AT&T says open models currently power ~25% of its overall AI usage, and expects that to rise to 70%-80% over time; AT&T uses an average of 45B AI tokens per day
The telecom company is staking its future on open models, both to control token cost and protect its proprietary data
Context & Ripple Effects
Enterprise AI deployment has been concentrated in leading proprietary providers: a Global 2000 survey found OpenAI in production at more companies than Anthropic. At the same time, Chinese models’ rising share of US-company tokens on OpenRouter shows that model access is broadening beyond the early incumbents.
AT&T is turning that diversification into a formal operating target at unusually high usage volume. Its stated rationale—lower token costs and tighter handling of proprietary data—makes model choice a procurement and governance decision, not just a technical one.
First-order effects
- AT&T will direct more model-selection, integration and governance work toward open-model deployments as it moves from roughly one-quarter of AI usage toward a 70%–80% share.
- At 45 billion daily tokens, AT&T has a strong immediate incentive to use open models where they reduce inference costs and keep proprietary data within its chosen controls.
Second-order effects
- Providers of proprietary models seeking AT&T workloads will have to compete more directly on total token cost, data protections and performance against open alternatives.
- AT&T’s target makes model-portfolio management more central: workloads will need to be assigned between open and proprietary models rather than routed by default to a single vendor.
Third-order effects
- If other high-volume enterprises adopt explicit open-model targets, enterprise AI buying shifts from selecting a primary model vendor to continuously optimizing a mixed model portfolio around cost, data control and task fit.
- The growing use of non-incumbent models, including Chinese models among US companies on OpenRouter, raises the importance of access and governance rules alongside model performance in enterprise procurement.
The trend: Enterprise AI is moving toward multi-model procurement, with open models gaining share where buyers can trade vendor dependence for lower costs and greater data control.