spare” baseband processing capacity during off-peak hours are fundamentally flawed. Hyperscaler data centres also experience idle periods during off-peak times, eliminating any shortage of cheap compute. What’s more, dynamically reallocating compute resources between complex radio network functions and external AI tasks introduces immense technical complexity for very little commercial gain.

 

The edge belongs to the cloud

 

The telecoms sector has a long record of attempting – and failing – to compete with hyperscalers by offering value-added services. MNOs consistently struggle in markets dominated by hyperscalers and over-the-top (OTT) providers, and edge computing is proving no different.

 

As discussions shift towards the 2030 rollout, 6G risks repeating 5G mistakes by selling technological promises disconnected from economic, physical and market realities. Edge computing will not be saved by 6G, nor will AI and sensing magically monetise network infrastructure.

 

Instead of representing a vital leap forward, pushing for edge-native 6G infrastructure increasingly looks like an expensive cycle fuelled by industry competition – while the true “edge” remains firmly anchored in hyperscaler data centres across major urban hubs.

---