Over the past two years, I’ve been leading a team of three analysts looking into AI edge computing. We’ve been charting the future role of telecom operators in a world of robots, AI voice services, sovereign AI, and on-premises industrial AI. We just released our first forecast in Telco Edge AI 2026 this week.
This is not MEC 2.0.
The biggest mistake in the mobile edge computing (MEC) era was the unfounded belief that low latency requirements would magically create revenue for the telcos. In psychology, this is known as “cognitive bias." In tech, it’s the tendency to think that your “better mousetrap” will automatically sell like hotcakes.
The problem with MEC was that very few customers were willing to pay for lower latency. Nobody bought the “better mousetrap”.
Looking at edge AI, we started by identifying several applications that people really will pay for. AI voice services, such as live translation or transcription. Industrial video analysis.
Robots and autonomous machines. AI CDNs generating real-time content. ISAC and other forms of environmental monitoring.
As we mapped these applications into network and compute infrastructure, we realized that the right place to perform the computing is often distributed across multiple locations. Most robotics developers intend to run their entire AI model on-board the robot, but need to outsource some high level strategic decisions to off-board AI models.
Many enterprises want to keep their data on-premises, but need training and data inputs from other locations. Sovereign and regional AI data centers will need authentication and policy control.
The edge is not a location. It’s a service where local customers will be steered to compute resources that are best for them.
That is the role that the telco can play: the traffic cop that decides which AI inferencing task should be sent to the regional sovereign AI factory, and which tasks should be done locally for performance, data privacy, or cost reasons.
My colleague Sebastian Barros coined the term the “AI Control Plane” to describe the function that the telco can serve here.
I’m sure that the industry will come up with a new name for this, but “control plane” is a good description. The telco can essentially play the role of broker for AI workloads. When a user pops up with an AI task, the telco’s “control plane” can:
- Identify and authenticate the person, device, or AI agent that initiated the task;
- Steer the task to the right model and compute location with the right mix of GPUs, TPUs, and CPUs;
- Provide the appropriate level of connectivity to match latency and reliability requirements;
- Govern the usage of each device or AI agent, to avoid runaway token costs.
Many CFOs were surprised by the million-dollar costs of tokens in 2025, and have now started to look for ways to economize on tokens. It’s not easy for an enterprise to control this…in theory, each employee should use the cheapest model that is right for the task. But in practice, employees use the most expensive frontier model for everything because they don’t know what level of model is needed. They’re spending more money and they’re sharing sensitive data on public models. Imagine how frustrated these CFOs are today.
There is real value in this role as a broker. Unlike MEC, which relied on new revenue creation, the telco can add value to edge AI by steering AI workloads instantly to the right network resources and the right compute resources. And telcos are in a good position to compete favorably. Telcos are already taking control of Sovereign AI in many countries. They already have the best authentication and policy control apparatus in the world. And they have the distributed presence that is required to interact with customers wherever the customer is located.
A few months ago, we published a groundbreaking report on The Economics of Kinetic Tokens, laying out a path for how the network can support outdoor robots. Our Telco Edge AI 2026 report and forecast takes it a step farther.
In our new forecast, we identify four main sources of telco revenue in a wide range of 10+ application areas:
- Regional and sovereign AI service revenues, where the telco has a local advantage in offering the AI or cloud computing service itself;
- Telco edge and metro inference services;
- Managed services for private cellular networks with Industrial AI workloads;
- Support of on-device AI models with authentication, policy control, or network assurance (we don’t count the connectivity revenue itself; we’re looking at the services that help on-board AI models to work better)
In the end, the telcos can be more than a “dumb pipe”. In countries with sovereign AI protections, we believe that the telcos have major opportunities to dominate the AI markets, and they will naturally provide the “control plane” to steer customers into their AI models. In other countries, the telcos can still take a pivotal role by leaning on their expertise in identity, policy control, and dealing with huge complex streams of data in real time.
Our expectations for telco revenue are modest for 2026-2030. But in the 2035 timeframe we see this “control plane” as the strategic high ground in the battle for customer share. This is the launchpad that the telcos are looking for.
Joe Madden is principal analyst at Mobile Experts, a network of market and technology experts that analyzes wireless markets.
Opinions from industry experts, analysts or our editorial staff do not represent the opinions of Fierce Network.