Summary Bullets:
• Nokia continues to build upon its AI RAN story, with new ecosystem partners and operator engagements as well as refined messaging.
• The company described a vision for training agents to be experts in specific network domains that also communicate across domains and promoted use cases including robotics and ISAC.
Nokia continues to build its AI RAN story. This week, the vendor used the Graphics Processing Unit Technology Conference (GTC) event in Berlin, created and hosted by AI giant NVIDIA – Nokia’s key partner in AI as well as a major shareholder – to burnish its message, with newly named partners, operator trials, use case demonstrations, and refined descriptions of how the AI RAN will work.
Building on its previous promotion of the use of “dApps” (third-party software applications that reside in the baseband unit, rather than a RAN Intelligent Controller, in order to minimize latency), Nokia named two US-based partners providing it with dApps, DeepSig and Zainar. DeepSig, a founding member of the AI RAN Alliance along with Nokia, is developing AI-aided spectrum-sensing software that optimizes the RAN, mitigates interference, and aids spectrum-sharing. ZaiNar, a physical AI startup that emerged from stealth mode earlier this year with a valuation over $1 billion and a $100 million investment, is integrating its positioning software with Nokia’s virtualized AI-RAN to enable precise network-based location capabilities. Nokia said it would name more dApp partners in the months to come.
Pointing to a growing ecosystem for AI RAN and promoting the idea of an open network (enabled by E3 interfaces) that can harness outside AI-powered innovation are both important ways to try to build operator confidence in the idea Nokia has put forth as key to its RAN strategy: the use of GPUs in the RAN, which has been met with skepticism from some operators.
This week Nokia also described the work it’s currently doing behind the scenes to arm NVIDIA’s Nemotron AI models with telecom network data, including details of Nokia’s products, operational data, and procedures as well as radio alarm and fault resolution data. Smaller models could yield agentic experts in specific network domains (RAN, core, transport, etc.) that could enable some degree of autonomous network operation – with guardrails that ensure operator control. These specialized expert agents would coordinate with agents elsewhere in the network for an end-to-end scope – a model multiple vendors have envisioned and operators have expressed interest in. And with a view to the importance of sovereignty, Nokia stressed the ability of operators to choose where different kinds of processing occur, whether in the operator network or in a hyperscaler cloud.
Addressing sovereignty is crucial for making operators comfortable with network AI, and Nokia’s emphasis on operator choice in the messaging of this point may also help it soothe concerns that operators may have about Nokia’s tight focus – narrower than some peers – on using its partner’s (and investor’s) GPUs in the RAN.
At GTC, Nokia and NVIDIA used robots to illustrate the concept of distributing AI compute in different places based on workload type and use case needs (an idea that is not specific to Nokia). In their demonstration, real-time perception and action tasks were processed on the robot itself. Fine-tuning and coordination across robots were processed at the distributed network edge. And larger training workloads were processed in more centralized compute infrastructure.
Nokia has previously emphasized the potential to integrate sensing with communication capabilities in the AI RAN (appealing to widespread and growing interest in ISAC, which is expected to be a prized 6G technology). This week, Nokia touted an AI RAN trial of an ISAC application with Finland’s Elisa, describing a dual-GPU approach within the same AI RAN platform, in which one GPU focuses on RAN operation and a separate GPU focuses on sensing and processing for an ISAC application. On previous occasions, Nokia has described GPUs in the RAN as having a dual purpose: RAN optimization and revenue growth (enabled by enterprise services or token sales). To operators that are wary of using GPUs in the RAN, the idea of using two GPUs for ISAC, whose key value is its integrated dual-purpose use of the same infrastructure, could raise questions about capital and operating costs.
Nokia also promoted recent and upcoming AI RAN trials with European operators. Finland’s Elisa has trialed the simultaneous operation of voice, mobile data and video analytics on the same RAN platform. England’s BT has agreed to a live-network trial in 2027. Germany’s Deutsche Telekom has lab-tested the use of AI-enhanced channel estimation to increase uplink throughput. And France’s Orange plans to test AI RAN in the lab and in the network. Beyond the European activity touted at this European event, Nokia separately announced plans this week for an AI RAN field trial with Japan’s Softbank in 2027.
Highlighting early operator trials are very important for Nokia’s strategy. Because AI RAN represents a real paradigm shift away from traditional RAN operation, many operators will look for trailblazing operators to do the difficult work of climbing the learning curve needed to prove out success in this area. As evidenced this week, Nokia is clearly doing that work, too.
