Telecom operator AT&T leverages Microsoft Foundry’s cloud platform and heterogeneous GPU infrastructure, including AMD Instinct MI300X chips, to build and deploy the next generation OTel2.0 telecom AI models at unprecedented scale. This approach addresses cost, flexibility, and operational complexity challenges inherent in developing domain-specific artificial intelligence.
- Unified platform managing 530+ GPUs across AMD and NVIDIA hardware
- Multi-model strategy enables tailored telecom AI with improved cost control
- Rapid deployment reduces provisioning from weeks to hours
Infrastructure signal
AT&T’s AI development for OTel2.0 demonstrates a shift towards highly scalable multi-cloud infrastructure tailored for telecom domain workloads. Using Microsoft Foundry Managed Compute, the team accesses over 530 GPUs with a heterogeneous mix including 430 AMD Instinct™ MI300X accelerators. This setup supports diverse workload requirements and optimization at scale.
The ability to deploy different GPU architectures within a single managed environment underlines the importance of flexible infrastructure that balances cost, performance, and scalability. Processing workloads at the trillion-token scale requires this level of operational agility without forcing teams to manage underlying infrastructure complexity.
Developer impact
Developers gain substantial benefits from Foundry’s unified platform that supports multi-model experimentation and rapid iteration. AT&T’s multi open-model approach—incorporating variants such as Phi-4 and Gemma-4 from open source collections—allows teams to fine-tune telecom-specific AI models tailored to distinct stages of the pipeline, from synthetic data generation to complex reasoning tasks.
This flexibility supports faster workflows by removing the need for manual orchestration of infrastructure provisioning. With flexible GPU selection and managed scale, AI teams can focus on model optimization while controlling compute costs, accelerating the path from prototype to production-ready telecom AI solutions.
What teams should watch
Organizations aiming to build specialized domain AI should consider platforms that integrate multi-vendor GPU architectures and offer managed compute to avoid siloed infrastructure. The AT&T case highlights how open models and cloud vendor collaboration can reduce operational overhead associated with deploying large-scale AI workloads while improving cost-efficiency and flexibility.
Teams in telecom and other verticals should monitor Microsoft Foundry’s evolving capabilities for hosting expansive AI workloads. Platforms that support rapid deployment and heterogeneous hardware choices enable faster development cycles and provide a foundation for agile experimentation critical in AI innovation.