Nvidia has released an open-source large telco model aimed at telecommunications operators that want AI trained specifically on their networks, terminology and processes. Unveiled at Mobile World Congress in Barcelona in 2026, the roughly 30-billion-parameter model is part of Nvidia’s Nemotron family of foundation models and represents an attempt to move the company from a behind-the-scenes hardware supplier to a direct provider of telecom-specific AI.
What a “large telco model” does
Where a general-purpose model is trained on broad internet text, a large telco model (LTM) is tuned to the language and workflows of running a network. Nvidia says the model reasons through tasks such as fault isolation — pinpointing which part of a system caused a failure — along with remediation planning and change validation, the process of approving alterations to a live network. The company frames this as a shift toward “reasoning” systems that work through multi-step problems rather than simply matching patterns, which matters in operations where a wrong call has real consequences. One reported benefit is sharper incident summaries: Nvidia says the model can roughly triple the accuracy of automated incident summaries (from around 20 percent to about 60 percent), cutting the volume of alerts that human engineers must review.
Agentic blueprints and energy
Alongside the model, Nvidia introduced an Intent-Driven RAN Energy Efficiency Blueprint — a closed-loop, agentic workflow designed to optimise the energy use of the radio access network — together with additional blueprints for tasks such as network configuration. The pitch is that operators could eventually combine reasoning models with agents that take actions on the network, with humans and AI sharing oversight rather than handing control entirely to software.
Distribution through the GSMA
Notably, Nvidia released the model and an implementation guide through the GSMA’s new Open Telco AI initiative, an industry effort to develop telecom-grade AI. Routing the release through the GSMA is significant because it puts the tooling within reach of mid-tier carriers that lack dedicated AI teams, not just the largest operators. Early adopters named in connection with Nvidia’s agentic network blueprints include Cassava Technologies, NTT DATA and Telenor Group, spanning deployments in Africa, Japan and maritime 5G settings. This sits within the same broader move toward enterprise reasoning agents seen elsewhere, such as the custom enterprise knowledge agents being built on other platforms.
Obstacles and what to watch
The ambition comes with real caveats. Industry analysts have noted that Nvidia is comparatively new to selling directly into the telecom domain, where established suppliers such as Ericsson and Nokia have deep, long-standing customer relationships. AI is also not a complete answer to the sector’s problems: bringing reasoning models to network operations can reduce complexity, but many of the harder, more nuanced challenges will not be solved by one vendor’s model alone. And although the model is open source, a practical question remains over whether telecom IT organisations can absorb and operationalise it quickly enough for it to matter. Open availability lowers the barrier to entry, but integration, data quality and staff skills still determine whether such tools deliver in production. The full announcement is available on the Nvidia blog, with context on the wider industry effort from the GSMA.
