AAI 2026: Open Telco AI Optimizes Models, Infrastructure and Token Costs

Using AMD hardware, AT&T trains additional open, telco-specific models in Microsoft Azure as the Open Telco AI initiative moves from industry vision to practical deployment.

Published July 23, 2026

AI Roundup Editorial summary, highlights, and FAQExpand

Key HighlightsSummaryFAQ

What exactly is the news?

AMD is highlighting an AT&T progress milestone. AT&T says it has processed more than 1 trillion tokens through Microsoft Azure-hosted AI models running on AMD Instinct GPUs and is training additional open, telecommunications-specific models using AMD hardware, Azure managed compute and GSMA data. The update shows the Open Telco AI initiative moving from an industry framework toward practical work at scale.

What is new since Open Telco AI launched at MWC 2026?

The MWC launch established the shared foundation: open telco models, data, compute, benchmarks, tools and an industry community led by GSMA. The Advancing AI 2026 update adds concrete execution: large-scale model processing in an Azure-hosted AMD environment and additional model training on AMD hardware. It is a new progress milestone within the existing initiative, not a separate or exclusive alliance.

Why do telecom operators need telco-specific AI models?

General-purpose models can struggle with network data, telecommunications standards and operational context. Telco-specific models can be trained and evaluated on domain data and realistic tasks, giving operators model options that are better matched to accuracy, performance and cost requirements. In this announcement, “telco-grade” describes the intended domain fitness, reliability, safety and controls for a use case; it is not a formal certification.

What is AMD’s specific role in the collaboration?

AMD provides the compute and open software foundation for training, fine-tuning, inference and evaluation. In this phase, AMD Instinct GPUs run the Azure-hosted models, while the open AMD ROCm software platform supports the development and deployment ecosystem. AMD’s broader portfolio across CPUs, GPUs, networking and adaptive computing gives operators options from core data centers to distributed edge environments. AT&T contributes models and telco expertise, GSMA contributes data and benchmarks, and Azure provides managed compute.

What happens next, and how does this fit AMD’s broader AI strategy?

The work moves forward through continued training, fine-tuning, evaluation, model publication and deployment testing. For AMD, this is a customer-led proof point for open software, cloud infrastructure choice and a compute portfolio that can scale from data center to edge.

AT&T has processed more than 1 trillion tokens through Azure-hosted AI models running on AMD Instinct GPUs.
AT&T processes an average of 45 billion tokens daily and uses an intelligent gateway to match each task with an appropriate model.
AT&T is training additional open, telco-specific models on AMD hardware using managed Azure compute and GSMA data.
AMD Instinct GPUs and the open ROCm software platform support model training and inference at scale.
AMD, AT&T, GSMA and other industry leaders launched the initiative at MWC 2026, with progress now shifting toward practical deployment and measurable outcomes.

AT&T has processed more than 1 trillion AI tokens on AMD Instinct GPUs in Microsoft Azure while training additional open, telecommunications-specific models using GSMA data. Its cache-aware AI gateway processes an average of 45 billion tokens daily and has reduced AI costs by as much as 80%, demonstrating the Open Telco AI initiative’s move toward scaled, practical deployment.

What’s the News? At Advancing AI 2026, AMD, AT&T and Microsoft announced OTel 2.0, an open-source model trained specifically for telecoms. AT&T has processed more than 1 trillion tokens through managed compute in Microsoft Foundry, leveraging AMD Instinct™ GPUs to train the OTel 2.0 models, moving telco-trained models from vision to scale. This progresses the dedication by AMD, AT&T, GSMA and other industry leaders to provide the telco industry with models specifically trained for its needs, announced at MWC 2026.

Why It Matters? The new OTel 2.0 model, trained on data provided by the Open Telco AI initiative, is the next step in bringing together models, datasets, benchmarks, tools and compute to help operators interpret network data, understand telco standards and support complex operations. OTel 2.0 is the largest and best performing open-source model for the telco industry to use. The approach can support more automated operations, faster troubleshooting, improved resiliency and better customer experiences.

What’s the Role of AMD?

AT&T’s progress shows what becomes possible when open collaboration, telecommunications expertise and right-sized compute come together. By combining AMD Instinct GPUs, the open AMD ROCm™ software platform and flexible compute from core to edge, we can help operators build and deploy telco-specific AI at scale while balancing accuracy, performance and cost.

— Dan McNamara, senior vice president and general manager, Compute and Enterprise AI, AMD

How Does AMD Enable Scale? AMD Instinct GPUs and the open AMD ROCm software platform provide AT&T a foundation for model training and inference. AT&T processed 1 trillion tokens through managed compute in Microsoft Foundry to identify the most meaningful 400 billion tokens and used those tokens to post-train the new OTel 2.0 model on AMD hardware. AT&T says the training on OTel 2.0 has made it the largest and best-performing open-source model in the Open Telco AI initiative.

What’s Next? The next chapter of telco AI will be measured by practical outcomes, not model size alone. Open collaboration, industry expertise and flexible compute can help operators automate network operations, troubleshoot issues faster, improve resiliency and deliver better customer experiences.

More: Advancing AI 2026 (Press Kit)

TopicsArtificial Intelligence Data Center Advanced Technologies #rocm #instinct #2026-aai

Share / cite

Press inquiries: corporate.pressinquiry@amd.com

More from this category

AAI 2026: AMD CEO Lisa Su Launches End-to-End AI Solutions (Keynote Livestream)

AAI 2026: AMD Launches AMD Helios Rackscale Solution for Frontier AI

AAI 2026: AMD Launches AMD Instinct MI400 Series GPUs for Frontier AI, HPC

AMD and OpenAI: Collaborating Across Every Layer of the Stack