The Strategic Evolution of AI-Native RAN and 6G Standards

The Rise of the Large Telecom Model (LTM) and Agentic AI

The most significant breakthrough in recent years is the transition from "AI for RAN" to "AI-Native RAN." In the former, AI was a bolt-on feature used for high-level orchestration or energy saving. In the latter, the radio interface itself is designed by and for machine learning. A pivotal moment in this journey occurred at MWC Barcelona 2026, where a collaboration between Northeastern University, SoftBank Corp., Keysight, and zTouch Networks demonstrated an LTM-powered autonomous agentic AI-RAN. This system utilizes a Large Telecom Model to interpret high-level operator intents and translate them into real-time radio configurations, effectively removing the need for manual parameter tuning.

From Deterministic to Generative Radio

Traditional RAN relies on fixed mathematical models for channel estimation and equalization. AI-native systems, however, employ Neural Receivers that learn the specific characteristics of a local environment. This allows for superior performance in non-line-of-sight conditions and high-mobility scenarios. The LTM acts as the brain of this operation, processing vast amounts of telemetry data to predict traffic patterns and interference before they manifest. This move toward Intent-Driven Networking allows operators to define outcomes—such as "prioritize 8K video for this sector"—while the AI autonomously manages the underlying physical layer resources to achieve it.

The Role of DeepSig and Neural Signal Processing

At the heart of this challenge to incumbents like Ericsson and Nokia is DeepSig, a pioneer in AI-native wireless communications. By replacing traditional signal processing blocks with deep learning models, DeepSig has demonstrated that it is possible to achieve significant gains in throughput and power efficiency. Their involvement in the OCUDU (Open Cloud-native Unit / Distributed Unit) initiative, alongside Intel and Nvidia, signals a shift toward a more modular, software-centric RAN where the "secret sauce" is no longer in the ASIC (Application-Specific Integrated Circuit) but in the weights of the neural network.

"The transition to AI-native RAN represents a 'Cambrian Explosion' for the telecom industry, where the rigid structures of the past are being replaced by fluid, self-evolving architectures that learn from the spectrum they inhabit."

The Geopolitical and Strategic Divide in 6G Standardization

As the industry looks toward the 6G horizon, a significant rift has emerged regarding the role of Open RAN in future standards. While the movement toward disaggregated, open interfaces has gained momentum in the 5G era, its mandatory inclusion in 6G is a point of intense contention. Huawei, for instance, has emerged as a vocal opponent of making Open RAN a mandatory feature of forthcoming 6G networks. This opposition is rooted in a strategic preference for tightly integrated vertical stacks, which Huawei argues offer superior performance and energy efficiency compared to the current state of disaggregated solutions.

The Open Telco AI Initiative

In response to the dominance of proprietary stacks, the GSMA launched the Open Telco AI initiative in March 2026. This global effort aims to accelerate the development of telco-grade AI through open collaboration across operators, vendors, and developers. Supported by a new family of models from AT&T and compute resources from AMD and TensorWave, the initiative seeks to democratize access to high-performance AI models tailored specifically for telecommunications. This move is designed to prevent a "vendor lock-in" scenario in the AI era, ensuring that even smaller operators can leverage advanced LTMs without being tethered to a single infrastructure provider.

Strategic Requirements for vRAN and AI-RAN

The path forward is being further defined by leading operators like SK Telecom (SKT) and NTT Docomo. In a joint white paper, these industry leaders identified three key technical requirements for the successful development of vRAN (virtualized RAN) and AI-RAN:

• Hardware Acceleration: The need for specialized accelerators (GPUs or TPUs) that can handle both RAN processing and AI workloads simultaneously.
• Resource Pooling: The ability to dynamically allocate compute resources between different network functions based on real-time demand.
• Standardized Interfaces: Ensuring that AI models from different vendors can interoperate within a single RAN environment.

Hardware Heterogeneity and the GPU-RAN Convergence

The integration of AI into the RAN has sparked a fierce debate over the underlying hardware architecture. For years, the industry has relied on ASIC-based hardware for its efficiency and low latency. However, the rise of AI-RAN has brought Nvidia and its GPUs into the center of the telecom ecosystem. The core of the debate lies in whether AI compute should be integrated directly into the RAN processing chain or kept separate.

The Case for Separate Compute

Some analysts argue that keeping AI separate from RAN compute is the most sensible approach for most telcos. This "sidecar" model allows operators to upgrade their AI capabilities independently of their radio hardware. It also avoids the complexities of trying to run deterministic, real-time RAN tasks on the same silicon as stochastic AI workloads. As noted in recent industry analysis, while Nvidia's GPUs are powerful, the overhead of managing them within a traditional RAN framework can be prohibitive for operators with limited technical resources.

The Convergence of Compute and Connectivity

Conversely, proponents of integrated AI-RAN argue that the latency requirements of 6G will necessitate a much tighter coupling of compute and connectivity. This is particularly true for applications like Integrated Sensing and Communications (ISAC), where the network must process radar-like sensing data in real-time to create a digital twin of the physical environment. The challenge for the industry is to develop a Cloud-Native architecture that can support this level of integration without sacrificing the "five-nines" reliability that telecom operators demand.

• GPU Utilization: Maximizing the duty cycle of expensive GPU assets by sharing them between RAN and edge AI tasks.
• Power Consumption: Addressing the massive energy footprint of AI-native networks, which could offset the efficiency gains of 6G.
• Latency Jitter: Mitigating the non-deterministic nature of AI processing to ensure stable 6G performance.

Integrated Sensing and Communications (ISAC): The 6G Frontier

One of the most transformative features of 6G is Integrated Sensing and Communications (ISAC). This technology allows the radio network to function as a high-resolution radar, detecting the position, velocity, and shape of objects in its environment. ETSI’s Industry Specification Group on ISAC has been at the forefront of defining the architectural foundations for this capability. Their report, ETSI GR ISC 003, examines 17 key challenges across system and RAN architectures, proposing a framework where sensing is a native service of the network rather than an afterthought.

Architectural Foundations for ISAC

Integrating sensing into the RAN requires a fundamental rethink of the lower-layer architecture. The network must be able to manage "sensing beams" alongside communication beams, requiring sophisticated interference management and resource scheduling. The ETSI report suggests that 6G RAN must support a multi-static sensing model, where multiple base stations work together to track objects with centimeter-level precision. This capability will be essential for future applications such as autonomous drone delivery, smart city traffic management, and industrial automation.

Security, Privacy, and Trustworthiness

The dual-purpose nature of ISAC introduces unprecedented security and privacy risks. If every base station is a radar, the potential for unauthorized surveillance is immense. ETSI GR ISC 004 identifies 19 key issues related to security, privacy, and sustainability in ISAC systems. These include:

• Data Anonymization: Ensuring that sensing data cannot be used to identify specific individuals without their consent.
• Sensing Spoofing: Protecting the network from malicious actors who might attempt to feed false sensing data into the system.
• Sustainability: Managing the additional power required for continuous environmental sensing.

"ISAC turns the network into a set of eyes for the digital world, but without a robust framework for privacy and trust, it risks becoming a tool for pervasive surveillance rather than a platform for innovation."

Security, Sustainability, and the "Open Telco AI" Framework

As AI becomes the backbone of the network, the definition of "telco-grade" is being expanded to include AI-specific metrics. The GSMA Open Telco AI initiative is critical here, as it establishes the benchmarks for what constitutes a secure and reliable telecom model. Unlike general-purpose LLMs, telco-grade AI must be deterministic, explainable, and capable of operating under extreme latency constraints. The family of models introduced by AT&T as part of this initiative is specifically trained on telecom datasets, ensuring they understand the nuances of 3GPP protocols and network topology.

The Sustainability Imperative

Sustainability has moved from a corporate social responsibility (CSR) goal to a core architectural requirement. The ETSI reports on 6G emphasize that the energy cost of AI and sensing must be offset by the efficiencies they create. AI-native RAN is expected to play a dual role here: while the compute itself is energy-intensive, the AI's ability to precisely manage power—turning off unused components in micro-seconds—could lead to a net reduction in the network's carbon footprint. This "AI for Green" vs. "Green AI" balance will be a defining metric for 6G success.

The Path to 2030

The road to 6G is paved with these complex technological and strategic trade-offs. The industry is currently in a phase of intense experimentation, as seen in the MWC 2026 demonstrations. The next four years will be critical for the 3GPP and ETSI as they move from reports and white papers to the actual freezing of 6G specifications. The outcome will determine whether the future of telecommunications is a fragmented landscape of proprietary "AI islands" or a unified, open, and intelligent global fabric.

• Standardization Timeline: 3GPP Study Items for 6G are expected to conclude by 2027, with the first set of specifications (Release 21) targeted for 2029-2030.
• Spectrum Allocation: The push for upper mid-band (7-15 GHz) spectrum will be essential to provide the bandwidth required for both high-capacity 6G and high-resolution ISAC.
• Ecosystem Collaboration: The success of initiatives like Open Telco AI will depend on the willingness of traditional rivals to share data and models for the collective benefit of the industry.