The GSMA has launched Open Telco AI, a global initiative aimed at developing AI models specifically for the telecommunications industry. The project is supported by AT&T and AMD, providing a portal for open-source models, datasets, and compute resources.

The initiative addresses a performance gap where general-purpose AI models often fail to accurately interpret network data or automate operations. According to GSMA Intelligence data from January 2026, only 16% of generative AI deployments in the sector have been successfully applied to network operations.

The Open Telco AI framework includes the following components:

• Telco Models: Open-weight models from AT&T and a radio-frequency language model (RFGPT) from Khalifa University
• Open Data: Knowledge graphs and fine-tuning datasets from Huawei, Orange, and various academic institutions
• Compute: Access to GPU platforms and open toolchains provided by AMD and TensorWave
• Benchmarks: A Telco Capability Index to measure model performance across seven industry-specific tasks

Louis Powell, Director of AI Initiatives at the GSMA, said current AI models fall short of the precision required for regulated telecom environments. He stated that the initiative provides a shared foundation to improve the accuracy and efficiency of automated networks.

AT&T has contributed a family of hardware-agnostic models trained on public data to the initiative. The GSMA stated that the collaborative approach is intended to help operators move from experimental demonstrations to production-grade AI performance at the network edge.