6G shifts from faster networks to intelligent ones: AI orchestrates non-terrestrial networks, tackling latency, spectrum, and compute offload at global scale.
It is almost a law by this point that every decade or so, the wireless world will promise something audacious.
The jump from analog to digital. From voice to the internet. From the web on your phone to the vast sprawling promise of 5G. But now, on the cusp of 6G, the rhetoric sounds different. What we’re hearing now is far less about speed and much more about intelligence.
If 5G was a disruptive reconfiguration of spectrum and access, 6G is being set up as a fully integrated digital world where AI is more than just an application riding the network, transforming instead into an organized base for how the network is built and operated.
That idea of AI as a structural layer of communications comes directly from researchers like Professor Michele Zorzi who reminds audiences that the promise of 6G lies in collapsing a fractured ecosystem of services into a single, heterogeneous infrastructure.
Where 5G optimized around broadband, ultra-low latency, and massive device access, 6G is about orchestration. And orchestration at planetary scale can’t be written in static protocol documents. It will be learned, adjusted, predicted, and remade in real time by AI.
This framing matters because, as he says, non-terrestrial networks (satellites, high-altitude platforms, drones) have moved from the periphery back to the center of communications research.
Once relegated to television broadcast or niche military use, these networks are now considered an integral part of the 6G blueprint. That is not because rural coverage suddenly became a glamorous research problem but because resilience and flexibility have become existential requirements. Hurricanes wipe out terrestrial base stations. Wars target fiber backbones. Even in wealthy nations, suburban broadband still fails under peak loads.
And even the most redundant datacenter fabrics strain against the sheer variability of AI-driven traffic.
In this world AI becomes essential. A satellite constellation cannot be managed like a cell grid of fixed towers. It moves and overlaps and drifts. The link budgets change with elevation angles, the interference patterns shift by the second, the delays stack up in unpredictable ways. Zorzi catalogues these constraints from latency, spectrum scarcity, random access breakdowns, and handover complexity. Each of these though is ripe for AI systems to learn the patterns of and adapt.
Latency, in particular, is the villain that refuses to be engineered away. Geosynchronous satellites will always sit over 30 km up, their signals arriving with quarter-second delays. Even low earth orbit cannot escape the physics. The only option is prediction and the only viable approach is to anticipate, to mask the lag with inference about what is likely to come next. In other words, AI.
Zorzi also points out that usable airwaves are already saturated. Military radar, weather satellites, earth observation systems and the like all stake claims on frequencies. Cellular systems have to squeeze themselves into whatever gaps are left, often at millimeter or soon terahertz bands, where the propagation characteristics turn hostile.
Cognitive radio which has been a research theme for two decades also came up in the talk, but it is only in the AI era that dynamic spectrum access looks feasible. The ability to sense, to learn interference patterns in real time, to negotiate coexistence across domains could present an opportunity.
Another thread in Zorzi’s presentation pulls AI in almost automatically and that’s computation offloading. He describes a future where resource-constrained vehicles and sensors hand their workloads to airborne or spaceborne platforms equipped with more powerful processors. The use case he sketches is of vehicles with limited onboard compute, satellites with surplus capacity.
This is edge AI extended vertically into the atmosphere where models don’t have to wait for a terrestrial cloud connection; they can be run on a high-altitude platform with solar panels, or pre-processed in orbit before being handed back down to earth.
Zorzi also points to the problem of variability in demand (think burstiness in workloads, unpredictability in geography and the like). A transformer model running inference across millions of phones at once looks a lot like a dense crowd at a stadium hammering a cellular network. A generative service that spikes at launch in one continent and then fades to another time zone looks like a moving cluster of users that the NTN must follow. In both cases, the need is not raw capacity but flexible capacity, guided by smart orchestration.
Resilience is also critical. Terrestrial networks can be swept away in hurricanes, severed in earthquakes, or destroyed in conflicts. Satellites though are harder to reach and therefore harder to kill. But resilience without adaptability is fragile. A static satellite system could still fail under shifting loads. Only an AI-governed system capable of rerouting traffic across constellations, allocating resources across orbits, and triaging priority in disaster scenarios can fulfill the promise of NTN resilience.
The open research questions he lists, from how to model new channels, how to redesign access protocols, how to coordinate thousands of moving satellites, are the research questions of AI in communications. They are optimization problems at scales and speeds that defeat conventional engineering and really, they are problems of learning and adaptation in dynamic environments.
And so the story of 6G is not “faster phones” but the emergence of a nervous system that is self-optimizing. AI will sit at every layer: predicting demand before it arrives to sidestep the speed-of-light problem; redistributing traffic so that inference spikes in one region do not overwhelm capacity; deciding in microseconds whether to process data locally, on a vehicle, or to offload it to a satellite-borne accelerator. The network becomes not just a pipe but a decision engine.
Of course, no algorithm can erase propagation delay or conjure infinite spectrum. But as Zorzi stresses, even incremental gains matter when the difference is whether connectivity collapses in the middle of a disaster or holds.
More broadly though, if AI is to govern the flow of information across earth and sky, then communications systems themselves are becoming agents. Not passive conduits, but active participants in shaping how, when, and where intelligence is delivered. A non-terrestrial network becomes not just an invisible extension of infrastructure but an actor with its own adaptive logic.
