6G's AI Question Shifts From Adoption to Architecture

IEEE Spectrum framed the debate this way: The Real Question for AI for 6G Isn't Whether—It's Where. The publication's choice of words marks the moment the telecom industry stopped asking whether artificial intelligence belongs in next-generation wireless and started asking where, exactly, it should sit.

The shift comes against a fixed timeline. The International Telecommunication Union (ITU) published its IMT-2030 framework in 2024, setting the outer requirements for 6G. Commercial launches are widely expected around 2030. With the standards window narrowing, placement decisions made in the next two to three years will shape the networks operators spend the next three decades running.

What does "where" actually cover?

The placement question runs through the entire network stack. On the radio access network (RAN), AI is already optimizing beamforming, scheduling, and energy use in 5G-Advanced deployments. 3GPP froze Release 18 in 2024 and made AI and machine learning a first-class feature of the air interface. Release 19 is extending that work into additional use cases.

Inside the core, operators have begun placing AI inside service-based architectures to handle dynamic policy control, anomaly detection, and traffic steering across network slices. Edge deployments push inference closer to base sites, reducing latency for the industrial and vehicular use cases that 6G is designed to support.

The "where" also spans the device-to-network boundary. Splitting AI workloads between handsets and base stations has become its own debate. Vendors have proposed different distributions to balance power consumption against privacy, and the choice will affect how operators dimension their radio and compute infrastructure.

Why the placement debate matters now

Standards bodies are drawing the architecture lines for 6G in the same window where AI hardware is changing quickly. The shift from training-heavy models to smaller, more efficient inference engines has made on-device AI feasible. It has also forced network planners to revisit long-held assumptions about where compute lives.

Some researchers argue that AI should sit at the radio interface, where it can react to channel conditions in microseconds. Others want AI in the core, where it can manage slices and orchestrate resources across regions. A third camp pushes AI to the edge, closer to users but further from the centralized intelligence they prefer.

The answer is unlikely to be a single location. Operators spent the past decade disaggregating their networks to escape vendor lock-in, and they are not about to centralize AI in one place.

Looking ahead

The IEEE Spectrum framing suggests the industry has accepted AI's role in 6G. The harder work — choosing the architecture, defining the interfaces, writing the standards — still sits ahead. With IMT-2030 setting the outer requirements and 3GPP expected to open its 6G study items later this decade, the placement choices made now will define the networks of 2030 and beyond.