6G AI rollout hinges on where—not whether—networks run intelligence

The shift toward AI in 6G reflects a broader industry push to squeeze more efficiency from wireless networks that have spent decades approaching theoretical performance limits. Nvidia's $1 billion investment in Nokia signals that hardware makers see significant commercial opportunity, and the proof-of-concept trials by Nokia, Ericsson, and Nvidia with T-Mobile demonstrate that AI workloads can technically coexist with radio network operations. However, the trials have not yet proven that AI must be deeply integrated into the core RAN architecture to deliver benefits—they show that AI and RAN can share resources, a narrower claim.

The crux of the debate, as framed by industry engineers like Kim Kyllesbech Larsen at United Group, is not ideological but pragmatic: operators need evidence that AI-RAN delivers capabilities or economics unavailable through simpler, existing approaches. Companies like Opanga Networks already deliver some of the efficiency gains that AI-RAN aims for without new hardware. The coordination risk—that multiple independent AI systems optimizing for different goals (energy, speed, coverage) could undermine each other—adds a technical complexity that goes beyond mere hardware deployment.

The September Madrid standards meeting will be consequential because it will establish the baseline architecture for 6G, influencing which vendors succeed and which operators are forced to invest in unproven systems. The question of where AI intelligence should reside—inside towers, at edge sites, or in regional data centers—is not yet settled, and the answer will shape both the cost and viability of 6G rollout.