AT&T's AI Shift: 80% Drop in Costs with Open Models!

AT&T Inc. is rapidly integrating OpenAI models, which currently account for 25% of its artificial intelligence usage, with plans to escalate this to 70-80% over time. This move is driven by significant cost savings, with open AI models reducing expenses by 80% to 90% in certain applications.

The company employs a "smart router" to select the most cost-effective model for processing an average of 45 billion AI tokens daily. AT&T has also developed its own customized open model, OTel, reporting a fivefold return on investment for its AI initiatives this year.

Concurrently, NVIDIA Corporation's CEO is actively promoting open-weight AI, a trend that could impact the broader AI economy. The core question is whether this open-model wave threatens or merely redirects AI infrastructure funding.

While NVIDIA's strategy relies on the assumption that increased usage will compensate for efficiency gains, a broad enterprise shift to smaller, cheaper open models might slow chip demand growth despite rising usage volumes. AT&T's adoption includes models from China, raising national security discussions, yet Gartner notes that even premium proprietary models face future cost unsustainability.