Strategic PartnersIssue 387 records
NVIDIA
NVIDIA made its Run:ai GPU orchestration platform open source on Oct 13. That matters to universities and state programs that share GPUs across labs, but the free edition comes without NVIDIA support, and its own documentation says Slurm remains the alternative for traditional HPC. Separately, NVIDIA's France roundup credited its Nemotron models for a French cooperative-purchasing deal covering local governments. The buyers' own announcements don't mention NVIDIA, describe a model-agnostic AI layer, and give no 2,500-organization count.
- Evidence records
- 7
- Cross-source patterns
- 2
- Evidence classes
- 5 vendor claim2 independent reporting
- Outcomes
- 6 emerging1 mixed
- Source freshness
- 5 new this fortnight1 undated1 older, newly relevant
- Research completed
- 2026-10-14
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Open-source Run:ai moves the cost from licenses to support and staffing
NVIDIA's blog says the community edition needs no subscription but includes no enterprise support and carries NVIDIA license terms on the NGC download. The repository confirms Apache 2.0 code, NGC-hosted images by default, no published roadmap, and Slurm as the alternative for bare-metal HPC. An independent analysis concludes production teams still pay for support. Interpretation: for campuses and state programs, the decision is about staffing and support, not license price.
Operating questionDoes the institution have the platform staff to run an unsupported GPU scheduler, or should it budget for partner or NVIDIA support?
Supporting evidenceNVIDIA makes the full Run:ai GPU orchestration platform open source as part of DSX AI FactoryNVIDIA (DSX AI Factory GitHub organization)alfw.ai
NVIDIA's attribution of a French public framework to Nemotron goes beyond what the buyers' documents say
NVIDIA says Twake.ai is built on Nemotron and serves more than 2,500 French public organizations. The LINAGORA–Docaposte release and Acteurs Publics report never mention NVIDIA, describe a model-agnostic AI layer using open-weight models, and give no count. LINAGORA's own July post shows the real link: its Luciole models adapt the Nemotron recipe. Interpretation: vendor ecosystem roundups can turn indirect lineage into procurement claims; verify with buyers.
Operating questionBefore citing an international public-sector deal as a reference, has the claim been checked against the buyer's own documents?
Supporting evidenceNVIDIA France roundup: Twake.ai on Nemotron for French public bodies, Côte-d'Or digital twin, Bull Vera Rubin Ultra productionLINAGORA and DocaposteActeurs PublicsLINAGORA / OpenLLM France
Evidence ledger
Full record · every source keeps its link and limitations
NVIDIA makes the full Run:ai GPU orchestration platform open source as part of DSX AI Factory
On Oct 13 NVIDIA released the complete Run:ai platform as open source, plus four separately adoptable components: KAI Scheduler (Apache 2.0, CNCF Sandbox), GPU Fractioning Operator (CUDA 13.4-native fractions; the proprietary implementation retired), KAI Resource Management (projects, departments, quotas, node pools) and Workload Map (a Go library). Eight DSX partners are integrating it: Crusoe, Mirantis, Nebius, OpenNebula Systems, Rafay, Saturn Cloud, Spectro Cloud and vCluster. The NGC Community Product requires accepting NVIDIA license terms and excludes enterprise support; paid editions continue through Mission Control and AI Enterprise.
Why it matters, evidence and limitations
- Why it matters
- High for research universities and state shared AI computing programs that pool GPUs across labs or agencies on Kubernetes.
- Evidence and measured results
- Vendor announcement; Model Streamer benchmark figures cited are vendor and partner measurements; no utilization results for the open release itself.
- Limitations and uncertainty
- Vendor claims; utilization and throughput benefits asserted without data for this release; community edition unsupported.
NVIDIA Run:ai open-source repository: Apache 2.0 license, components, install paths and stated fit limits
The README describes a monorepo with the control plane, cluster services, installer, CLI, MCP server, client packages, scale tests and docs. It is licensed Apache 2.0 with trademarks excluded and nested MIT and CC-BY-4.0 components. It supports Kubernetes or OpenShift, connected or air-gapped, with images and Helm charts pulled from NGC (nvcr.io/nvidia/runai) by default. It says Slurm is an alternative for HPC-style batch on bare metal and that single-team dedicated GPUs may need only native Kubernetes. No long-range roadmap is published.
Why it matters, evidence and limitations
- Why it matters
- Moderate to high for university research computing teams deciding between Kubernetes GPU orchestration and existing Slurm clusters.
- Evidence and measured results
- Primary project documentation; does not include performance data.
- Limitations and uncertainty
- Project self-description; support and end-of-life details sit on linked pages not reviewed line by line.
alfw.ai analysis: Run:ai code is free, support is not
The analysis says the release creates three offers: self-supported open source, partner-integrated or managed versions (eight DSX partners named; Lambda separately added Run:ai to its Kubernetes service), and NVIDIA's supported commercial Run:ai. It notes NVIDIA closed the acquisition in December 2024, when the founders said open-sourcing could extend reach beyond NVIDIA hardware. It recommends judging the release on adoption: KAI commit history, CNCF progress, and uptake by clouds outside NVIDIA's partner list. It advises teams needing production support, or with only a few GPUs, to skip self-supported use.
Why it matters, evidence and limitations
- Why it matters
- Moderate: helps public institutions frame Run:ai as an operating-cost and support decision.
- Evidence and measured results
- Analysis citing NVIDIA's blog, Reuters, CTech, vCluster, Lambda and Mirantis announcements; no independent testing.
- Limitations and uncertainty
- Commentary rather than measurement; acquisition price estimates differ by source ($700 million vs. $800 million).
NVIDIA France roundup: Twake.ai on Nemotron for French public bodies, Côte-d'Or digital twin, Bull Vera Rubin Ultra production
Published Oct 13, the roundup says LINAGORA and Docaposte secured the CANUT framework agreement, making Twake.ai, built on NVIDIA Nemotron, the sole referenced sovereign digital workplace for more than 2,500 French public organizations. It says a ThinkDeep and Akila digital-twin agent in Côte-d'Or manages nearly 5,800 km of roads and buildings, cutting total operating costs 10% and targeting 30%. Other items: Bull extending European production to Vera Rubin Ultra NVL72 (planned 2028), NVIDIA investments in Doctolib, Entalpic, Genesis AI and Pigment, and a Fondation Inria working group reporting in spring 2027.
Why it matters, evidence and limitations
- Why it matters
- Low to moderate: international local-government and public-procurement signals with limited transferability to U.S. SLED.
- Evidence and measured results
- Vendor roundup without methods, baselines or links to buyer documents for the public-sector claims.
- Limitations and uncertainty
- Vendor claims; the 2,500 figure and the Nemotron attribution do not appear in the CANUT press release or Acteurs Publics report; the Côte-d'Or figure is unsourced and an independent article could not be opened.
LINAGORA and Docaposte win CANUT's sovereign collaborative environment framework (press release)
CANUT, a nonprofit cooperative purchasing body for local governments and other eligible entities, selected the LINAGORA–Docaposte group and Twake.ai after more than ten months of competitive dialogue comparing sovereign proprietary and open-source solutions. The suite covers mail, chat, video, storage and office tools and integrates State suite components. Its AI relies on open-source, open-weight models with compute hosted in France and is model-agnostic. The framework also targets social, medical-social and educational institutions. Région Occitanie is the first large reference. The release does not mention NVIDIA or a count of eligible organizations.
Why it matters, evidence and limitations
- Why it matters
- Moderate as an international model of cooperative purchasing for local and regional government collaboration and AI tools.
- Evidence and measured results
- Vendor press release describing the procurement; no adoption or outcome data.
- Limitations and uncertainty
- Vendor-authored; framework terms and pricing not published; why now: NVIDIA's Oct 13 roundup attributed this framework to Nemotron.
Acteurs Publics: CANUT adopts a sovereign collaborative suite to move local governments away from Microsoft
Acteurs Publics reports CANUT inaugurated the framework on Oct 7 as an alternative to Microsoft and Google tools. Twake Workplace covers the Office 365 scope and integrates State suite components. It is hosted at OVH and can run on-premises or in SecNumCloud at Outscale. AI features query user data for reply suggestions and scheduling. Bidders worked with Région Occitanie and Seine-Maritime. The framework lasts four years, and CANUT's director says culture change may take two to five years. The article does not mention NVIDIA, Nemotron, or a count of eligible organizations.
Why it matters, evidence and limitations
- Why it matters
- Moderate as an international reference for cooperative purchasing and change-management timelines in local government.
- Evidence and measured results
- Trade-press reporting with named quotes from CANUT and LINAGORA leadership.
- Limitations and uncertainty
- Reporting relies on supplier and buyer quotes; no adoption numbers.
LINAGORA: Luciole open models adapt the Nemotron recipe, trained on Jean Zay
The post says LINAGORA and OpenLLM France build open multilingual models with GENCI and AI Factory France support on the Jean Zay supercomputer. Luciole comes in 1B, 8B (hybrid Mamba-Transformer) and 23B versions, all Apache 2.0, with checkpoints, training data and code published and a corpus of around 5 trillion tokens. All Luciole architectures are adapted from Nemotron, and model cards credit a custom adaptation of the Nemotron 3 Nano recipe and training with NeMo on Jean Zay H100s. Twake.ai is listed in the same open-source ecosystem.
Why it matters, evidence and limitations
- Why it matters
- Moderate for state AI offices and universities considering open models built on public compute; explains the Nemotron link behind NVIDIA's Twake.ai claim.
- Evidence and measured results
- Vendor-affiliated description; no benchmarks.
- Limitations and uncertainty
- Why now: it is the documented basis for NVIDIA's Oct 13 'built on Nemotron' attribution; it does not say which models Twake.ai deployments run.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.