NVIDIA Unveils 30-Billion-Parameter Nemotron 3 Large Telco Model; Survey Finds 89% of Operators Say Open Models Important

Executive Summary

• Telecom operators are increasingly building their AI strategies on open models, with 89% of respondents to NVIDIA's latest State of AI in Telecommunications report saying open source models and software are important to their company's AI strategy, according to NVIDIA Blog.
• NVIDIA announced the 30-billion-parameter Nemotron 3 Large Telco Model (LTM), fine-tuned by AdaptKey on open source telecom datasets to improve accuracy on telecom-specific tasks, per NVIDIA Blog.
• NVIDIA also released a full recipe covering the end-to-end fine-tuning pipeline for adapting open models to operator-specific networks, customers and procedures using NVIDIA NeMo open libraries, NVIDIA Blog reports.
• SoftBank Corp., AT&T and Indosat Ooredoo Hutchison are named in the source as operators working with open models on telecom-specific AI, model-choice strategy and local-language AI respectively, according to NVIDIA Blog.

Key Takeaways

• The strategic case operators make for open models is fivefold in NVIDIA's framing: lower-cost access to frontier-level intelligence, telco-specific customization, trustworthy and governable AI, flexible and secure deployment across cloud, private and edge environments, and the ability to deliver locally adapted AI services.
• NVIDIA positions Nemotron as an open foundation layer for telecom operations, with open weights, training data and recipes, plus speech capabilities for voice applications.
• Named operators describe different priorities rather than a single use case: SoftBank Corp. points to network operations, design and management; AT&T frames the issue as matching each workload to the right mix of performance, cost and control; Indosat Ooredoo Hutchison emphasizes Indonesian language, culture and data.
• NVIDIA argues models alone are insufficient, pointing to data pipelines, agent orchestration, secure runtimes and simulation as the additional layers required to move autonomous telecom operations into production.

Why Operators Are Choosing Open Models Over Single-Vendor Lock-In

The source frames the shift around control rather than price alone. NVIDIA states that open models give operators the ability to trust, control and customize AI across critical workloads spanning autonomous networks to customer care, and that the strategic value is fivefold: expanding access to frontier-level intelligence at lower cost so closed models can be reserved for the highest-value workloads; supporting telco-specific customization through open weights and training recipes; enabling trustworthy AI through greater visibility into and control over model artifacts and behavior; enabling flexible, secure deployment across public clouds, private infrastructure and edge environments; and unlocking locally adapted AI services for enterprise and government customers.

The cost argument is qualified rather than absolute. NVIDIA says lower-cost open access lets operators reserve closed models for workloads where those models drive the most value, a portfolio approach rather than wholesale replacement. The source also cites independent benchmarks, the Artificial Analysis Intelligence Index v4.3.2, as showing leading open models becoming more competitive across demanding reasoning, coding, scientific and agentic workloads. That benchmark reference is presented by NVIDIA; the underlying methodology is not detailed in the source text.

The governance argument carries direct procurement weight. NVIDIA states that open models can be evaluated, adapted and governed in alignment with regulations and business policies because operators gain visibility into and control over model artifacts and behavior. For regulated network operators, that framing positions open weights as a compliance and audit mechanism, not only a technical preference.

Nemotron 3 Large Telco Model and the Fine-Tuning Recipe

The concrete product announcement in the source is the 30-billion-parameter Nemotron 3 Large Telco Model, fine-tuned by AdaptKey on open source telecom datasets. NVIDIA says the model is intended to give operators an open baseline that can understand telecom industry terminology and reason through telecom operations workflows such as network configuration and customer incident triage.

NVIDIA also released a full recipe walking through the end-to-end fine-tuning pipeline for adapting open models to operator-specific networks, customers and procedures using NVIDIA NeMo open libraries. The company describes the Nemotron family as providing frontier-level reasoning performance optimized for agentic workflows, as well as speech capabilities for voice applications, with open weights, training data and recipes. Those performance and capability descriptions are NVIDIA's characterizations; the source does not provide independent verification of the model's benchmark results.

The practical distinction is between a starting point and a finished system. NVIDIA presents Nemotron 3 LTM as an open baseline and the recipe as a customization path, leaving operators to supply their own operational data. The source does not state what accuracy gains were measured, on which tasks, or under what evaluation conditions beyond describing the fine-tuning as delivering accuracy gains for telecom-specific tasks.

Related: Critical Minerals Market Size, Trends and Forecast Statistics by Country 2026-2030

SoftBank, AT&T and Indosat Ooredoo Hutchison on Model Choice

SoftBank Corp. is described in the source as an illustration of an operator using open models as a basis for developing and continuously advancing telecom-specific AI capabilities. Rajeev Koodli, principal fellow of SoftBank Corp. and senior vice president of SB Telecom America, said open models allow SoftBank Corp. to build on the rapid progress of global foundation models while applying network knowledge and operational expertise accumulated over many years. He added that the company is using open foundations extensively, including NVIDIA Nemotron models and others, in developing its SoftBank Large Telecom Model for telecom-specific use cases such as network operations, design and overall management.

AT&T's framing emphasizes workload matching. Andy Markus, chief data and AI officer of AT&T, said the future of AI is not about choosing a single model but about intelligently matching every workload to the right combination of performance, cost and control, and that open models are essential to that approach. He said AT&T can bring that model-choice strategy into production with NVIDIA at the scale, reliability and governance its business requires.

Indosat Ooredoo Hutchison represents the localization case. Chirag Sukhadia, chief data and AI officer of Indosat Ooredoo Hutchison, said the value of open models for countries like Indonesia goes beyond access to powerful AI and is about adapting that intelligence to local language, culture, data and real-world needs. He said the operator's Sahabat-AI family of open source models uses open models as a foundation to build AI that understands Indonesia.

For deeper context, see our Health Tech analysis: "Scale AI Partners With Mayo Clinic to Deploy Clinical AI Systems".

Production Workflows and Local AI Platforms in Telecom

NVIDIA's argument is that open models are a critical building block but insufficient alone to bring autonomous telecom operations safely into production. The source identifies two requirements beyond the model layer: pipelines that prepare and protect data for fine-tuning by anonymizing sensitive records and generating privacy-preserving synthetic datasets, and a platform that turns open models into governed, autonomous agentic workflows.

NVIDIA states that it provides that end-to-end platform powered by NVIDIA AI Enterprise software and NVIDIA Agent Toolkit, spanning data pipelines, open models, agent orchestration, secure runtimes and simulation, supported by a partner ecosystem building at each layer. The company describes this as a clear path for operators to translate open model benefits into production-ready AI workflows. This is a vendor-stated capability description; the source does not include deployment timelines, operator-reported production results or independent evaluation of the platform.

The second commercial thread is local AI services. NVIDIA says operators building AI infrastructure aligned with national AI strategies can use open models to deliver services tailored to local languages, industries, regulations and data governance requirements, and can host open models on their trusted platforms for customers to consume directly or build on with their own data and applications. Indosat Ooredoo Hutchison's Sahabat-AI is the named example of this pattern.

Additional coverage: Alcolase Raises €1.

What This Means for Practitioners

For telecom procurement and platform teams, the source reframes model selection as portfolio allocation rather than a single-vendor decision. AT&T's stated approach of matching each workload to a combination of performance, cost and control implies operators should define which workloads justify closed-model spend and which can run on fine-tuned open foundations. The gating factors NVIDIA identifies are operational, not just model quality: anonymized data pipelines, synthetic dataset generation, agent orchestration, secure runtimes and governance. Teams evaluating Nemotron 3 LTM should treat it as an open baseline requiring their own network and customer data, and should ask what governance evidence the platform can produce, since the source offers vendor-stated capability rather than measured production outcomes.

NVIDIA Implementation Risks

The source does not disclose deployment costs, licensing terms, support commitments, model accuracy figures, evaluation methodology or production timelines for Nemotron 3 LTM, the fine-tuning recipe or the NVIDIA AI Enterprise and Agent Toolkit platform layers. Operators therefore cannot, from this source alone, size total cost of ownership or compare open and closed model economics on a like-for-like basis. The 89% figure reflects respondent sentiment in NVIDIA's own State of AI in Telecommunications report rather than measured adoption or deployment success. Claims about frontier-level reasoning performance and accuracy gains for telecom-specific tasks are attributed to NVIDIA and its collaborators; the source provides no independent verification, no named benchmark results for the LTM, and no operator-reported outcomes from production deployments. The Artificial Analysis Intelligence Index reference is cited without methodology detail. Data governance obligations around the anonymization and synthetic dataset steps are stated as requirements but not specified.

Editorial independence disclosure: this article is based solely on the supplied NVIDIA Blog source and does not include independent reporting or verification. Source note: NVIDIA Blog, October 6, 2026.

Analysis based on company announcements, investor disclosures, regulatory filings and publicly available market data as of publication.