AI news ·
Nvidia survey finds 89% of telecom operators prioritize open AI models for cost and customization
89% of telecom operators now see open source AI models as critical to their strategy, per NVIDIA's 2026 report. The shift lets carriers run inference on their own infrastructure and avoid unpredictable per-token API pricing at scale.
NVIDIA's State of AI in Telecommunications report, published October 6, 2026, found that 89% of telecom operators now view open source AI models and software as critical to their strategy. The survey highlights a sector-wide push to cut costs, customize models with proprietary data, and maintain tighter control over AI behavior in production environments.
The report identifies five drivers behind the shift: lower costs for accessing frontier-level intelligence, the ability to fine-tune models on internal data, greater trust and transparency in model outputs, reduced vendor lock-in, and faster deployment for localized services. These factors are reshaping how carriers approach everything from network operations to customer-facing tools.
What operators are building with open models
SoftBank Corp. is using open models to develop its SoftBank Large Telecom Model, designed specifically for network operations. AT&T emphasized model flexibility that aligns with shifting business priorities rather than committing to a single vendor's roadmap. Indosat Ooredoo Hutchison built its Sahabat-AI models by adapting open foundations to local languages and cultural contexts - a capability that closed, off-the-shelf models rarely support.
NVIDIA is positioning its Nemotron family of open models as a foundation for telecom-specific AI workloads. The company pairs these models with its Agent Toolkit and NVIDIA AI Enterprise software stack, offering an end-to-end pipeline for operators moving from experimentation to production. The toolkit handles orchestration, guardrails, and deployment - areas where enterprise teams often stall without dedicated infrastructure.
Why open models change the cost equation
Running frontier-level AI through third-party APIs can become expensive at telecom scale, where millions of customer interactions and network events generate constant inference demand. Open models let operators run inference on their own infrastructure, fine-tune on proprietary data without exposing it externally, and avoid per-token pricing models that grow unpredictable over time.
For customer care teams, this means AI assistants trained on actual call transcripts, billing policies, and troubleshooting guides - not generic knowledge bases. Finance departments gain models that understand internal forecasting logic and contract terms. Sales teams can deploy assistants that reflect real product catalogs and pricing, updated as conditions change.
Why this matters for customer support, finance, and sales teams
The shift toward open, fine-tuned models directly affects frontline business functions. Customer support teams can expect AI tools that reference actual account histories and internal policy documents rather than producing plausible but incorrect answers. Finance professionals will work with models that understand company-specific forecasting methods and compliance requirements - not generic financial advice. Sales teams gain assistants that can quote accurate pricing, check real-time inventory, and adapt to regional product variations without hallucinating discontinued SKUs. When operators own the model, they own the accuracy.