Telecommunications and Artificial Intelligence: Machine Learning Inside the Network, Language Models at the Customer Edge, and a 6G Standard Designed Around AI

Purpose. Telecommunications is both a user of artificial intelligence, in running networks and serving customers, and the carrier on which AI services run. This article asks where AI stands in the industry in 2026 and where it may be by about 2035. Design and methods. A review combining the share of the telecommunications research literature that involves AI (OpenAlex, 2010–2025), Eurostat's survey of AI use in EU telecommunications enterprises (2021–2025), the standards work of 3GPP and the International Telecommunication Union, a benchmark of language models on telecom knowledge, an operator's reported energy-saving trial, and the regulatory position. Claims taken from the literature were checked against the sources cited. The scenarios are judgements built on this evidence. Findings. The share of telecommunications works that mention AI rose from 4.2% in 2018 to 20.9% in 2025, 5.0 times as much, and unlike most industries only 8.0% of these AI works involve language models, which shows that AI in networks is mostly not generative AI. In the EU, 39.8% of telecommunications enterprises used AI in 2025, double the 19.9% for all enterprises, after being flat between 2021 (19.8%) and 2023 (19.2%). Their use for ICT security (13.6% of enterprises, against 3.9% for all enterprises) stands out. Standards bodies have written AI into the technology: the first release of 5G-Advanced included study and work items on AI, including AI/ML for the radio interface and for network energy saving, load balancing and mobility, and the ITU framework for 6G adopted in 2023 includes a usage scenario for integrated AI and communication. Reported benefits are mostly from vendors and operators: one announcement reported AI-based energy saving of 13.5% per site against 2.8% for baseline features, without stating the number of sites or the duration. General language models were found to struggle with complex questions on telecom standards. We expect AI-assisted operations and energy saving to be routine in networks by 2028, AI-native design to be built into the 6G standard in 2029–2032, and the commercial question of how operators earn from AI traffic to be open after that. Originality. The article separates the machine learning inside networks from the language models at the customer edge, shows that the research literature in this industry is less about generative AI than in others, and is explicit that most reported benefits are vendor or operator claims.