What the post establishes
NVIDIA's October 6 post says 89 percent of respondents to its own State of AI in Telecommunications report consider open source models and software important to their company's AI strategy. Around that figure it arranges a telecom model, a fine-tuning recipe, a software platform and quotes attributed to executives at SoftBank, AT&T and Indosat Ooredoo Hutchison. That establishes NVIDIA's account of its offer and three operators' stated positions. It does not establish how many operators run these tools in production or what they got for it.
Source: Why Telecom Operators Are Building Their AI Strategy on Open Models · 2026-10-06 · Primary source context · attributed account
The concrete part: a model and a recipe
The checkable commitments are two. NVIDIA says it announced the 30-billion-parameter Nemotron 3 Large Telco Model, fine-tuned by AdaptKey on open source telecom datasets, which NVIDIA says delivers accuracy gains on telecom-specific tasks and gives operators a baseline for workflows such as network configuration and customer incident triage. It also says it released the full recipe for the fine-tuning pipeline, so an operator can adapt open models to its own networks, customers and procedures with NVIDIA's NeMo libraries. Silver Owl view: the recipe is the more interesting of the two, because a documented process can be reproduced and a model announcement can only be admired. The practical question is how much of that pipeline a team can actually run, and what it still has to validate itself.
Source: Why Telecom Operators Are Building Their AI Strategy on Open Models · 2026-10-06 · Primary source context · attributed account
Three operators, three different ambitions
In NVIDIA's post, SoftBank's Rajeev Koodli says the company uses open foundations, Nemotron among them, to develop its SoftBank Large Telecom Model. AT&T's Andy Markus describes matching every workload to the right combination of performance, cost and control. Indosat Ooredoo Hutchison's Chirag Sukhadia describes building AI that understands Indonesia through its Sahabat-AI family of open source models. Our reading: that is development work, a sourcing philosophy and a national adaptation program, each presented inside one vendor's frame. They make good evaluation questions. They do not share a unit of delivery, so none of them should be read as the result of another.
Source: Why Telecom Operators Are Building Their AI Strategy on Open Models · 2026-10-06 · Primary source context · attributed account
The seller's wider offer
NVIDIA presents its AI Enterprise software and Agent Toolkit as an end-to-end platform spanning data pipelines, open models, agent orchestration, secure runtimes and simulation. Our reading: the post is a commercial argument for the platform around the open models as much as for the models. A fair test, and our suggestion: have a team reproduce the tuning recipe on one telecom task, document its own evaluation, and list which NVIDIA software the workflow actually required. That turns a broad pitch into inspectable work, and it tells you whether open means something you can leave with.
Source: Why Telecom Operators Are Building Their AI Strategy on Open Models · 2026-10-06 · Primary source context · attributed account