Telecom Carriers Shift AI Plans Toward Open Models, NVIDIA Survey Finds

Telecom operators are increasingly routing their AI strategies through open models rather than proprietary APIs, according to a new NVIDIA report on AI adoption in the telecommunications sector. The survey points to a shift in how carriers weigh model choice, with control and customization ranking alongside cost as deciding factors. Operators are applying the technology to some of their most sensitive workloads, including autonomous network management and customer care.

The reasoning goes beyond licensing fees. Closed models route data and inference through a vendor's infrastructure, which limits how far an operator can tune a model to its own traffic patterns, equipment and internal processes. Open-weight models can be run inside a carrier's own data centers, letting engineering teams inspect behavior, fine-tune on proprietary network data and keep subscriber information on-premises. For carriers bound by national data-residency rules, that distinction can settle the build-versus-buy question before price enters the conversation.

The workloads in question raise the stakes. Autonomous networks depend on models that make or recommend configuration changes in near real time, where an unexplained output can disrupt service for thousands of subscribers. Customer-care deployments handle billing records, account details and call transcripts, all of which fall under privacy and telecom regulations that vary by country. An operator that cannot audit a model's weights or retrain it on local data has limited ability to answer to a regulator or explain a failure.

NVIDIA has a commercial interest in the trend it is documenting. Open-weight releases such as the Llama family, Mistral's models and NVIDIA's own Nemotron line run on its GPUs, and the report arrives as the company courts telecom operators with hardware and software for network automation. The framing competes with a different pitch from cloud providers, who argue that managed closed models cut operational burden for carriers that lack large in-house AI teams. Which argument wins appears to depend on how much AI engineering a given operator can staff.

The shift matters beyond telecom. Carriers run some of the largest private infrastructure networks in the world, and their procurement choices influence which model formats get long-term support from vendors. If operators standardize on open weights for regulated workloads, that sets a reference point for banks, hospitals and government agencies weighing similar trade-offs between control and convenience.

What to watch next is whether the survey's stated preference translates into deployed systems. Operator interest in open models has run ahead of production rollouts in past cycles, and the deciding factor is likely to be talent: carriers that can fine-tune and evaluate models in-house will move faster than those that cannot. The next round of earnings calls and vendor disclosures should show whether the shift is real or still aspirational.