Nebius Customer Atlas Atlas changes New customers, relationship updates and selected company news. Dates below are when the Atlas published each change.
2026-10-20 Multiply Labs: Trains robotics models on Nebius H100 GPUs
Original source 2026-10-20 KnowledgePilot: Uses Nebius for AI question answering
Original source 2026-10-20 CamoCopy: Nebius hosting for AI chat and personal-data storage
Original source 2026-10-20 Arker: Nebius GPU compute in the EEA
Original source 2026-10-20 Cortecs: Nebius inference in a European LLM gateway
Original source 2026-10-20 Nolla Health: Post-trained clinical models on 32 Nebius B200 GPUs
Original source · 2026-10-15 2026-10-20 Cloop: Nebius inference for website chat and sales messages
Original source 2026-10-20 TypeSafe AI: Customer AI prompts processed on Nebius compute
Original source 2026-10-20 TextCortex: Routes DeepSeek V4 Pro requests through Nebius.
Original source 2026-10-20 NeoDemos: Token Factory embeddings for council document search
Original source 2026-10-20 Weisscam: AI inference for its shared business workspace
Original source 2026-10-20 Norma: Tavily search for firm research during onboarding and refreshes
Original source 2026-10-20 SentientPro: Tavily web search for research and industry news
Original source 2026-10-20 Transcripta Bio: Nebius compute for transcriptomic drug discovery
Original source · 2025-06-25 2026-10-20 Lity Club: Serverless document recognition for a legal copilot
Original source 2026-10-20 Aikium: Nebius GPU cloud resources for drug discovery
Original source · 2025-06-25 2026-10-20 EUrouter: Routes model requests through Nebius Token Factory
Original source 2026-10-20 Robot Company: Robot Company's repository reports 17 successful simulated picks in 18 environment-tuned trials using chained Nebius jobs and 89.4% success for optimized soft foam under sourcing noise. Its residual policy matched the open-loop policy, which remains deployed.
Original source 2026-10-20 Recraft: In an undated testimonial, founder and CEO Anna Veronika Dorogush said Recraft used Nebius for ongoing data processing as well as model training.
Original source 2026-10-19 Elastic: Tavily search for security advisory drafting
Original source · 2026-06-23 2026-10-19 Inherent: Trained its Faraday AI Scientist on Nebius AI Cloud
Original source 2026-10-19 BusinessMCP: Tavily web research for its assistant and sales
Original source 2026-10-19 EstateView: Tavily search for AI-assisted estate planning
Original source 2026-10-19 trefolio: Tavily search for investment research
Original source 2026-10-19 General Input: Uses Tavily for web search.
Original source 2026-10-19 Bryn: Tavily web search for enrichment
Original source 2026-10-19 Ataraxis AI: Deploying Nebius GPU resources for cancer diagnostics
Original source · 2025-06-25 2026-10-19 Together AI: Tavily search for Open Deep Research
Original source · 2025-04-16 2026-10-19 Odin: Tavily web search for industry and market research
Original source 2026-10-19 Candor: Audience research with Tavily web search
Original source · 2026-09-29 2026-10-19 Lioma: Tavily web search for company research
Original source 2026-10-19 botts.ai: Token Factory inference for open-weight models
Original source 2026-10-19 University of Washington: Token Factory for mathematical research pipelines
Original source · 2026-07-22 2026-10-19 Cotal: Hosts agent sandboxes on Nebius
Original source 2026-10-19 LangWatch: Agent simulation tests with Nebius-served models
Original source 2026-10-19 Rasa: Token Factory inference for voice-agent demonstrations
Original source 2026-10-19 SocialCrawl: Tavily web search, extraction, mapping and crawling
Original source 2026-10-19 Robot Company: Robot simulation and training on Nebius serverless GPUs
Original source 2026-10-19 Temple University: Nebius compute supports collaborative mathematics research
Original source 2026-10-19 Rechnungskit: Token Factory for Kit, product suggestions and support drafts
Original source 2026-10-19 everyais: Routes model inference requests through Nebius Token Factory
Original source 2026-10-19 Neuralwatt: Uses Nebius GPU infrastructure and AI compute services
Original source 2026-10-19 Perforated: Nebius compute for an eight-week research pilot
Original source · 2026-10-12 2026-10-19 Anemos Bio: Nebius compute for an eight-week cancer research pilot
Original source · 2026-10-12 2026-10-19 Inceptron: Nebius infrastructure for EU inference workloads
Original source 2026-10-19 Indraprastha Institute of Information Technology, Delhi: Nebius compute for speech AI research
Original source 2026-10-19 CloudThrill: Multimodal inference labs and Kubernetes model serving
Original source · 2026-07-07 2026-10-19 2comply: Nebius hosting for its compliance platform
Original source 2026-10-19 Caltech: Cloud compute for mathematical research
Original source 2026-10-19 Solo Tech: Nebius infrastructure for physical AI
Original source · 2026-05-16 2026-10-19 Anonimizator: Personal-data detection with Nebius Token Factory
Original source 2026-10-19 SkyPilot Tokens: Model serving on Nebius GPUs and Token Factory
Original source 2026-10-19 Accenture: Nebius infrastructure and Token Factory for applied AI
Original source 2026-10-18 Rivet: Dependency upgrades with Token Factory and Tavily
Original source 2026-10-18 RunRobot: Runs robot policy simulations on Nebius GPUs
Original source · 2026-03-15 2026-10-18 Apostle Labs: Nebius inference and RAG embeddings
Original source 2026-10-19 NVIDIA: NVIDIA’s Bryan Catanzaro identifies himself as a Nebius customer and says NVIDIA’s work with Nebius as a GPU infrastructure provider was instrumental to creating its Nemotron open models.
Original source 2026-10-19 NVIDIA: Tavily’s October 6, 2026 article says NVIDIA uses Tavily in Nemotron 3 Ultra post-training.
Original source · 2026-10-06 2026-10-19 NVIDIA: NVIDIA announced an agreement to acquire the AI model platform Hugging Face.
Original source · 2026-09-03 2026-10-19 Decart: The Jerusalem Post, citing Bloomberg, reported that Anthropic abandoned its proposed roughly $6 billion acquisition of Decart after due diligence. The companies could still consider other collaboration, and both declined to comment.
Original source · 2026-09-08 2026-10-19 Mastercard: Nebius’s June 19, 2026 presentation reported that Mastercard achieved a high detection rate and reduced reaction time for money-laundering detection using Tavily’s online signals.
Original source · 2026-06-19 2026-10-19 WEKA: WEKA’s September 24, 2026 benchmark report measured median end-to-end response time of 7.3 seconds with Augmented Memory Grid versus 25.1 seconds with GPU memory alone on Nebius. It also reported 53,651 requests meeting a five-second time-to-first-token threshold versus 9,976.
Original source · 2026-09-24 2026-10-19 GDELT Cloud: GDELT Cloud’s changelog identifies Tavily fallback extraction in its V2 entity-tone scorer. The scorer supports concurrent processing with a per-run Tavily budget, allowing hourly dogfood ingest to cover thousands of queued entity-story cases.
Original source 2026-10-19 LynxKite: Nebius’s historical Lynx Analytics case study reports that a graph-based RAG benchmark increased retrieval accuracy from 5% to 15%. It also describes a GNN trained in LynxKite and powered by Nebius AI Cloud serving as a graph-based retriever.
Original source 2026-10-19 London Institute for Mathematical Sciences: Nebius’s customer story lists Virtual Private Cloud among the services used for LIMS’s BERT rule-abstraction research.
Original source 2026-10-19 Custodio Legal: Custodio Legal’s privacy policy says Nebius replaced z.ai as its text-generation provider. It also describes how document excerpts and matter context support drafting, cited answers, fact extraction and automatic matter reevaluation.
Original source 2026-10-19 Pleias: Pleias’s Nano model card documents additional RAG model training on TractoAI using Nebius’s ISEG cluster. Pierre-Carl Langlais also reports that Common Corpus 2 included 11 million OpenAlex articles, about 175 billion tokens, processed with PDF parsing models on Nebius.
Original source 2026-10-19 TheStage AI: The founder credited Nebius GPU infrastructure for TheStage’s Gemma4 compression release and said evaluations ran on Modal. A Recraft research case study also credited Nebius infrastructure. Nebius’s existing customer story reports uninterrupted GPU operation and says similar optimization work with Praktika’s models tripled GPU cost savings.
Original source 2026-10-19 NVIDIA: Tavily’s September 29, 2026 article says NVIDIA used its live web results to generate research trajectories for fine-tuning Nemotron 3 Super.
Original source · 2026-09-29 2026-10-19 MongoDB: MongoDB’s August 26, 2025 blog describes using Tavily in an editorial application to enrich topic suggestions with related contextual information and source URLs.
Original source · 2025-08-26 2026-10-19 Milestone Systems: Milestone's training account reports evaluation gains of 19.4% in traffic flow and direction understanding, 8.9% in condition identification and 4.4% in alert verification accuracy after post-training Cosmos Reason for smart-city video tasks on Nebius.
Original source 2026-10-19 Caltech: The authors of “The Two-Hump Problem: Bridging the Difficulty Gap in Mathematical Reinforcement Learning” acknowledge cloud computing resources provided by Nebius through Nebius Academy’s Research Program.
Original source 2026-10-19 Antioch: Nebius’s customer story reports that a warehouse-surveying robot customer cut time to results from weeks to hours while improving model accuracy using Antioch on Nebius. It also reports a 25% improvement in Antioch developer efficiency through CLI control of cloud orchestration and parallelization.
Original source 2026-10-19 Solo Tech: Founder Dhruv Diddi describes Solo Tech’s AGIBOT X2 demonstration at Nebius’s first APJ event, with robot reasoning powered by Nemotron on Nebius Token Factory.
Original source 2026-10-19 LMSYS · SGLang: Nebius’s undated customer case reports that SGLang’s DeepSeek R1 experiments maintained nearly flat throughput across 16, 32 and 64 concurrent requests. It also says the optimization iterations took days.
Original source 2026-10-19 Luma: In a September 17, 2024 interview with Cerebral Valley, Nebius said it was the key cloud provider for inference at Dream Machine’s launch. It also described supplying storage, fast networking and reliable infrastructure for Luma’s video generation.
Original source · 2024-09-17 2026-10-19 AIMultiple: AIMultiple reports benchmarking Nebius inference through OpenRouter using GPT-OSS-120B. Its benchmark sends 108 questions every five minutes and uses a reference question to measure first-token and end-to-end latency.
Original source 2026-10-19 Lightning AI: Lightning AI announced emergency maintenance on October 16 and subsequently identified a security incident as the reason its service was down. Its official status page continued to list the incident as unresolved on October 19.
Original source · 2026-10-16 2026-10-19 NVIDIA: Tavily’s August 7, 2026 article says NVIDIA used its search in science-reasoning examples, used its search and browsing in the BrowseComp evaluation harness, and used its web search for a finance-agent benchmark.
Original source · 2026-08-07 2026-10-19 Higgsfield AI: NVIDIA's case study reports that Higgsfield's Blackwell migration across Nebius and other NVIDIA Cloud Partners supported 22 million users and over 6 million pieces of AI-generated content per day.
Original source 2026-10-19 Inceptron: Inceptron announced that Hedy’s EU cloud AI processing now runs on its infrastructure. Hedy’s EU region generates approximately 6 billion tokens monthly, with EU-only processing covered by the companies’ data processing agreement.
Original source · 2026-09-29 2026-10-19 Voxel51: Nebius reports that the automotive team generated and quality-checked hundreds of realistic scene variations through the Voxel51 pipeline, reducing the time needed to create model-ready training data.
Original source 2026-10-19 JetBrains: Tavily's customer story reports 10–20 minutes saved per avoided wrong-API debugging cycle in Junie and claims zero hallucinated APIs on recent and niche libraries. Its April 8, 2026 post also describes Tavily working without configuration through the Junie CLI.
Original source 2026-10-18 Photoroom: ChannelX reported on October 13, 2026 that photos processed by Photoroom for Vestiaire Collective also ran in performance marketing ads, which became 2% more efficient. Digital authentication was unaffected.
Original source · 2026-10-13 2026-10-18 Microsoft: Reuters reported, citing The Telegraph, that Microsoft terminated its agreement for Nscale’s planned data center in Loughton, Essex. Microsoft and Nscale had not responded to Reuters’ requests for comment.
Original source · 2026-10-18 Atlas update 2026-10-18 The earnings-call transcript does not reliably identify Genies, Inc. as a Nebius customer.
Original source Atlas update 2026-10-18 RoboCo is an individually maintained homelab working prototype; its documented Nebius sandbox use does not establish an eligible commercial operator.
Original source 2026-10-18 xAID: Nebius’s undated customer story reports that xAID’s medical-imaging model achieved a macro-average F1 score of 0.8 overall and 0.86 for five acute conditions relevant to emergency care.
Original source 2026-10-18 Voxel51: Nebius explicitly describes the synthetic-data pipeline built with Voxel51 and NVIDIA as a production pipeline, adding production status to the previously documented reference architecture.
Original source 2026-10-18 vLLM project: Nebius's existing customer story documents RLHF workloads on its compute clusters for synthetic data generation, reward modeling and real-time inference evaluation. It also states that vLLM plans to continue using Nebius infrastructure for inference development.
Original source 2026-10-18 Openbenchmarks: A September 2 benchmark established that Openbenchmarks tested Nebius's self-serve inference API on 600 lookup, classification and extraction requests. Its published results reported 269.3 tokens/s median visible-answer speed, 84.83% task success and 14.83% operational failures.
Original source · 2026-09-02 2026-10-18 Meta: The London Institute for Mathematical Sciences describes Nebius as supporting Meta’s training and deployment of frontier models at scale.
Original source 2026-10-18 Liner: A July 2026 disclosure established that Liner also benchmarked Tavily Research and Tavily's advanced answer, extending the previously documented basic-search testing.
Original source · 2026-07-30 2026-10-18 Converge Bio: An August 4, 2025 interview with Nebius’s Ilya Burkov reported that Converge Bio used Nebius’s AI platform to analyze 36 million single cells in days rather than the months typically required.
Original source · 2025-08-04 2026-10-18 BladeRunner Labs: Named Nebius GPU cloud customer
Original source · 2025-05-08 2026-10-18 Requesty: Routes model inference requests to Nebius
Original source 2026-10-18 Custodio Legal: Legal document generation and embeddings with Token Factory
Original source 2026-10-18 Pleias: TractoAI for model training and data processing
Original source 2026-10-18 Kagi Assistant: Uses Nebius model APIs to generate Assistant responses
Original source 2026-10-18 Lurus: Operates AI models on Nebius Token Factory
Original source 