AI RAN in 2026: What lies ahead?

• GPUs at the cell site? Nvidia is pushing AI RAN hard, but operators remain cautious as GPU economics still raise capex and opex questions
• CPUs aren’t standing still: Next-gen x86 processors are gaining AI muscle, and vendors say CPUs may be “good enough” for many RAN workloads
• AI RAN is already improving 5G today and setting the stage for AI-native networks by 2026

Nvidia made a splash with SoftBank when they announced the AI RAN Alliance at Mobile World Congress Barcelona in 2024. Since then, Nvidia has further embedded itself in the telecom space, investing $1 billion in Nokia and using 6G as an entry point.

But are we really going to see GPUs in base stations? The jury is still out on that.

“The industry is carefully evaluating the total cost of ownership (TCO) model and business cases associated with introduction of GPUs in the base station,” said Alok Shah, vice president of Strategy and Marketing for the networks division of Samsung Electronics America. “To date, the wholesale shift to GPU compute at the cell site has shown challenges in both capex and opex, but innovation in this space is active. We may find that there is a subset of locations where the math works in the coming years.”

“Certain algorithms or business models may require the installation of GPUs, whereas CPUs may be more than sufficient to meet RAN compute needs,” Shah said. “We expect operators and suppliers to carefully evaluate the benefits and considerations of these different options, just as they do today in other parts of the network.”

Bring on network efficiencies

Beyond the GPU debate, AI is certainly considered critical in today’s telecom networks – and it’s already having an impact.

“There are lots of great AI applications that can help mobile network operators (MNOs) right now on a wide range of use cases, such as accelerating root cause analysis, reducing energy consumption or planning more efficient networks,” said Rob Hughes, head of Wireless Marketing for 1Finity, a Fujitsu company.

Sure, operators can achieve some network improvements without AI, but AI does them faster and more efficiently. “Over the long term, MNOs that do not embrace AI will become increasingly more and more inefficient relative to their competitors,” he said.

At Mobile World Congress 2025, 1Finity partnered with Nvidia and SoftBank to show how they could use an AI algorithm to improve channel estimation error, boosting uplink by 20% – and in some cases, up to 50%. “These kinds of capabilities enable MNOs to offer a superior experience to their customers,” he said.

AI RAN: No need to wait

There’s a similar mindset at Dell Technologies.

“AI is becoming indispensable in telecom because networks are about to reach a ‘beyond human scale’ threshold,” said Sandro Tavares, director of Telecom Systems Marketing at Dell. “Without AI, operators cannot transform high-volume network data into timely and actionable decisions at scale. AI unlocks capabilities that are simply unattainable through manual intervention.”

Although the expectation is that AI will be baked into the 6G standard, it’s already materializing in 5G through intent-based assistants, analytics-driven automation and AI-enabled tooling that bridge RF engineering and software operations.

“The role of AI will certainly deepen with each new generation of networks. In 5G, AI acts mostly as an optimization layer. In 5G Advanced, AI becomes more integrated and proactive. For 6G, networks are expected to be AI native, with AI built directly into the core functions of control, resource management and service orchestration,” he said.

In sum, the consensus among these vendors is to expect great things from AI RAN in 2026, even if it’s still early days for the creation of 6G. 

“AI-RAN is poised to revolutionize network operations in 2026 by enabling new levels of automation and efficiency,” concluded Samsung’s Shah. “We anticipate that AI-RAN will drive significant advancements in network performance, energy efficiency and user experience.”