Microsoft Foundry scales AT&T telecom AI on Azure

Introduction

Microsoft is using AT&T’s OTel2.0 project to show what production-scale, domain-specific AI looks like on Azure. For IT leaders and platform teams, the announcement matters because it demonstrates how Microsoft Foundry Managed Compute can help organizations run very large AI workloads faster, with more model flexibility, and at lower cost.

What’s new

AT&T built the next generation of its telecom-focused AI models, OTel2.0, using Microsoft Foundry Managed Compute on Azure.

Key details from the deployment include:

• Trillion-token scale: AT&T processed about 1 trillion tokens for OTel2.0 development.
• Open-model strategy: The company used models from the Hugging Face ecosystem, including Phi-4, OSS-120B, and Gemma 4.
• Heavy Phi-4 usage: Phi-4 alone processed more than 700 billion tokens per month for data preparation and synthetic data generation.
• Large GPU footprint: AT&T used about 530 GPUs through Foundry Managed Compute, including 430 AMD Instinct MI300X GPUs.
• Faster deployment: Foundry reportedly let teams deploy and scale in days instead of weeks.
• Lower AI costs: Microsoft says AT&T saved tens of millions of dollars by using open-source models instead of frontier models for data generation.

Why it matters for Azure customers

This is more than a telecom success story. It shows how Azure is positioning Microsoft Foundry as a platform for organizations that need to build industry-specific AI systems without managing all the underlying infrastructure themselves.

For enterprise IT and AI platform teams, the main takeaways are:

• Model choice matters: Teams can match different models to different tasks instead of forcing one model to do everything.
• Infrastructure flexibility matters: Support for mixed GPU environments, including AMD and NVIDIA, can improve cost and performance planning.
• Operational simplicity matters: Managed GPU access reduces provisioning delays and infrastructure overhead.
• Cost optimization is becoming strategic: At large scale, open models may offer a more sustainable path for training, synthetic data generation, and experimentation.

Impact on administrators and technical decision-makers

Azure administrators, AI engineers, and infrastructure teams should view this as a reference architecture for production AI. If your organization is moving beyond pilots, Foundry Managed Compute may be relevant for workloads that require dedicated GPU capacity, governance, and faster scaling.

This is especially important for teams building AI in regulated or highly specialized industries where domain tuning, approved data use, and cost control are critical.

Next steps

• Evaluate whether Microsoft Foundry Managed Compute fits your AI platform roadmap.
• Review where open models could reduce inference or training costs.
• Assess whether heterogeneous GPU strategies can improve flexibility in Azure.
• Use this AT&T example as a benchmark for planning production-scale AI deployments.

Microsoft’s message is clear: Azure wants to be the platform where enterprises build their own specialized AI systems at scale.