The GSM Association (GSMA) has formally launched Open Telco AI, a global industry initiative aimed at addressing a persistent and widening gap between the capabilities of general-purpose artificial intelligence models and the highly specialised demands of the telecommunications sector. Unveiled at Mobile World Congress Barcelona on March 2, 2026, the initiative introduces a dedicated portal that consolidates open models, datasets, compute resources, benchmarks, and community tools to accelerate the development and evaluation of artificial intelligence solutions built specifically for telecom environments.
The impetus behind Open Telco AI is a well-documented limitation: while frontier artificial intelligence models have advanced considerably in general domains, they continue to underperform on telecom-specific tasks. Many of these models struggle to interpret network data, parse standards documentation, or automate network operations with the accuracy that operators require. According to GSMA Intelligence data, only 16 percent of telecom generative artificial intelligence deployments have so far been applied to network operations, a figure that underscores how far the industry still has to go in translating artificial intelligence promise into operational reality. Louis Powell, Director of Artificial Intelligence Initiatives at GSMA, noted that artificial intelligence does not yet speak the language of telecom, and that establishing shared benchmarks while collaborating on datasets, models, and agentic systems is essential to closing this gap.
As founding supporters of the initiative, AT&T and AMD are making significant technical contributions from day one. AT&T is releasing a family of open telco models developed and trained on publicly available data, designed to be hardware and cloud-agnostic, with the goal of demonstrating that artificial intelligence can deliver tangible value across projects regardless of size or computer availability. AMD, for its part, is providing compute capacity for model training, fine-tuning, inference, and evaluation through its graphics processing unit platforms, cloud partner TensorWave, and open toolchains. The Open Telco AI portal will also host a leaderboard assessing model performance across seven telecom-specific benchmarks, alongside tools for evaluating and submitting models from local environments. Additional models featured on the platform include a radio-frequency language model called RFGPT from Khalifa University and a Large Telco Model developed by AdaptKey Artificial Intelligence built on NVIDIA Nemotron. A library of knowledge graphs, embeddings, and fine-tuning datasets has been contributed by institutions including Huawei Technologies France, Purdue University, Yale University, the University of Texas at Dallas, the University of Leeds, and others, with synthetic data pipelines provided by NVIDIA.
The initiative draws support from a broad coalition of contributing and participant partners spanning operators, technology vendors, and academic institutions, including SK Telecom, Softbank, Ooredoo, Turkcell, KPN, Deutsche Telekom, Telefónica, Vodafone, Google Cloud, IBM, China Mobile, China Telecom, China Unicom, and several universities. Community programmes form a key pillar of the effort, including competitions such as the AI Telco Troubleshooting Challenge, which drew over 1,000 registrations and announced its winners at MWC26 Barcelona. Andy Markus, Chief Data and Artificial Intelligence Officer at AT&T, described the collaboration as accelerating the industry’s path toward intelligent, automated networks by building the datasets, models, and evaluation frameworks necessary for telco-grade artificial intelligence at scale. New partners interested in participating can register through GSMA.com/open-telco-ai.
Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.
