6G After MWC26 Barcelona Signals a Shift from Vision to Influence in 3GPP

During MWC26 Barcelona, there was virtually no discussion about 6G; only three keynote speakers discussed 6G during the show. However, this does not indicate that 6G has not progressed. It instead indicates the substantial transition from pre-emptive positioning to more direct and practical efforts behind closed doors.

In the absence of coordinated vendor announcements, vendors are pursuing independent initiatives. For instance, there is an initiative by NVIDIA to develop Artificial Intelligence (AI)-enabled physical infrastructure products. Qualcomm is attempting to improve the relationships between its ecosystem and the companies that manufacture devices. AMD has introduced telecommunications platforms with integrated AI functionality. However, what is more significant than these vendor-specific initiatives is their conversion into substantive early-stage contributions associated with the working groups of The 3rd Generation Partnership Project (3GPP): RAN1 (the physical-layer evolution), RAN2 (protocol architecture), and SA2 (system architecture). While developing their early-stage proposals on AI-native interfaces and compute integrations related to Release 21, there's now informal alignment among these three Radio Access Network (RAN) working groups.

This apparent fragmentation is consistent with the current state of development by continuing through the 3GPP process. With Release 21 still several years away, this time frame gives vendors a chance to develop competing technologies and solutions for determining Release 21 technologies and features. Influence at this time will be less about being formally involved in approving these technologies and more about influencing study items, technical reports, and early workshop assessments to establish the direction of later standards/specifications.

At the same time, operators remain focused on 5G-Advanced deployments, automation, and cloud-native transformation, with little urgency to define or commercialize 6G. Work already underway in Release 18 and 19 on AI/Machine Learning (ML) for RAN, such as model lifecycle management and data frameworks, is effectively acting as a precursor to 6G.

MWC26's most significant change, so far, was that 6G is progressing from a new type of air interface goal to an emerging competition over technology stack control points. First, AI will move from being an optimization layer to the core components of network architecture. The vendors’ positioning indicates the need for supporting AI workloads as embedded functions and includes defined interfaces for inference, data access, and real-time decisions. This creates new standardization requirements for how AI interacts with RAN and core layers.

Second, the compute component is changing significantly. A growing focus on accelerated and distributed compute highlights a tighter coupling between telco networks and external compute. Thus, there are open questions about functional splits, latency management, and orchestration of various heterogeneous environments; this will eventually become major battle areas of 3GPP.

Third, intelligence at the device level is emerging as another influencing factor. Qualcomm's emphasis on AI-enabled chipsets shows a shift away from centralizing decision-making on either the network or the device level, focusing instead on distributing decision-making, especially for radio optimization and resource allocation.

Fourth, the differentiation of infrastructure is increasingly taking place below the level of standardization. Ericsson’s ongoing development of its own proprietary silicon, as well as embedded AI, confirms that even though 3GPP will define the various interfaces, the vendors will compete based on implementation. This results in a situation where interoperability will be defined and standardized, but performance and efficiency will be defined and remain based on the vendor.

Underlying these shifts is a structural lesson from 5G. The limited success of enterprise monetization was driven less by technical limitations and more by complexity, cost, and unclear value delivery. As a result, 6G is being shaped less as a vehicle for new services and more as a platform for simplifying operations, integrating compute, and improving economic efficiency. Release 21 is, therefore, likely to define architectural primitives rather than a complete system, increasing the strategic importance of early influence within 3GPP.

Vendors must focus on the control points developing within the 3GPP process to ensure that their 6G strategy has the necessary influence. This will require more than positioning and submitting technical proposals detailing how AI tasks can be executed as part of the network function, how distributed computing can be orchestrated between the RAN and Core, and how accelerators can be accessed by standardized interfaces. In practical terms, it is about designing models for real-time inference at the RAN level, compute-aware scheduling, and the abstraction of the heterogeneous hardware being used. The goal is to ensure that their architecture is integrated into these foundational configurations before the release of Release 21 that will restrict flexibility, while maintaining differentiation between proprietary silicon and software at the interface level.

Operators, on the other hand, must leverage their 5G-Advanced deployments to start influencing their future architecture. This will require systematic experimentation with key architectural trade-offs including centralized versus edge-based AI processing and identifying the breakdowns that occur in practical multi-vendor, cloud-native environments. The information obtained from these experiments will then be used to generate performance, cost, and energy efficiency metrics that can be fed directly into the work of 3GPP. Operators that can connect real-life constraints to proposed standards are going to develop solutions that can be implemented on a larger scale and will create a more economically viable outcome for all involved; those operators that can’t will inherit architectures that do not represent real-life operational scenarios.