Trend Shift
AT&T Uses Open Models to Cap Closed-Model Bills
The Information reports that AT&T plans to use more open models, including NVIDIA Nemotron, to keep employee spending on Anthropic and OpenAI models flat in coming years.
According to The Information, AT&T plans to keep employee spending on Anthropic and OpenAI models flat over the coming years by increasing use of open-source models such as NVIDIA Nemotron. The signal is not that AT&T is abandoning proprietary models, but that a large enterprise may be treating model procurement as a cost structure that can be segmented by task, substituted, and actively managed.
The change is a managed ceiling on model spending
The Information reports that AT&T plans to use more open models, including NVIDIA Nemotron, to prevent employee spending on Anthropic and OpenAI models from rising in coming years. This points to a change in procurement rules: proprietary models are not described as being eliminated, but as one component within a spending pool intended to remain flat. For a company with many employees and complex internal workflows, that makes AI usage a budget-governed operating decision rather than an unconstrained departmental expansion.
The mechanism is task allocation across supply layers
Open models offer an alternative to proprietary APIs priced by use. Enterprises can retain frontier external models for tasks requiring higher performance, reliability, or security while moving standardized, cost-sensitive, or internally deployable work to open models. AT&T’s reference to NVIDIA Nemotron suggests that the choice need not mean retreating from AI workflows. Instead, model routing and task segmentation can limit the share of expensive proprietary calls within overall usage.
Enterprise model markets may become portfolio markets
If more large companies follow this approach, proprietary suppliers will face not only a capability contest but continuing scrutiny over whether each workflow should remain on a paid interface. The strongest countercase is that AT&T’s plan may reflect its own scale, technical resources, and telecom operating structure rather than a broader enterprise pattern. Whether open models can reliably substitute for proprietary ones on complex tasks will also depend on deployment outcomes.
What to watch next
Next observable evidence includes whether AT&T discloses the scope, cost targets, or usage share of its open-model deployments; whether other large enterprises publicly cap proprietary-model budgets while adopting similar routing strategies; and whether Anthropic, OpenAI, or NVIDIA increasingly address open-model substitution and task-level pricing in enterprise materials. Those developments would test whether model procurement is systematically becoming more portfolio-based.