AI news videos

NVIDIA Says Telecom Operators Are Building Their AI Strategy On Open Models

NVIDIA published a post on its own blog arguing that telecom operators are increasingly building their AI strategies on open models, for trust, control and customization rather than cost alone. The post is dated October 6, 2026.

Watch on YouTube

Transcript

Sources

Every claim in this video comes from the top ranking coverage of this topic. The claims and where each one came from:

• Backing sentences, quoted from the page: "Telecom operators are increasingly building their AI strategies on open models, and the reasons go beyond mere cost." and the standfirst "Telcos are turning to open models to build trusted, production-ready AI workflows they can deploy, customize and govern across their most critical workloads." NOTE ON QUOTING: the source sets off clauses with em dashes, and a gate bars every dash form from every field of this file, so source em dashes are rendered here as commas and the source's curly apostrophes are rendered as straight ASCII apostrophes. No word of the quoted text is otherwise altered. EDITORIAL DECISION: hook sentence one states the plainest fact of the post, that NVIDIA says telecom operators are building AI strategies on open models, and attributes it out loud with "NVIDIA says" because the claim is NVIDIA's own reading of its own market. Hook sentence two, "The post is on NVIDIA's own blog", puts the provenance inside the first five seconds, which is required here because nothing on the page is checked by an outside party and no corroboration of any kind was obtainable. The hook is exactly two sentences, because the renderer splits the hook into two on-screen beats and a one sentence hook renders the second beat as undefined and forces a full voiceover and timing redo. The hook is eighteen words, under the twenty word ceiling, and it speaks no figure at all, which keeps the script's single spoken figure out of the opening and off the hook card. The hook says nothing about when anything was launched: it is a claim about where the post appeared, not about any product. NOT NARRATED: the author's name, the categories and the seven tags are real page facts held out of narration for word budget; the byline and the full tag list are carried on the scene 4 card, because authorship is provenance. REFUSED: the source clause "the reasons go beyond mere cost" is a figure of speech and is not narrated or paraphrased into a new one, and the related framing "NVIDIA's latest State of AI in Telecommunications report reflects this shift" is refused on the same ground. The standfirst's "deploy, customize and govern" is a three item list and is barred from narration, so it sits on the scene 1 card. No superlative is narrated: the page evidences no ranking, no market share and no leadership claim about open models, and the script asserts none.(HOOK BACKING. First-party NVIDIA corporate blog post on blogs.nvidia.com, headlined "Why Telecom Operators Are Building Their AI Strategy on Open Models", dated October 6, 2026 on the page itself, by Kanika Atri, filed under categories AI and Software with tags for agent style AI, the Nemotron model family, NVIDIA AI Enterprise, NVIDIA NeMo, Open Source, Telecommunications and Trustworthy AI. Captured by our news watcher at 2026-10-06T13:06:22.393Z and read for this script only from that saved snapshot; the manager independently re-fetched the live URL and reports the live page still matches the snapshot headline, byline, date and body. This is NVIDIA writing on NVIDIA's own blog, by an NVIDIA author, recommending NVIDIA's own models and NVIDIA's own platform, so the page is the sole and self-interested source for every narrated claim in this Short.)
• Backing sentence, quoted from the page: "NVIDIA's latest State of AI in Telecommunications report reflects this shift, with 89% of respondents reporting that open source models and software are important to their company's AI strategy." EDITORIAL DECISION: this scene carries the script's ONE spoken figure, "eighty nine percent", spoken as words because a gate bars digits from every narration field. It is the page's headline number and the only figure on the page that is not a model specification, a version string or a parameter count, all of which our voice engine garbles. It is placed in scene 1 so that the hook before it and scenes 2, 3, 4, 5 and the endcard after it are completely figure free, which satisfies the rule that two spoken figures must be separated by at least two figure free sentences; in this script there is no second spoken figure at all. That choice is deliberate and not merely conservative: our transcriber falls into a repetition loop when two figures are spoken close together and swallows the middle of the transcript, which costs a full render. The 30 billion parameter count of the telecom model was the obvious second candidate figure and was deliberately NOT spoken for exactly that reason; it is on the scene 3 card instead. The figure is attributed out loud in the same sentence with "NVIDIA's own survey report says", so a listener hears the provenance and the number together rather than hearing the number as a neutral industry statistic. NOT NARRATED: the report's title, State of AI in Telecommunications, and the page's exact wording "open source models and software" are both on the silent card; narration compresses to "open models" for budget, which is the page's own term throughout the rest of the post. The standfirst list "deploy, customize and govern" and the phrase "production-ready AI workflows" are on the card. REFUSED: the clause "reflects this shift" is a figure of speech and is not narrated in any form. No criticism is invented: the script does not call the figure false, inflated, misleading or unrepresentative, and does not attribute motive to NVIDIA for publishing it. It states only what the page states, plus the plain fact recorded in scene 4 that no sample size is given, which is an observation about the page rather than a charge against the number. The word "marketing" does not appear in any narrated line of this file.(SCENE 1 BACKING. Same first-party NVIDIA blog post, the survey paragraph. The 89% figure is NVIDIA's own survey result from NVIDIA's own report, and the page gives no sample size, no methodology, no date range and no definition of who the respondents were.)
• Backing sentences, quoted from the page: "Open models give telcos the ability to trust, control and customize AI across their most critical workloads, from autonomous networks to customer care." and "For operators, the strategic value of open models is fivefold:" followed by the five items, quoted on the card and in full here: "They expand access to frontier-level intelligence at lower cost, allowing operators to reserve closed models for the workloads where they drive the most value. Independent benchmarks such as the Artificial Analysis Intelligence Index v4.3.2 show that leading open models are becoming more competitive across demanding reasoning, coding, scientific and agentic workloads.", "They support telco-specific customization, with open weights and training recipes that operators can fine-tune for their own operations using network, customer and industry data.", "They enable trustworthy AI by giving telcos greater visibility into and control over model artifacts and behavior, so models can be evaluated, adapted and governed in alignment with regulations and business policies.", "They enable flexible, secure deployment: teams can size and optimize open models to run across public clouds, private infrastructure and edge environments." and "They unlock the opportunity for telcos to deliver locally adapted AI services to enterprise and government customers by hosting and fine-tuning open models." EDITORIAL DECISION: this is the single biggest timing risk in this story, because the source is an explicit listicle. A multi-item comma list reads at about 2.65 spoken words per second against about 2.18 for plain prose and will blow the sixty second audio ceiling even at a legal word count. So no narrated sentence in this script enumerates the five strategic values, and no narrated sentence anywhere contains a comma separated list of three or more items. Scene 2 narration is one twelve word sentence carrying exactly two items, "control and customization", joined by "and" with NO comma between them, and the entire fivefold list including all five numbered items and every sub list inside them is carried on the silent card, where it costs zero seconds. The page's own trio "trust, control and customize" was cut to two items for the same reason, and "trust" was the item dropped because the trustworthiness point is the hardest to state without implying an outside verification that the page does not provide. The claim is attributed with "NVIDIA says" because it is NVIDIA's assertion about its customers' interests. NOT NARRATED: frontier level intelligence at lower cost, the reservation of closed models for the highest value workloads, open weights and training recipes, visibility into model artifacts and behavior, governance in alignment with regulations and business policies, deployment across public clouds and private infrastructure and edge environments, locally adapted AI services for enterprise and government customers, and the autonomous networks and customer care workloads are all real page facts dropped from narration for word budget and for the no list rule, and all are on the card. The cited benchmark name and version are on the card only, never spoken: it is a dotted version string and our voice engine garbles those. REFUSED: the word meaning agent like is barred from every narration field in this file because our voice engine garbles it in all four prompt configurations and no vocabulary edit fixes it; the page uses it repeatedly, in "agentic workloads" and "agentic workflows" and in its own tag list, and it appears in this file only on silent cards. The figures of speech "at every layer", "a clear path", "building blocks" and "goes beyond mere cost" are refused in every form. The vendor aspiration verb "unlock the opportunity" is not narrated. The script does not say open models are better, cheaper or more capable than closed models: the page's own claim is only that leading open models are "becoming more competitive", and even that is left on the card rather than narrated, because the only evidence offered for it is a benchmark we did not read.(SCENE 2 BACKING. Same first-party NVIDIA blog post, the fivefold strategic value list and the opening paragraph. These are NVIDIA's own assertions about why operators should choose open models. The one benchmark the page offers as independent evidence, the Artificial Analysis Intelligence Index v4.3.2, was not fetched or verified by us.)
• Backing sentences, quoted from the page: "NVIDIA announced the 30-billion-parameter Nemotron 3 Large Telco Model (LTM), fine-tuned by AdaptKey on open source telecom datasets to deliver accuracy gains for telecom-specific tasks. This gives operators an open baseline that can understand telecom industry terminology and reason effectively through telecom operations workflows such as network configuration and customer incident triage." and "To help operators customize the Nemotron 3 LTM and other open models with their own operational data, NVIDIA released the full recipe that walks through the end-to-end fine-tuning pipeline for adapting open models to operator-specific networks, customers and procedures using NVIDIA NeMo open libraries." and "The NVIDIA Nemotron family of open models provides frontier-level reasoning performance optimized for agentic workflows, as well as speech capabilities for voice applications, with open weights, training data and recipes." EDITORIAL DECISION: the narration says NVIDIA "announced" the model and "released" the recipe, which is exactly what the page says, and it attaches NO time claim of any kind to either. It does not say announced today, just announced, just launched, released today or new, because the announcement itself lives behind a hyperlink to a page we never fetched, and the only thing verified as same day here is this blog post. This is a strategy and positioning post, not a launch post, and the script is worded so that a listener cannot come away believing a product shipped today. No brand name is spoken: "Nemotron", "Nemotron 3", "Nemotron 3 Large Telco Model", "LTM", "AdaptKey", "NeMo" and "NVIDIA NeMo" are glued brand plus digit tokens, short glued letter tokens or invented names, all of which our voice engine garbles, so narration refers to "a telecom specific open model" and "a recipe" and every real name is carried silently on the card. The 30 billion parameter count is NOT spoken: it would be the script's second spoken figure, and two figures spoken close together send our transcriber into a repetition loop that swallows the middle of the transcript and costs a full render, so the count is on the card as digits where nothing is spoken. "telecom specific" is written without a hyphen, unlike the source's "telecom-specific", because hyphenated compounds are a known garble risk for our voice engine. "fine-tune" is spoken once and is a known safe token. The sentence carries two clauses joined by "and", not a list of three. NOT NARRATED: the parameter count, the partner that performed the fine-tuning, the open source telecom datasets, the claimed accuracy gains, the open baseline framing, telecom industry terminology, network configuration, customer incident triage, the end to end pipeline description, the named NVIDIA open libraries, and the Nemotron family's open weights and training data and speech capabilities for voice applications are all real page facts dropped from narration for word budget, for the no list rule and for the spoken brand name rules, and all are on the card. REFUSED: "frontier-level reasoning performance" and "reason effectively" are vendor performance claims with no benchmark shown on the page and are not narrated. "turn open models into practical building blocks" and "walks through" are figures of speech and are refused in every form. The script does not say the model is accurate, best in class or production proven: the page offers no benchmark table, no sample, no error rate and no independent evaluation for it. No approval, availability date or price is claimed, because the page states none.(SCENE 3 BACKING. Same first-party NVIDIA blog post, the section headed "Open Foundations, Tuned for Telecom Operations". CRITICAL PROVENANCE POINT: the word "announced" in the model sentence is a hyperlink on the page pointing at a SEPARATE announcement page that we did NOT fetch, so this script attaches no date whatsoever to the model or to the recipe, and says nothing about either being new.)
• Backing facts, confirmed by reading the whole document as captured at 2026-10-06T13:06:22.393Z. Publisher and authorship are stated by the page itself: it is published on blogs.nvidia.com and the byline reads "by Kanika Atri", filed under categories AI and Software. Survey ownership is stated by the page: "NVIDIA's latest State of AI in Telecommunications report reflects this shift, with 89% of respondents reporting that open source models and software are important to their company's AI strategy." The page gives no sample size, no respondent definition, no field dates, no geography breakdown, no margin of error and no methodology note for that figure, and links no methodology document that we fetched. Commercial self-interest is stated by the page throughout: "The NVIDIA Nemotron family of open models provides frontier-level reasoning performance", "NVIDIA announced the 30-billion-parameter Nemotron 3 Large Telco Model (LTM)", "NVIDIA released the full recipe ... using NVIDIA NeMo open libraries", "NVIDIA provides that end-to-end platform, powered by NVIDIA AI Enterprise software and NVIDIA Agent Toolkit, and spanning data pipelines, open models, agent orchestration, secure runtimes and simulation", and the closing call to action "Learn more about NVIDIA technologies for telecommunications." EDITORIAL DECISION: scene 4 is spent entirely on who is making these statements, because that is the central editorial fact of a chip and software vendor publishing a strategy post that concludes operators should adopt that vendor's own models and platform. It is two sentences of twenty one words, each a single plain provenance fact with no comma list: that NVIDIA wrote this on its own blog and that the figure is from its own survey report, and that no sample size is given. This satisfies the requirement that the provenance disclosure be narrated out loud rather than carried only on a card. The scene speaks no figure and no date. "No sample size is given" is a literal, checkable statement about the page, not an accusation: the page genuinely prints no sample size. No criticism is invented. The 89% is not called false, inflated or misleading; no motive is attributed to NVIDIA or to any operator; the word "marketing" appears in no narrated line; and no narrated line says or implies that NVIDIA is hiding anything. NOT NARRATED: the author's name, the categories, the seven tags, the report's title, the named NVIDIA platform components and the closing call to action are all real page facts carried silently on the card for word budget. The missing methodology, date range and respondent definition are on the card; narration names only the missing sample size, because that is the single most load bearing omission and the budget allowed one. REFUSED: no countdown, deadline, urgency or expiry phrasing appears in any field of this file, because a downstream validator hard fails those phrases; the GTC Berlin event registration furniture, including its "Register Now" imperative and its date range, was excluded for that reason as well as because it is advertising aimed at NVIDIA's readers. No second person hype and no rhetorical question appears in any copy field. No em dash, en dash or double hyphen appears in any field, and the source's own em dashes were rewritten as commas wherever source text is quoted in this file. The page's framing "Together with a strong partner ecosystem building at every layer of this platform, telecom operators have a clear path to translate the benefits of open models into production-ready AI workflows" is vendor aspiration and contains two figures of speech, and it is refused in every form.(SCENE 4 BACKING, MANDATORY PROVENANCE DISCLOSURE, part one. The page itself is the evidence. A first-party post on NVIDIA's own corporate blog, by an NVIDIA author, recommending NVIDIA's own open model family and NVIDIA's own commercial platform, in which the headline statistic comes from a survey report NVIDIA publishes itself with no sample size, no methodology, no date range and no respondent definition stated anywhere on the page.)
• Backing quotations, quoted from the page: "Open models allow SoftBank Corp. to build on the rapid progress of global foundation models while applying the network knowledge and operational expertise we have accumulated over many years," said Rajeev Koodli, principal fellow of SoftBank Corp. and senior vice president of SB Telecom America. "We are using open foundations extensively, including NVIDIA Nemotron models and others, in developing our SoftBank Large Telecom Model, continuously advancing it for telecom-specific use cases such as network operations, design and overall management." And: "At AT&T, we believe the future of AI is not about choosing a single model; it's about intelligently matching every workload to the right combination of performance, cost and control," said Andy Markus, chief data and AI officer of AT&T. "Open models are essential to that approach, and with NVIDIA, we can bring that model-choice strategy into production with the scale, reliability and governance our business requires." And: "For countries like Indonesia, the value of open models goes beyond access to powerful AI. It is about adapting that intelligence to our own language, culture, data and real-world needs," said Chirag Sukhadia, chief data and AI officer of Indosat Ooredoo Hutchison. "Sahabat-AI puts this into practice, using open models as a foundation to build AI that understands Indonesia and can be developed for local applications. This allows us not only to adopt AI, but to build capabilities around it and enable more Indonesians to create with AI on their own terms." EDITORIAL DECISION: the narration states the one fact that governs how a viewer should weigh all three quotations, that every operator quoted is an NVIDIA partner or customer, and then states what this video actually read. "We read only that post" is a literal statement about this production: only the captured article text was read, and the linked announcement page behind the word announced, the benchmark index, the closing telecommunications link and every item in the Recent News and Related News lists were left unfetched, so this script makes no claim about any figure, capability or availability that is not in the captured text. No operator name and no executive name is spoken: "SoftBank Corp.", "SB Telecom America", "AT&T", "Indosat Ooredoo Hutchison", "Sahabat-AI", "Rajeev Koodli", "Andy Markus" and "Chirag Sukhadia" are short glued letter tokens, hyphenated compounds or names our voice engine garbles, and all are carried silently on this card. REGIONAL FRAMING, recorded deliberately: our channel is United States focused, and this story qualifies because the publisher is a United States company and the page quotes a large United States carrier. Two of the three operators quoted are foreign, a Japanese carrier and an Indonesian carrier, and the Indonesian local language model is a real and substantial part of the page, but none of it is narrated: narration stays on NVIDIA, the telecom model, the published fine-tuning recipe and the survey figure, and the foreign operators appear on this silent card only. This Short is not about Indonesia or Japan. NOT NARRATED: all three quotations in full, every executive title, the Japanese carrier's own large telecom model, the United States carrier's model choice strategy, the Indonesian open source model family, and the page's paragraph about data anonymization and privacy preserving synthetic datasets for fine-tuning are real page facts dropped from narration for word budget, for the no list rule and for the spoken name rules. REFUSED: "the future of AI is not about choosing a single model", "on their own terms", "puts this into practice", "build on the rapid progress" and "goes beyond access to powerful AI" are figures of speech or vendor aspiration and none is narrated or paraphrased into a new figure of speech. "Powerful AI" and "frontier-level" are unevidenced superlatives on this page and are not narrated. The script does not say any operator has deployed anything in production, achieved any saving or measured any result, because the page reports no deployment outcome, no metric and no case study result for any of the three.(SCENE 5 BACKING, MANDATORY PROVENANCE DISCLOSURE, part two. Every operator quoted on the page is an NVIDIA partner or customer: a Japanese carrier building on NVIDIA's own model family, a large United States carrier describing a production strategy it runs with NVIDIA, and an Indonesian carrier building on open models. No independent audit, no regulator, no journalist and no outside reviewer is named anywhere on the page, and only this one page was read for this script.)
• No source sentence backs the endcard. It is the standard Inboxsmith pitch, "Inboxsmith builds AI receptionists for small business. Please like and subscribe for more news." EDITORIAL DECISION: the endcard carries no NVIDIA claim, no operator name, no model name, no availability statement and no figure of any kind, so no vendor assertion is spoken at the point where narration stops attributing. Keeping it figure free also keeps the script's single spoken figure, in scene 1, eight sentences clear of the close, so no two figures can land near each other at the end of the read. The two required sentences are present verbatim, including both full stops, and "Please like and subscribe for more news." is the final sentence of the whole script. endcard.onscreen is exactly "inboxsmith.com". No countdown, deadline or urgency phrasing, no rhetorical question and no second person hype appears here. The source's own closing call to action, "Learn more about NVIDIA technologies for telecommunications.", is page furniture aimed at NVIDIA's readers and is deliberately not carried into our endcard or anywhere else in this file.(ENDCARD. Inboxsmith house sign off. Carries no claim from the source and is deliberately not traceable to the article.)
• No corroboration of any kind was obtainable, so this single NVIDIA blog post is the sole source for every narrated claim. Recorded plainly, from the files themselves: the YouTube transcript pull returned 0 of 5 videos, because the yt-dlp build installed on this box is too old to fetch captions. That is a known local tooling limitation, not bot blocking, and it says nothing about the story. transcripts.json for this slug holds an empty videos array, so not one sentence of transcript text was available to check anything against, and sources.json for this slug holds an empty sources array, which is normal for this pipeline and is not a red flag. Because any claim absent from the article would require two or more independent transcript confirmations and there are zero transcripts, every narrated claim in this script comes from the captured article and nothing else, and nothing was inferred, estimated, averaged or filled in. The eighteen dated probes in timeline-probes.json, fetchedAt 2026-10-06T13:10:15.788Z, were read in full for weak existence context only and not one word of them is narrated. They are unrelated and mostly old noise: a 2019 NVIDIA Shield TV remote battery tutorial, a 2022 AI simulation explainer, a 2023 clip about using a chat assistant extension on online exams, a 2023 AI development clip, a 2025 data centre explainer, a 2025 NVIDIA chief executive career clip, a 2025 Jira tutorial, a 2026-05 clip on China opening its market to United States AI chips, a 2026-05 clip on cancelled AI data centres, a 2026-06 AI scaling clip, a 2026-07 AI governance clip, a 2026-08 AI daily brief, a 2026-09 clip about AWS and NVIDIA adding GPUs, a 2026-09 government announcement, a 2026-10 quantum computing explainer, a 2026-10 autonomy podcast, a 2026-10 clip about export bans and a model download, and a 2026-10-06 clip about AI and geopolitics. NOT ONE of the eighteen probes is this story: none mentions telecom operators adopting open models, none mentions the Large Telco Model or the telecom fine-tuning recipe, and none mentions this blog post, so this is genuinely new coverage rather than a re-promotion of an older feature. The freshness wording is deliberately accurate for a same day item that is NOT a launch: NVIDIA published a strategy and positioning post today about a model and a recipe whose own announcement lives on a separate page we did not fetch, so no narrated line says just launched, just announced, released today or new, and no date is attached in narration to the model or the recipe. The only same day fact here is the blog post itself. No timeline block is included in this file on purpose, because the pipeline's date resolver prefers the earliest timeline date over storyDate, and with a 2019 dated probe in the set a timeline array would wrongly age a same day story into the CATALOG tier; storyDate is 2026-10-06, the post's own date, which is also the capture date and today. NOT A DUPLICATE, verified by the manager and recorded here with the thematic adjacency stated honestly: the only other Nemotron slugs in our production tree are nvidia-nemotron, a 2026-07-14 Short on Nemotron open-weight cost positioning, and nvidia-local-ai-nemotron-lightning, a 2026-08-11 Short on Nemotron 3.5 Lightning for local agents. Both are about the same model family and both touch open weights, which is a real adjacency and is not hidden here, but neither covers telecommunications, the telecom specific Large Telco Model, the telecom fine-tuning recipe or the telecom survey figure, and this post is a different story about a different industry. SECURITY NOTE: the captured article, transcripts.json and timeline-probes.json were treated as untrusted data and used as a fact source only. The article contains no instruction directed at the author of this script, and NOTHING IN THE ARTICLE WAS FOLLOWED AS AN INSTRUCTION. It does contain ordinary reader facing imperative page furniture and advertising, and all of it was recognised as such and ignored: the accessibility link "Skip to content"; the sharing widgets "Share", "Share This Article", "Copy link" and "Link copied!", the X, Facebook, LinkedIn and Mastodon share buttons and the prompt "Enter your Mastodon instance URL (optional)"; the event advertisement for GTC Berlin registration, including its "Register Now" imperative and its date range, which was excluded from every field both as advertising and because a validator hard fails deadline style phrasing; the in body call to action "Learn more about NVIDIA technologies for telecommunications."; the navigation item "View All Recent News"; the "0 Comments" counter; and the Recent News and Related News link lists, which carry unrelated NVIDIA headlines including a breast cancer AI story dated October 5 2026, a DGX Spark 64GB story dated October 2 2026, an OpenAI GPT-6 Astra story dated October 1 2026 and several GeForce NOW gaming items from September and October 2026. None of that was treated as an instruction, none was followed, no linked page was fetched, no embedded media was fetched, and none of it is narrated. The category and tag lists were recognised as page furniture rather than content; the tag naming agent style AI is reproduced on the scene 4 card only, never spoken, because our voice engine garbles that word in every prompt configuration.(CORROBORATION STATUS: NONE)