Samsung Networks: AI ‘tsunami’ will reshape networks but GPU economics still unclear

• Samsung says AI will fundamentally reshape telecom networks and operations
• GPUs have a role in AI-RAN, but the business case is yet to be determined
• 6G needs to be incremental and software-driven, says head of Samsung Networks

Telecom operators face a “tsunami” of change as they look to build AI-native networks – but not every use case requires the deployment of GPU-based infrastructure, according to president and head of networks business at Samsung Electronics, Woojune Kim.

With AI-RAN high on the agenda during and following the recent MWC26 event in Barcelona, operators are assessing their network investment strategies to make sure they can take full advantage of AI and deal with the data traffic they might need to support in the AI era. 

Speaking to select journalists during a breakfast briefing in Barcelona, Kim said: “The biggest question around AI has been, is it real? Can I really use it? Now operators are in the middle of trying to carry out a wholesale transformation.”

Kim, who leads Samsung’s networks business, went on to say that he views MWC26 as “transformational” for telcos. “I hope they go back and realise that a big tsunami is coming, and this is different from any previous generation transitions.”

This transition, he added, is moving much faster than the usual telecom product cycle, which is a big challenge for Samsung customers. He cited the rapid shift from simple large language models (LLMs) to copilots to agentic AI – all in the past four years. This is being accelerated by the use of AI for coding and other workloads that previously took longer.

“The three-year plan that you have right now is based on what you thought five months ago,” he added. “Next year, that three-year plan could be thrown out completely.

“Operators have now stopped thinking about how to change their networks to do that stuff and are instead asking us if our systems are ready for it,” he said.

Operators are still in the exploratory stage, trying to determine what is coming down the line and identify the tools they will need for their networks to be able to cope with the demand. For Samsung, its position as a huge tech firm means it is developing AI across various verticals, including developing AI chips. “That gives us a stronger base of AI expertise to help them with their questions,” he added.

The network, he expects, will follow the wider AI trends, developing from AI chatbots to copilot and agentic solutions, then ultimately with more complex “robots” or “digital humans” that could control the network operations.

Kim added: “These days, organisations have software agents that help do the workload. You tell the agent what you want and when you come back later, it is done. The network side will follow these trends.”

The GPU conundrum

Yet while Kim acknowledged that the direction of travel is towards AI-driven and increasingly autonomous networks, he also suggested that some of the more immediate infrastructure questions – particularly around the deployment of graphics processing units (GPUs) in the radio access network (RAN) – remain unresolved.

The idea that GPUs will be a key component in future RAN deployments was pitched by the founders of the AI-RAN Alliance in early 2024: Those founders included Nvidia, the world’s leading developer and supplier of GPUs, and a number of other influential RAN market players, including Ericsson, Nokia and Samsung. 

Since then, the debate has raged as to whether such deployments make strategic and economic sense, though there is a steady trickle of operators that are showing interest as Nokia (in which Nvidia took an equity stake last year) pitches the case for its RAN platform based on Nvidia’s Aerial RAN Computer Pro (Arc-Pro) platform. More recently, Nvidia added to its case for the deployment of GPUs in wide area network infrastructure, including the RAN, with the launch of AI grids, an architecture that is being put through its paces by a number of major operators – see Nvidia plots AI grids with operators, partners.

Kim confirmed that Samsung has explored the potential of the AI-RAN model, and was keen to stress that the South Korean firm has a good relationship with Nvidia – a company with which it has worked for more than 25 years and with which it is building an AI-focused semiconductor factory.

Even at MWC26, Samsung was advancing its relationship with the chip firm, announcing a multi-cell test combining its virtualised RAN software with Nvidia’s AI platform and jointly demonstrating AI-RAN running on Nvidia infrastructure.

“We do have a close partnership with Nvidia so we look at it together,” Kim said when asked about GPUs. “We just haven’t solved the economics of it.”

He added that Nvidia’s approach reflected a broader strategy to extend accelerated computing across multiple industries, placing long-term bets on where AI workloads will emerge.

“Their goal is to put GPUs everywhere in the whole industrial ecosystem. Then people start using it, and that locks them in,” he added. “AI does not equal GPU.”

He indicated that this does not translate into a need for GPU deployment across all parts of the RAN. Instead, the use of GPUs is likely to be more targeted – notably in high-density or high-value areas, where the economics make more sense.

“We’re not saying you don’t need GPUs at the edge. We have a platform that shows you can get performance enhancements… [but] there are a lot of AI use cases where we have the AI running on the CPU [central processing unit]. GPUs have a higher hurdle to prove that you get value.”

Kim added: “If you were to put it in a hotspot, a stadium… maybe putting a GPU there is going to make a lot of sense. But does it make sense in the middle of suburbia? I’m not so sure… we are exploring those applications.”

Activating 6G

Kim suggested the shift towards more flexible, software-defined networks could reshape operator investment cycles – most notably around 6G.

“6G should be activated, not installed,” he noted. “I don’t think we are going to go into the traditional 10-year bump, but rather see a steady state of investment.”

Rather than triggering a new wave of large-scale infrastructure upgrades, he indicated that future network evolution is more likely to be incremental, with operators enabling new capabilities through software rather than wholesale hardware replacement.

Kim also questioned whether 6G will deliver a fundamentally new service proposition, arguing that much of the core connectivity challenge has already been addressed.

“Tell me, is that a new service, or is it just more of what you have?” he said.

While operators acknowledge the need to continue investing in their networks, Samsung’s partners are focusing on improving coverage, performance and efficiency with existing infrastructure – a trend reflected by TelecomTV’s coverage of MWC26.

- James Pearce, Contributing Editor, TelecomTV