An open-source telecom AI model helps operators analyse network data, improve performance, and scale AI workloads using GPU acceleration.
AT&T has launched OTel 2.0, which is an open-source AI model that will help telecom operators better analyse their massive amounts of network data. Created as part of the Open Telco AI initiative in collaboration with AMD, Microsoft, and other tech companies, the model is designed specifically to analyse the telecom dataset.
This latest version runs on AMD GPU-based hardware and according to AT&T the company has processed more than a trillion tokens in both training and post-training phases. Moreover, AT&T says that the company processes more than one trillion tokens each month across its AI workloads for the expansion of generative AI applications in network planning, customer service, fraud prevention, employee onboarding, and cell tower installation.
As a telecom-oriented model, OTel 2.0 includes domain-specific training and GPU-based acceleration, improving model performance while supporting large-scale AI inference and optimisation. AT&T notes that the previous version was downloaded more than 18 million times and the latest version introduces additional improvements, but the company has not shared any technical benchmarks yet.
The program is a response to the industry’s attempt to create open AI models for telecommunications, instead of using only general language models. According to AT&T, the open-source initiative allows collaboration and gives operators an opportunity to adjust AI models to fit their own infrastructure.
For its part, AMD supports the initiative through Instinct GPUs and ROCm software platform. It provides the computing power needed to build and use telecom-specific AI models.
The Open Telco AI programme continues to expand with the participation of several telecommunication companies and technology providers. With OTel 2.0 now freely available, the participants want to inspire the creation of more tools for managing networks, automating telecom workflows, and promoting AI across telecom infrastructure.
