The 6G Core Divide: Why 3GPP's AI Domain Debate Is Open RAN's Next Battleground
- October 4, 2026
- 9 mins
- Technology
- 3gpp 6g ai ran network architecture open ran standards telecom
The 6G Core Divide: Why 3GPP’s AI Domain Debate Is Open RAN’s Next Battleground
While the industry fixates on AI-RAN field trials and GPU-accelerated baseband, a quieter but arguably more consequential battle is unfolding inside 3GPP’s SA2 working group. Light Reading’s recent report on “The 6G core divide” lays out a genuine architectural fork in the road: how the 6G core network should handle AI, and by extension, what that means for every device, RAN, and core component that connects to it. For Open RAN operators who have spent years disaggregating the radio access network, this debate is not academic. The decisions made in 3GPP over the coming months will determine whether the intelligence fabric being built across Open RAN, edge, and cloud can actually reach the end user — or whether it gets walled off behind a new generation of proprietary interfaces.
Here are five things this 6G core divide means for Open RAN operators, drawn from what we know about the SA2 study phase and the positions staked out by the major players.
1. The Fork in the Road: Direction 1 vs. the Alternatives
The 3GPP SA2 study phase has identified three candidate directions for how AI functionality should be integrated into the 6G core. The most prominent — and the one with the most heavyweight backing — is Direction 1, which calls for a separate AI domain for intent handling, independent of the packet-switched network. In this model, the AMF (Access and Mobility Management Function) continues to route NAS signaling, but AI fulfillment happens via service-based interfaces (SBI).
This direction is backed by an unusually broad coalition: Nokia, T-Mobile USA, Verizon, NVIDIA, Deutsche Telekom, Apple, Qualcomm, Ericsson, and NEC. That lineup spans infrastructure vendors, chip designers, device makers, and operators — a rare alignment that signals just how seriously the industry takes this architectural question.
But Direction 1 comes with a significant catch. As Light Reading notes, implementation requires complex UE AI domain client integration. In plain terms, it means reworking the device side of the network in ways that go far beyond a firmware update. The other candidate directions, while less detailed in the public reporting, presumably offer simpler paths that avoid some of this complexity — but likely at the cost of less architectural flexibility.
For Open RAN operators, the key question is not which direction wins in the abstract. It is whether the winning direction preserves the multi-vendor, interface-driven model that Open RAN has painstakingly built — or whether it reintroduces the kind of tight coupling between core and device that Open RAN was designed to eliminate.
2. The NAS Signaling Rewrite Nobody Is Talking About
Here is the detail that should make every Open RAN operator sit up: Direction 1 is not just tweaking how a modem connects to the RAN. It standardizes UE AI domain client interfaces to the core by creating an application-layer intent client — one that understands what the user or agent wants — and a 3GPP AI domain client that translates that intent into secure network signaling.
Historically, mobile devices negotiated radio access and sessions via NAS signaling, while applications handled everything else. That separation is one of the foundational assumptions of mobile network architecture. Direction 1 fundamentally changes device architecture by formalizing an intent-driven AI domain in both the core network and the device.
The commercial implications are substantial. Reworking NAS requires modem redesign, core security procedure overhaul, and device backward-compatibility planning. This is not a software patch. It is a generational change in how devices and networks talk to each other.
For Open RAN operators, this matters because the RAN sits squarely in the middle of this signaling path. If the UE AI domain client generates new types of intent signaling that must traverse the RAN to reach the core, the RAN must be able to parse, prioritize, and potentially act on that signaling. That is a new requirement for Open RAN software stacks — and an opportunity for vendors who can deliver it in a disaggregated, multi-vendor way.
3. Why the RAN Cannot Be an Afterthought
The 6G core divide is often framed as a core network and device story. That framing is a mistake. The RAN is where the intent meets the air interface, and it is where the architectural rubber hits the road.
Consider what an intent-driven AI domain implies for the RAN. If the core is routing AI fulfillment via SBI while the AMF handles NAS signaling, the RAN must be able to distinguish between traditional session traffic and AI-domain signaling. It must apply different quality-of-service and security treatment to each. And it must do so in a multi-vendor environment where the RAN components may come from different suppliers than the core.
This is precisely the scenario Open RAN was built for — but only if the interfaces are standardized in a way that allows the RAN to participate in the AI domain rather than being a passive pipe. The E3 interface work that Nokia and others have championed for dApps is one example of how this could work. But E3 is an O-RAN Alliance construct, and the 3GPP SA2 debate is happening in a different standards body. The risk is that the two efforts diverge, leaving operators to stitch together incompatible approaches.
Operators should be asking their vendors a direct question: how will your RAN software handle UE AI domain signaling, and will it do so through open interfaces or proprietary extensions? The answer will reveal whether a vendor is building for an open 6G or a walled garden.
4. The Coalition Behind Direction 1 Is a Signal — and a Warning
The list of companies backing Direction 1 is striking for its breadth. Nokia, Ericsson, and NEC represent the traditional infrastructure vendor community. NVIDIA represents the GPU-accelerated computing wave that is reshaping AI-RAN. Qualcomm and Apple represent the device silicon and platform layer. T-Mobile USA, Verizon, and Deutsche Telekom represent operators with significant 5G-Advanced and 6G ambitions.
On one level, this coalition is reassuring. It suggests that the industry is converging on a direction that can accommodate both AI-native operations and the multi-vendor reality that operators increasingly demand. A broad coalition is more likely to produce a standard that actually gets implemented.
On another level, it is a warning. Broad coalitions often paper over deep disagreements about implementation. The fact that Direction 1 requires complex UE AI domain client integration means that the device-side work — which is largely controlled by Apple and Qualcomm — could become a bottleneck. If device makers delay or implement the AI domain client in proprietary ways, operators could find themselves with a core network ready for intent-driven AI but a device population that cannot use it.
For Open RAN operators, the lesson is to engage early with both the standards process and their device partners. The RAN is not the only place where disaggregation matters. The device-to-core signaling path is about to become disaggregated in new ways, and operators need visibility into how that will work.
5. What Operators Should Do Now
The 3GPP SA2 study phase is not yet complete, and no final decision has been made. That means operators still have time to influence the outcome — but not much. Here are four concrete actions Open RAN operators should take in the coming months.
First, map your current core and RAN signaling architecture against the Direction 1 model. Identify where AI-domain signaling would need to traverse your RAN, and assess whether your existing Open RAN software stack can handle it. If it cannot, that is a gap to raise with your vendors now, not after the standard is frozen.
Second, engage with the O-RAN Alliance and 3GPP coordination efforts. The E3 interface, the SMO framework, and the RIC are all potential integration points for AI-domain signaling. Operators who understand how these pieces fit together will be better positioned to demand coherent standards rather than fragmented ones.
Third, pressure-test your vendors’ roadmaps. Ask specifically how they plan to support UE AI domain client signaling in their RAN and core products. Vendors who have a clear answer are likely already engaged in the standards process. Vendors who do not may be planning to wait and see — which means operators will be waiting with them.
Fourth, think about the device population. The UE AI domain client is a device-side requirement, and device refresh cycles are measured in years. Operators should be talking to their device partners about timelines and backward-compatibility strategies. A 6G core that supports intent-driven AI is of limited value if the installed base of devices cannot use it.
The Bigger Picture
The 6G core divide is not just a standards debate. It is a test of whether the telecom industry has learned the lessons of 5G. The 5G era showed that architectural decisions made early — about network slicing, about core virtualization, about RAN disaggregation — have long-lasting consequences for competition, innovation, and operator economics.
Open RAN was born from a desire to break free from single-vendor lock-in and to create an environment where innovation could come from many sources. The 6G core debate is the next frontier of that same struggle. If Direction 1 wins with an architecture that preserves open interfaces and multi-vendor participation, it could be a major win for Open RAN operators. If it wins with an architecture that concentrates control in a few vendors’ hands, it could undermine the very principles Open RAN was built on.
The good news is that the coalition behind Direction 1 includes companies that have publicly committed to open interfaces. The bad news is that coalitions are not guarantees. Operators who want an open 6G core need to show up, ask hard questions, and make their requirements known. The window to do so is open now. It will not stay open forever.