As doors close on the twentieth MWC in Barcelona, Contributing Editor Annie Turner looks at the potential implications of what happened for future networks.

This year’s MWC had a high degree of excitement and purpose about it. This was in no small part due to the huge presence of the world’s largest company by market cap, NVIDIA, and its charismatic CEO, Jensen Huang. The company set the scene ahead of the show, publishing its fourth annual State of AI in Telecommunications report in mid-February.

AI everywhere?

NVIDIA’s report suggests that telecoms is embracing AI more than most other sectors. For this report, it interviewed 1,038 professionals between September and November last year, 27% of whom work for operators. Here are some of the highlights from its survey:

• 90% said AI is helping increase annual revenue and drive down costs
• 77% said they expect to see AI-native networks launch before the deployment of 6G
• 65% of telecom operators said network automation is being driven by AI
• 60% said their organisation is using or assessing generative AI, up from 49% in 2024
• 89% said open source models and software are important to their AI strategy
• 89% of telcos plan to boost AI spending in 2026, up from 65% a year ago.

  

And yet operators’ reported gains from AI investments have not made step changes in profitability, which is not to say that small improvements cannot make big differences.

This was underlined by Orange’s group CEO, Christel Heydemann when the operator reported it earnings for 2025 in February. She highlighted “significant” operational improvements driven by AI, which contributed to a 2.7% increase in interest, tax, depreciation and amortisation, after leases ( EBITDAaL) and a continued focus on reducing churn.

She pointed out that 1% reduction in churn could lead to as much as a €40 million improvement in earnings before EBITDA in France alone. The group’s aim is reduce churn rates by 3% points across its European opcos.

Accenture released research at MWC was also food for thought, namely that about 79% of all operators are still at Levels 0 or 1 in network automation. That said, a relative handful of operator groups (out of many hundreds) serve most of the world’s subscribers – for example, the top 10 operators serve collectively serve 3.9 billion consumers (out of a total of about 8.3 billion).

All of which suggests: a level of immaturity among those who are going for it with AI in the network and operations; that there is a substantial market yet untapped; the hype gap persists and we need more empirical evidence on ROI; and that is takes time to apply AI and things don’t stand still in the meantime. That brings us to the next big theme.

Telecoms according to NVIDIA

Huang’s grand plan is to turn the network into an intelligent grid, from end to end, as AI descends from the cloud and 6G integrates with the physical world (definite shades of predications made about use cases for 5G that have not materialised). The GPUs will simultaneously deliver connectivity, AI inference and autonomous network operations in the RAN.

The new network architecture will be built around accelerated computing which uses specialised hardware to handle intensive tasks while central processing units (CPUs) orchestrate and manage functionality.

The most famous type of specialised hardware for accelerated computing is GPUs, which NVIDIA utterly dominates, but also includes field-programmable gateway arrays (FPGAs) for programmable logic and application-specific integrated circuits (ASICs). They all use parallel processing for speed, while CPUs operate sequentially.

And Huang’s vision is not just theory. Last October Nokia signed a strategic partnership deal with NVIDIA through which the chipmaker will invest $1 billion in Nokia to accelerate AI-RAN innovation and the two intend to lead the transition from 5G to 6G. At MWC Nokia, demo’d AI-RAN on GPU-based platforms that showed tangible improvement in baseband processing and network optimisation.

The partners are working with operators, including BT, Elisa, NTT DOCOMO and Vodafone Group to adopt AI-RAN technologies. Further, the partners have undertaken new customer integrations and successful functional tests of Nokia anyRAN software on NVIDIA GPU-accelerated AI-RAN platform with T-Mobile, Indosat and SoftBank Corp.

Justin Hotard, President and CEO of Nokia commented, “AI is the new workload reshaping networks. That shift requires architectural change across every layer – including the radio.

“AI-RAN transforms RAN into a software-driven platform optimized for AI, and with NVIDIA and a growing ecosystem of partners we are progressing from validation to commercial deployment. This is a foundational step toward AI-native networks and 6G.”

Open RAN is dead, long live open RAN II?

There’s another important strand in this new architecture proposed by Huang, which has echoes of open RAN. As the noise around open RAN died down, support for an open platform in which operators could integrate the best-on-class components from multiple suppliers has evaporated. The idea was to increase competition, boost innovation and reduce cost.

The reality was high integration costs, complex interoperability issues and perceived better performance from the established, single-vendor, closed systems.

Now, alongside the proposed rethink of the RAN’s architecture and purpose, as outlined above, there a shift towards ‘open compute’ in the shape of the Open Centralized Unit Distributed Unit (OCUDU) Ecosystem Foundation under the aegis of the Linux Foundation. It has been described as the “Linux of RAN”.

The OCUDU project’s mission is to speed up innovation in AI-RAN. In particular, it will be concerned with standardising how accelerated infrastructure runs telecoms workloads. Put another way, it wants to define a unified, industry-wide approach to the compute layer that will runs the network.

The two biggest potential icebergs for this Titanic ambition are integration and operations, but geopolitical tensions could also result in fragmentation. You can read more about OCUDU’s origins here.