Recently, Nokia provided a major update about integrating AI into radio access network gear — AI-RAN — in a commercial launch. Nokia’s vision is that by turning the RAN into an AI-accelerated network, it can bring a software-defined pace of innovation to cellular telephony.

A big part of this new strategy is pairing Nokia’s anyRAN software with NVIDIA’s Aerial AI-RAN platform. The real value that Nokia is trying to communicate with its new AI-RAN offering is that it believes it can double existing spectrum capacity by 2028, effectively lowering the cost per bit while also improving the user experience and creating a software-upgradeable path to 6G.

AI-Native Networks for AI-Accelerated Workloads

AI-RAN is Nokia’s approach to the reality that AI is quickly becoming a dominant workload in mobile networks, demanding greater capacity and better network economics. AI is also quickly becoming how networks themselves answer the challenges of AI. The long partnership between Nokia and NVIDIA, which was supercharged by NVIDIA’s $1 billion investment in October of 2025 is critical for both companies. NVIDIA wants to grow its telecom business and expand the role of AI in the network, while Nokia wants to accelerate its transformation and growth by partnering with a company as successful in AI as NVIDIA. Not to mention, partnering with NVIDIA should improve its chances of having enough supply of gear to satisfy its customers’ demand.

Nokia’s Ambitious Performance Goals

While Nokia’s goal of 100% spectral efficiency gain by 2028 may seem hyperbolic to some, the company is giving itself some checkpoints along the way to make attaining it possible. For starters, it says it has already gained 20% in spectral efficiency through AI-driven radio innovations. These include using AI to accelerate channel estimation, link adaptation, and carrier aggregation.

Nokia says that it will target 50% spectral gains through more AI enhancements by 2027, and that it expects to reach the 100% number by 2028. This would allow operators to improve uplink performance to address the increased needs of AI workloads without needing to refarm or acquire significant spectrum. Refarming is the act of taking existing licensed spectrum that has been used by older technology (like 2G, 3G, or 4G) and repurposing it for a newer application, usually the next generation (in this case, 5G or 6G).

Nokia’s AI-RAN Architecture Has Multiple Paths

The new AI-RAN architecture is designed to support 4G, 5G, and 6G workloads. It’s also built to be fully O-RAN compliant, enabling multi-vendor operability — which is crucial for any major operator. Nokia intends to provide three different ways of deploying anyRAN software and AI-RAN as a whole. The first is through an AirScale plug-in unit, which adds a GPU plug-in within existing AirScale baseband RAN gear. This solution should create the least friction for operators that already have Nokia AirScale units deployed and want to add AI-RAN capabilities through an upgrade that requires only minor effort.

The alternative solution is a standalone AI-RAN node completely powered by NVIDIA GPUs. This would be the industry’s first GPU-accelerated standalone node and would support 4G, 5G, and 6G workloads, which should mean that it will also offer the most scale and flexibility. Besides the standalone option, these nodes can also be deployed in clustered configurations, or alongside an AirScale as a single logical base station. This solution looks to be the most capable path, and it’s where I believe many of Nokia’s partners will move in the future, although in the short term, it may be too big a lift.

The third option in the works is a cloud-native AI-RAN COTS server solution. This would involve running anyRAN software on a Nokia pre-certified partner’s cloud-native server. While this would theoretically be silicon-agnostic, it would most likely still run on NVIDIA GPUs, though it might also support chips from other suppliers such as Marvell. This would open up an alternative supply chain while still offering cloud-native flexibility and delivering enough performance to run AI-native RAN.

Nokia’s Transition to an Update Cycle from a Refresh Cycle

Nokia’s new AI-RAN solution is also intended to help the company transition to a subscription-based business model that offers continuous software updates. Among other benefits, this may help operators manage their capital expenditure challenges when building and upgrading their networks. This subscription-based model would offer more frequent updates that enhance the AI algorithms running the network, further improving spectral efficiency and other network KPIs. This new subscription model should also, in theory, improve network performance while lowering TCO.

Operators already have long hardware refresh cycles, so it may behoove them to continue to pay for software-driven updates to their RAN if that results in better economics. This also bodes well for Nokia’s business because it moves the company away from the highly cyclical nature of network hardware upgrades, which spike for a few years whenever a new cellular generation launches, but then fall off quickly thereafter. Nokia says that pilot deployments of AI-RAN will begin at the end of 2026, with full commercial availability in 2027.