Telecom operators may be racing to put AI into their networks, but AWS VP Jan Hofmeyr says the industry could run into a cost reckoning if it treats generic large language models as a cure-all.
The issue isn’t just token volume, he argues, but token efficiency: LLMs were built to consume text, images and video, while networks speak in protocols, telemetry and context layers. Without a network-specific layer that can translate that complexity, operators risk paying heavily for AI outputs that still lack the accuracy, efficiency and operational value they need.
Hofmeyr says the path forward starts with breaking down data silos and using AI agents on top of a network-aware context model.
Get more of Hofmeyr’s insights – including what AI payoffs will look like - in the video above. And check out additional content from our extended interview with Hofmeyr in these stories:
Telcos are living in a bubble – and it’s holding them back
Telcos find AI’s next big use case: energy management
Telco automation and autonomy are not the same thing
Read and watch all of our coverage from DTW Ignite 2026 right here.