Dominic Jainy stands at the forefront of the digital revolution, bridging the gap between cutting-edge artificial intelligence and the complex world of mobile network infrastructure. With an extensive background in machine learning and emerging technologies, Jainy offers a unique perspective on how the telecommunications sector can evolve without crumbling under the weight of legacy systems. This conversation delves into the strategic transition toward AI-RAN, exploring how operators can harvest immediate operational efficiencies while laying the groundwork for a fully AI-native 6G landscape. We examine the critical role of shared infrastructure, the pragmatism required for GPU deployments, and the essential economic metrics that will dictate the pace of this global technological shift.
How can operators realistically balance immediate operational gains like energy savings and spectral efficiency without committing to a massive, full-scale network overhaul?
The beauty of the current trajectory is that operators don’t have to “rip and replace” their entire setup to see a difference in their bottom line. By focusing on AI for RAN initially, they can visualize immediate savings in operating costs and energy consumption directly on their existing infrastructure. It’s a pragmatic first step where you can deploy software-driven capabilities for fault detection and spectral efficiency without a total modernization of the radio network. I’ve seen how these small, calculated injections of AI can make a network breathe easier, optimizing performance in real-time while keeping the legacy hardware humming along. This phased approach allows companies to feel the tangible benefits of reduced opex before they take the plunge into more capital-intensive upgrades.
In what ways are traditional self-organizing networks shifting toward AI-enhanced platforms, and what does this mean for Open RAN environments?
We are seeing a fascinating evolution where traditional self-organizing networks, or SON, are essentially being “brained up” by AI-enhanced platforms. In the Open RAN space, this is manifesting through service management and orchestration systems that utilize AI-powered xApps and rApps. These aren’t just for show; they are functional tools that can manage both modern and conventional radio networks simultaneously, creating a bridge between old and new. It’s like giving a veteran athlete a high-tech wearable—the core strength is there, but now the movements are precise and data-driven. This allows for a much more fluid management of resources across the board, ensuring that even older parts of the network benefit from the intelligence of the new orchestration layer.
Could you walk us through the three-stage roadmap for AI-RAN and what specifically characterizes each phase as we move toward 2030?
The journey is best visualized in three distinct chapters, starting with “AI for RAN,” which is our current primary focus through roughly 2027. During this stage, we are purely looking at operational optimization and using AI as a tool to sharpen existing radio functions. The second phase, “AI and RAN,” which we expect to see blossom between 2027 and 2030, is where the real magic happens as AI and radio workloads begin to share common, unified infrastructure. This is the period where we will see a surge in proof-of-concepts, heavily influenced by the research maturing around 6G standards. Finally, after 2030, we enter the “AI on RAN” era, where the radio access network transforms into a native platform for AI applications, opening up entirely new revenue streams that we are only just beginning to imagine.
What is the strategic reasoning behind being selective with GPU acceleration rather than deploying it universally across the radio network?
There is a common misconception that we need to sprinkle GPUs everywhere like fairy dust, but the reality is much more calculated. We should not think about GPU everywhere in the RAN; instead, it’s about targeted deployments where the business case is undeniably strong, particularly for AI inferencing at the edge. By starting with a few selected sites, operators can test the waters and run Layer 1 and Layer 2 functions without needing a specialized real-time kernel, which simplifies the stack significantly. Working with partners like SoftBank, Fujitsu, and Nvidia has shown us that this selective acceleration can provide massive technical advantages without the waste of over-provisioning. It’s about placing the power where the demand is, rather than blanketing the network in expensive hardware that might sit idle.
How do current collaborations between major tech players and telecom giants influence the timeline for commercial 6G and AI-native services?
The ecosystem collaborations we are seeing now, such as Red Hat’s work with SoftBank and Nvidia, are the actual engines driving us toward that 2030 horizon. These partnerships allow us to expand AI-RAN proof-of-concepts in a controlled environment, essentially using the remainder of this decade as a high-stakes laboratory. By extending initiatives like the AI Grid as a RAN-ready infrastructure platform at the edge, we are building the “AI fabric” that will eventually span the data center and the core. T-Mobile and SoftBank are already leading the charge here, proving that shared infrastructure isn’t just a theoretical dream but a functional reality. These early movers are setting the pace, ensuring that by the time 6G is ready for commercial rollout, the underlying AI-native infrastructure is already battle-tested and stable.
At what point does the deployment of AI-RAN move from a technical experiment to a clear-cut economic necessity for a mobile operator?
The shift happens the moment an operator can prove that AI-RAN creates measurable business returns and improves network quality in a way customers can actually feel. AI-RAN only makes sense if it provides both technological and economic benefits; it’s not enough to just be “cool” tech. One effective strategy is to start with AI inferencing workloads at the network edge to see how much revenue they can actually pull in before deciding how much GPU capacity to dedicate to radio functions. It’s a “show me the money” moment for the industry, where monetization potential and operational savings must outweigh the costs of implementation. If an operator can see their churn rates drop because of better network quality or find a new way to charge for edge AI services, the transition becomes a financial no-brainer.
What is your forecast for the role of AI-native transformation in the broader telecom landscape over the next decade?
I believe we will see a total shift where AI-RAN is no longer viewed as a standalone radio project, but as one piece of a much larger, AI-native puzzle. Operators will start applying the harsh lessons they’ve learned from deploying AI in their core and OSS/BSS systems to the radio architecture, creating a seamless cloud-native platform that covers everything from the data center to the edge. By 2030, the “clutter” of separate silos will give way to a consistent AI fabric, making the network far more autonomous and responsive than anything we have today. We are moving toward a future where the network doesn’t just carry data, it thinks about the data it carries, optimizing itself every millisecond to provide a level of service that feels almost invisible to the end user. It’s an exciting time because we are moving from building pipes to building an intelligent, living infrastructure.