← Back

Wire · founder news, decoded · opportunities

AT&T and Microsoft scale trillion-token workloads with Microsoft Foundry and AMD

Published

23 July 2026

Topic

opportunities

Geography

United States

Source

Read at azure.microsoft.com

Verified

Fusion42 · 23 July 2026 · Fusion42 review

AT&T built Open Telco 2.0, a telecom-domain AI model, using Microsoft Foundry Managed Compute with 530 GPUs (430 AMD MI300X) to process trillion-token workloads across multiple open models including Phi-4, enabling faster iteration and lower costs than traditional infrastructure approaches.

The Wire takeaway

You've just watched a $2bn telecoms operator choose managed GPU over owning infrastructure, and do it faster and cheaper. That's the death knell for captive AI compute—the market is moving to rental, which means you can now build AI products without owning silicon.

Related on Wire

Topics

gpu-capacity · foundational-models · telecom-ai · cost-optimization · managed-compute