Justin Hotard, the company’s CEO, urged a network transformation so they can handle the variability in mobile traffic tied to AI — namely dynamic uplink and downlink, unlike traditional mobile traffic. “We’re still at the beginning of the AI supercycle, because most traffic — sessions and tokens — involves human-enabled interaction, human-machine in nature. That includes chatbot usage, early-stage agents and still-potential physical AI,” the executive said.
In terms of figures and hard data, there are currently around 1.3 trillion AI sessions per year (discrete AI interactions), more than 100 trillion tokens per day (the minimum unit of text an AI model uses to process language), 77 Exabytes of monthly AI traffic and 53.5% of the latter on mobile.
“What’s coming next is machine-to-machine interaction. That’s when we’ll see a tremendous acceleration in the number of tokens and sessions, and that’s why planning the network transformation is essential,” Hotard warned.
Under this vision, in the presentation ahead of the start of MWC, Nokia outlined milestones following the agreement reached with Nvidia in 2025. The Finnish company carried out functional trials of AI workloads on the GPU-accelerated AI-RAN platform powered by Nvidia, with T-Mobile, NTT Docomo, Indosat, Vodafone, Elisa, SoftBank and BT.
In the case of T-Mobile in the United States, the trial was conducted at the operator’s AI-RAN Innovation Center in Seattle, with applications such as video streaming, generative AI queries and AI-based video subtitles. The test demonstrated simultaneous AI and RAN processing on a single Nvidia Grace Hopper 200 server, using accelerated AI-RAN workloads.
Pallavi Mahajan, the Finnish company’s CTO, said commercial trials are scheduled by the end of 2026, with the first commercial launch expected in 2027.
In Indonesia, operator Indosat announced a year ago that it was working on trials with Nokia and Nvidia. Yesterday, the company’s CEO, Vikram Sinha, took part in a panel discussion during Nokia’s event. “We started with AI-RAN and will evolve toward an AI Grid (in reference to a ‘network’). We have 55,000 sites, which will become a grid.”