The era of AI-native RAN starts now

Innovation happens when engineers, scientists and researchers build on the work of their peers and predecessors. The easier it is to build on an idea, the more consequential the innovation can become.
 
One of the best examples of this is the internet. 

The internet was built on open standards, such as TCP/IP and HTTP, that anybody could use. The innovative minds responsible for the internet did not invent Google, Amazon or Facebook (to name just three examples), but by building on their work those companies are now worth almost $9 trillion and employ just shy of 2 million employees worldwide. 

The inventors of the internet provided an open, programmable foundation. Once millions of developers could build on that foundation, innovation exploded. In other words, the platform creators enabled far more value than they could have created themselves. 

Nokia’s ambition is to apply those same principles to networks in the AI era.

We have developed the industry’s first AI-native RAN platform. Because it is open, programmable, extensible and secure, developers, enterprises, cloud providers, startups and operators can use it to create applications and services that neither Nokia nor network owners could have imagined individually.

With AI-RAN, we deliver:

• An open, programmable platform that supports distributed real-time millisecond-level applications (dApps) at the cell site, exposes data for third-party AI applications and is ready for integrated sensing and physical AI.
• A step change in spectral efficiency with advanced AI algorithms running on AI-accelerated merchant silicon at radio timescales.
• Three deployment paths on one AI-RAN platform, all using the same software foundation and sharing the same 6G destination.
• RAN innovation at software speed on our anyRAN software platform, which is ready for AI models that continue evolving in the future.

An open, programmable platform

The AI era creates a unique opportunity to fundamentally reimagine the RAN as a programmable platform for innovation. By exposing data for real-time AI applications, supporting third-party dApps and enabling open interfaces and APIs, AI-RAN creates a foundation for innovation and emerging capabilities such as integrated sensing, distributed intelligence and physical AI.

We see a broader customer need for open, scalable and increasingly autonomous networks, including better use of network data. Through our work with the AI-RAN Alliance and 3GPP, we help ensure that software-defined evolution is built on industry-defined reference architectures and open, standards-based processes that customers can trust. Furthermore, we provide full support for O-RAN and any O-RAN-compliant radios from the Open RAN ecosystem partners. 

A step change in spectral efficiency

Spectrum remains one of the scarcest and most expensive resources in mobile networks. Spectral efficiency is becoming a boardroom metric – and as efficiency increases, so does the extensibility of a network, and the potential number of use cases and capabilities that it can support.

Instead of waiting for the next spectrum auction, operators must unlock more capacity from the spectrum they already own, especially for the most demanding cells.

At Nokia, we are committed to delivering twice as much spectral efficiency by 2028 from the spectrum assets that telecom providers already own. We are validating our work in structured proof-of-concept programs with leading telecom providers including T-Mobile U.S., SoftBank and Indosat. 

With AI-RAN, we introduce an architectural shift that pushes spectral efficiency beyond conventional RAN architectures. Our software-defined baseband utilizes accelerated merchant silicon, and we have an expanding portfolio of advanced AI algorithms running at radio timescales. These algorithms can continuously improve how the network learns, predicts, schedules, transmits and receives radio signals.

We have already proven 20% combined spectral efficiency gains with AI-assisted RAN algorithms on existing platforms:

• AI-assisted link adaptation continuously optimizes how the network interprets radio conditions and improves modulation decisions. Customer field trials have demonstrated up to 13% throughput gains across all users and up to 18% gains for fixed wireless access users.
• AI-powered channel estimation improves the network’s understanding of radio conditions in real time, enabling better signal recovery and more effective use of uplink spectrum. We have measured 5-10% uplink spectral efficiency gains in demanding propagation environments.
• With AI-based carrier aggregation, the network learns which carrier combinations are likely to deliver the highest spectral efficiency and the best results for each user. In initial tests, we observed 5% improvements in user throughput.

These results prove that AI brings real improvements in the RAN. But incremental optimization is starting to reach its limits. We need AI for millisecond-level optimization in frequency and time domains across multiple antenna layers and beams serving an increasing number of connected devices and AI systems.

Teams across Nokia Bell Labs and product development are already working on the next AI-powered innovations that can further improve RAN performance and spectral efficiency. These include:

• Deep receivers (DeepRx) improve the network's ability to recover and decode data in challenging radio conditions, reducing pilot symbol overhead and enabling more radio resources to be dedicated to user traffic. Bell Labs has demonstrated 10% throughput gains compared to conventional receiver architectures.
• Deep transmitters (DeepTx) use AI to make transmission decisions more intelligently, improving beamforming accuracy, anticipating channel variations and reducing radio overhead. DeepTx concepts have indicated throughput gains in the order of 10% through more efficient signal design, creating a foundation for future AI-native air interface innovations.
• CSI feedback compression uses machine learning to reduce the amount of wireless channel information that a device needs to send back to the network – like sending a ZIP file rather than an entire folder of information. Proof of concept with some of Nokia’s co-innovators showed spectral efficiency gains of 10% for the same feedback budget.

Together, these algorithms create a compounding effect across the radio stack, converting a greater share of available spectrum into usable network capacity. And we are not stopping there. Our teams are already researching and testing the potential gains with AI-assisted RRM parameter optimization and inter-cell interference coordination. 

Three deployment paths on one AI-RAN platform

Our AI-RAN platform is founded on the different deployment realities of radio access networks. It is fully compliant with Open RAN standards, supports multi-vendor deployments and provides telecommunication providers with a shared anyRAN software roadmap across three hardware deployment paths:

• First, AirScale ABIG builds on what our customers already have. It is an AI-accelerated plug-in card that fits our customers’ AirScale baseband installed base, enabling a lean, investment-protected migration path to AI-RAN. 
• Second, the Accelerated AI-RAN node is a completely new product category for our industry. It is a high-capacity, all-in-one solution that can be deployed standalone or alongside AirScale as a single logical base station, providing a scalable path from 4G and 5G to 6G.
• Third, COTS server Cloud AI-RAN is designed for distributed and centralized cloud-native deployments. It runs on GPU-powered servers from leading vendors, including Dell, Supermicro and Quanta, introducing accelerated computing capacity for distributed inference and AI Grid deployments, where infrastructure can be shared across AI workloads and RAN.

Telecom providers are starting from different places. Some need the fastest route from an existing installed base, some need high-capacity expansion and standalone nodes, and some have a cloud-native strategy. With a common anyRAN software roadmap, these become three deployment choices on a single software-defined platform, providing a seamless path to 6G.

RAN innovation at software speed

The AI era moves too quickly for innovation – and security – to remain tied to traditional hardware refresh cycles. New AI models, threats and optimization techniques emerge continuously, while network infrastructure has historically evolved over years. That growing mismatch creates a strategic challenge for telecom providers who want to provide a platform trusted by customers.

Built on our anyRAN software foundation, Nokia AI-RAN enables networks to evolve through software. New capabilities, AI algorithms, security enhancements and performance improvements can be introduced continuously. 

With AI-RAN software subscriptions, customers can turn software innovation, spectral efficiency enhancements, AI-driven optimization capabilities and security upgrades into real network gains without waiting for the next hardware generation.

The beginning of the AI-native era

Networks are moving from carrying traffic to adapting around intelligence. We are evolving from RAN architectures optimized for prior-generation traffic and hardware-cycle upgrades to software-defined, AI-accelerated RAN. 

Building on that foundation, our open, programmable, extensible and secure AI-native platform embeds intelligence deeper into the network while providing transformational economics and a practical bridge to 6G. It will be an invaluable asset to innovators, engineers and builders for the duration of the AI supercycle and beyond.

This is the beginning of the AI-native era. This is Nokia AI-RAN.