Mark Austin, vice president of AT&T, stated that the company plans to maintain the expenditures incurred by employees calling on closed-source models from Anthropic and OpenAI at the current level over the next few years by expanding the use of open-source large models such as NVIDIA Nemotron. Currently, 40% of AI requests initiated by AT&T employees are already handled by open-source models, and the company plans to increase this proportion to 60% to 70%. Open-source models perform as well as or even better than older closed-source models in handling tasks with low computing power requirements and are less costly. For example, after integrating with Lite LLM, a routing service provider, the cost of some advanced AI tasks has dropped by up to 56%, while the quality of AI output has only declined by 2%.