Managing the cost of AI at scale: how AT&T cut AI deployment costs by nearly 90%

Telecoms operators are scaling generative AI rapidly, but the cost of large-language-model (LLM) inference is becoming the main constraint on deployment. AT&T, which processed around 8 billion AI tokens a day, has become a leading example of how to control those costs while preserving accuracy and latency.

This case study provides:

• an overview of how AT&T cut AI inference costs by nearly 90% at massive scale
• insight into the model and orchestration choices that AT&T had to make to achieve these savings and other benefits 
• an analysis of how operators can build a repeatable framework to manage AI deployment costs
• the results that AT&T has achieved from deploying such a repeatable framework.