**AT&T and Microsoft Foundry: Powering Next-Generation AI for Telecom**
AT&T, one of the world’s largest telecommunications providers, is actively pursuing artificial intelligence to enhance its operations, from customer service to network optimization. However, the unique complexities of telecom demand more than off-the-shelf AI. Unlike other industries, telecommunications involves highly specialized networks, strict standards, and intricate operational workflows. Generic models, trained on broad internet data, often miss the mark—struggling to understand network topology, signal protocols, or billing systems.
To solve this challenge, AT&T created **Open Telco (OTel)**—a family of AI models purpose-built for telecom. OTel represents a shift from general-purpose AI to domain-specific systems that understand the telco landscape. Developing OTel2.0, the latest iteration, required massive compute power, flexible infrastructure, and a scalable platform to manage costs without sacrificing innovation. The solution came from **Microsoft Foundry Managed Compute**, enabling AT&T to deploy, test, and refine large-scale AI workloads with unprecedented efficiency.
—
### Model Choice Meets Infrastructure Flexibility
Building OTel2.0 required more than training a single large language model—it demanded a flexible, multi-model strategy. AT&T leveraged **open-source models** from Hugging Face, selecting specialized models for different stages of development:
– **Phi-4** handled massive data preparation, processing over **700 billion tokens per month** for synthetic data generation and preprocessing.
– **OSS-120B** supported higher-reasoning tasks, such as complex analysis and inference.
– **Gemma-4** played a key role in OTel2.0 development workflows.
This heterogeneous approach allowed AT&T to match each model to its ideal workload—optimizing both performance and cost.
> “Every company in the world needs to build its own AI, and that is only possible with open models and open source. AT&T is championing this vision… Microsoft Foundry makes this practical at scale.”
> — **Jeff Boudier**, Vice President of Product, Hugging Face
To support these models, AT&T utilized approximately **530 GPUs** through Microsoft Foundry Managed Compute, including **430 AMD Instinct MI300X GPUs**. This infrastructure provided the compute power needed for large-scale training and inference while maintaining flexibility as requirements evolved.
—
### Optimizing Cost Without Limiting Innovation
AI scalability introduces significant cost challenges. For AT&T, the goal was clear: **reduce model consumption costs while preserving innovation**. By adopting open models and leveraging Foundry Managed Compute, AT&T achieved a more efficient economic model centered on dedicated GPU infrastructure.
The impact was substantial:
– Processed **1 trillion tokens** for OTel2.0, including raw GSMA documents and synthetic data.
– Generated data using open-source models like Phi-4 saved **tens of millions of dollars** compared to frontier models.
– Enabled large-scale experimentation while maintaining operational efficiency and business focus.
> “When you are processing hundreds of billions of tokens, infrastructure becomes part of the problem you solve. Foundry Managed Compute gave us access to GPU capacity at scale so our teams could focus on advancing OTel2.0 instead of managing infrastructure.”
> — **Mark Austin**, Vice President, Data Science and AI at AT&T
At this scale, infrastructure is not just an operational concern—it’s a strategic asset.
—
### Accelerating the Next Wave of Production-Scale AI
OTel 2.0 demonstrates how organizations can combine **open models, scalable infrastructure, and domain expertise** to build production-ready AI systems. By aligning the right model with the right workload and optimizing infrastructure for cost and performance, AT&T processed trillions of tokens while maintaining efficiency.
As enterprises move from experimentation to production deployment, they need platforms that offer:
– **Model flexibility**—choose the best model for each task.
– **Infrastructure scalability**—deploy quickly without lengthy provisioning.
– **Cost efficiency**—optimize performance without limiting innovation.
Microsoft Foundry and Foundry Managed Compute unify these capabilities, helping organizations accelerate AI adoption at scale.
—
### FAQ
**Q: What is Open Telco (OTel)?**
A: Open Telco (OTel) is AT&T’s family of AI models specifically designed for telecommunications. It helps AI systems understand telecom networks, standards, and operations with domain-specific expertise.
**Q: Why did AT&T choose open models for OTel2.0?**
A: Open models provide flexibility, allow customization with approved telecom data, support large-scale experimentation, and reduce costs compared to proprietary models.
**Q: How did Microsoft Foundry Managed Compute help AT&T?**
A: Foundry Managed Compute gave AT&T scalable GPU capacity, streamlined deployment, and cost-efficient access to diverse GPU architectures—enabling faster experimentation and execution.
**Q: How many GPUs were used in the OTel2.0 project?**
A: Approximately **530 GPUs**, including **430 AMD Instinct MI300X GPUs**, were used through Microsoft Foundry Managed Compute.
**Q: How much token data was processed for OTel2.0?**
A: Around **1 trillion tokens** were processed, including raw documents and synthetic data generated with open-source models.
**Q: What cost benefits did AT&T achieve with open models?**
A: Using open-source models like Phi-4 saved **tens of millions of dollars** compared to frontier models, while supporting large-scale data preparation and training.
—
### Conclusion
AT&T’s OTel2.0 project highlights how purpose-built AI models, combined with scalable infrastructure, can transform enterprise AI strategy. By leveraging open models and Microsoft Foundry Managed Compute, AT&T processed massive telecom datasets efficiently, reduced costs, and accelerated innovation. As AI moves from experimentation to production, organizations need platforms that deliver flexibility, performance, and economic efficiency—and AT&T’s approach offers a blueprint for the future of telecom AI.
