Nebius Customer Atlas Atlas changes New customers, relationship updates and selected company news. Dates below are when the Atlas published each change.
2026-10-27 Super Protocol: Confidential healthcare AI on Nebius Blackwell GPUs
Original source 2026-10-27 Ultimarii: Lists Tavily as an AI subprocessor
Original source 2026-10-27 Serna Bio: RNA structure-function mapping on Nebius
Original source · 2025-07-18 2026-10-27 Permit.io: Tested DeepSeek R1 for authorization tuple generation
Original source · 2025-02-24 2026-10-27 Tenshen: Tavily news search for curated articles
Original source 2026-10-27 NanoGPT: Tavily search adds current web results to model prompts
Original source · 2025-12-23 2026-10-27 codai: Lists Tavily for conditional web search
Original source 2026-10-27 Quincy: Trains atypical-speech recognition on Nebius
Original source 2026-10-27 ForestVPN: Nebius powers its support assistant’s language model
Original source 2026-10-27 Leash Bio: Nebius H100s for protein-ligand model research
Original source 2026-10-27 oneclick Assist: Token Factory for product support responses
Original source · 2024-07-17 2026-10-27 Duoc UC: Instructor training through Nebius Academy
Original source 2026-10-27 CloudSigma AI Mission Control: Lists Nebius AI for LLM inference
Original source 2026-10-27 Taski: Tavily web search during AI chatbot voice calls
Original source 2026-10-27 Voxel51: Porsche, Volkswagen Group brands and CARIAD announced a shared autonomous vehicle data flywheel integrating Voxel51, NVIDIA and Microsoft. Voxel51 provides data search, curation, labeling and validation workflows.
Original source · 2026-10-21 2026-10-27 AIMultiple: In its December 2025 search benchmark, AIMultiple measured Tavily latency of 998 ms. Its MCP benchmark separately reported 38% success and 14-second completion time for Tavily’s single-agent web search and extraction tasks.
Original source 2026-10-27 Together AI: Together AI’s July 17, 2025 FutureBench article states that its question-generation agent uses Tavily for additional context when needed. Its baseline SmolAgents prediction agents also use Tavily to find recent information and expert analysis.
Original source · 2025-07-17 2026-10-27 Stanford · CRISPR-GPT: In a Beyond Longevity interview, a Nebius representative said Nebius-powered CRISPR-GPT was used to seek a cure for a rare disease and had an impact on a three-year-old patient.
Original source 2026-10-27 Positronic Robotics: Competition rules state that Positronic runs and scores submissions on Nebius GPUs. The competition privacy policy identifies Nebius as its provider of compute, storage, container registry and secret storage, with storage in Finland.
Original source 2026-10-27 Prima Mente: Prima Mente’s October 24, 2026 Pleiades-Neuron research article credits Nebius with backend infrastructure and compute for training and inference in its work on detecting ultra-rare neuronal cell-free DNA in blood.
Original source · 2026-10-24 2026-10-27 Photoroom: ChannelX reported on October 13, 2026 that a randomized comparison of Photoroom-edited Vestiaire Collective listings against its in-house clipping tool found 5% higher listing liquidity and 3–7% more unique product-page views.
Original source · 2026-10-13 2026-10-27 Nucleo: In an August 4, 2026 interview, Nucleo said it planned to use Nebius compute to lay the groundwork for oncology foundation models for pan-cancer detection, lesion quantification and classification.
Original source · 2026-08-04 2026-10-27 Cotal: Cotal's August 28, 2026 post reported the first live run of its Nebius integration during a two-day hackathon, which also served as the first external test of its hosted mesh.
Original source · 2026-08-28 2026-10-26 VESSL Cloud: Nebius compute for customer-selected workloads
Original source 2026-10-26 Gysho: Transient AI inference on Nebius
Original source 2026-10-26 LoyJoy: Tavily web search for customer-facing AI agents
Original source 2026-10-26 EvaluAgent: Hosts LLMs on Nebius
Original source 2026-10-26 Stonetotem: Token Factory for AI language processing
Original source 2026-10-26 Daytona: Lists Nebius as a cloud infrastructure subprocessor
Original source 2026-10-26 Modal: Lists Nebius as a cloud hosting subprocessor
Original source 2026-10-26 Applied Compute: Nebius cloud infrastructure for AI
Original source 2026-10-26 RespawnHost.com: AI support replies via Eden AI and Nebius
Original source 2026-10-26 mandari: Token Factory for council-document AI and a Work assistant
Original source 2026-10-26 Inception: Nebius infrastructure for LLM training and serving
Original source 2026-10-26 9brains: Nebius models for business chat and document analysis
Original source 2026-10-26 VESSL Cloud: At its strategy presentation, VESSL disclosed first-half 2026 revenue of KRW 130 billion and forecast full-year revenue of KRW 250 billion. It outlined plans for eight Korean AI data center hubs and 200,000 GPUs by 2030.
Original source · 2026-10-15 2026-10-26 VESSL Cloud: VESSL announced that it operates global inference for Upstage's Solar Pro 4 using GPU servers in Korea, Europe and the United States. It reported average processing speeds doubled and per-token costs fell by up to half through caching and quantization.
Original source · 2026-10-08 2026-10-26 VESSL Cloud: The Korea Economic Daily reported, citing investment banking sources, that VESSL and lead investor KB Securities were recruiting investors for a roughly KRW 200 billion Series C. National Growth Fund participation remained under consideration.
Original source · 2026-09-23 2026-10-26 VESSL Cloud: VESSL announced a cloud supply agreement covering 128 NVIDIA B200 GPUs across 16 nodes for SK Biopharm. Service began in August for drug research and medical foundation model training.
Original source · 2026-09-02 2026-10-26 VESSL Cloud: The Bell reported that VESSL was pursuing up to KRW 400 billion in GPU financing and had submitted a rack order request to Supermicro. Discussions with foreign investment banks were ongoing.
Original source · 2026-08-26 2026-10-26 VESSL Cloud: The Bell reported that VESSL secured more than KRW 100 billion in orders during 2026, entirely from its GPU cloud business.
Original source · 2026-08-21 2026-10-26 VESSL Cloud: VESSL AI announced that it secured more than 1,000 NVIDIA B200 GPUs through cooperation with SK Telecom. It planned phased supply through VESSL Cloud beginning in August.
Original source · 2026-08-07 2026-10-26 Modal: Modal raised a $750 million Series D led by Accel at a $15.75 billion post-money valuation. The company also disclosed annualized revenue above $1 billion, tripled usage since its Series C and growth exceeding 15 times year over year.
Original source · 2026-10-14 2026-10-26 Inception: Inception made Mercury 2 available through Baseten as part of Baseten for Model Labs. The distribution arrangement lets existing Baseten customers access the model through their current provider.
Original source · 2026-07-29 2026-10-25 Aline: Lists Tavily for LLM and RAG search
Original source 2026-10-25 Opverio: Uses Tavily for watchlist search.
Original source 2026-10-25 Levano: Tavily web search for its legal AI assistant
Original source 2026-10-25 Zyler: Tavily search for desktop AI features
Original source 2026-10-25 OloLand: Tavily web search for finance-agent benchmark evaluations
Original source 2026-10-25 Artificial Analysis: Benchmarks Nebius inference endpoints and Tavily search
Original source · 2024-11-28 2026-10-25 Keenable: Evaluates Tavily in its NEEDLE search benchmark
Original source · 2026-08-27 2026-10-25 Baljia AI: Lists Tavily for web search, crawling and research
Original source 2026-10-25 Backdrop: Named in Tavily’s trusted-user list
Original source 2026-10-25 Fact Engineering: Tavily web research for live fact-checking
Original source 2026-10-25 Everygen: Tavily web search and page reading in Studio
Original source 2026-10-25 Ever Rec: Tavily search and page extraction for research agents
Original source 2026-10-25 MindMaking: Tavily web search for chat and research features
Original source 2026-10-25 Signal: Lists Tavily among its public web research tools
Original source · 2026-08-01 2026-10-25 Zola Analytics: Tavily web search for web-data lookups
Original source 2026-10-25 TrainLoop: Trains clinical reasoning models on Nebius GPUs
Original source 2026-10-25 MyQandA: Nebius infrastructure for website chatbot model requests
Original source 2026-10-25 Kaon AI: Custom inference infrastructure on Nebius
Original source · 2026-07-08 2026-10-25 Praxis Repina: Uses Nebius to generate vectors for internal inquiry search.
Original source 2026-10-25 University of Miami: Nebius compute supports shared mathematics research
Original source · 2026-08-26 2026-10-25 azena: Open-model inference for Model Garden and Brand Studio
Original source · 2026-10-21 2026-10-25 Prolumios: Token Factory and Tavily for pre-meeting research
Original source 2026-10-25 Sereact: Sereact and Teradyne Robotics announced a partnership combining Cortex with Universal Robots arms for retail, logistics and manufacturing. The companies say they have already shipped hundreds of robots in dual-arm systems to European customer sites.
Original source · 2026-10-21 2026-10-25 Requesty: Requesty’s May 9, 2026 disclosure reported that median Nebius request latency fell from 22.1 seconds in April 2025 to 2.3 seconds in April 2026, an 89% reduction, with at least 50,000 requests in each month.
Original source · 2026-05-09 2026-10-25 Rechnungskit: Rechnungskit's privacy policy documents tax firms using Kit in its partner portal, with Nebius processing partner users' chats and requested information about clients who have authorized access.
Original source 2026-10-25 Jinbay: Jinbay's March 26, 2025 announcement says its recruiting-agent workflow can complete candidate research in minutes that previously took a day.
Original source · 2025-03-26 2026-10-25 Sereact: Sereact, Zalando and CEVA Logistics launched automated returns processing in Greven, Germany, and Świebodzin, Poland. Cortex-powered dual-arm systems are operating live to grasp, identify and sort returned fashion items under a Robotics-as-a-Service model.
Original source · 2026-10-07 Atlas update 2026-10-25 EUrouter announced on October 23, 2026 that it stopped routing requests to Nebius and replaced its Nebius-only model route.
Original source 2026-10-25 WEKA: WEKA’s September 24, 2026 benchmark article reported that 82.5% of requests met a two-second time-to-first-token threshold with Augmented Memory Grid, compared with 16.2% using GPU memory alone on Nebius B300 infrastructure.
Original source · 2026-09-24 2026-10-25 SocialCrawl: SocialCrawl's GitHub documentation describes Search Everywhere sending a query to parallel sources including Tavily and returning ranked, clustered results enriched with top comments.
Original source 2026-10-25 RoboForce: Nebius’s October 21, 2026 announcement names RoboForce among customers already using Physical AI Workbench to deploy physical AI in the real world.
Original source · 2026-10-21 2026-10-25 Prosus: Nebius’s October 21, 2026 announcement quotes Prosus’s chief AI officer confirming Nebius AI Cloud use for fine-tuning. It also identifies lifestyle and e-commerce businesses as the scope supported by Prosus’s Token Factory workloads.
Original source · 2026-10-21 2026-10-25 Ocean: Tavily’s October 14, 2026 case study reported that over four months, the number of customers using Ocean’s agents grew 1.5x and investigation volume nearly doubled, with no additional human web research effort. The integration also let Ocean avoid building and running its own web research layer.
Original source · 2026-10-14 2026-10-25 Lightning AI: In a July 16, 2026 post, Lightning AI reported an 88% success rate for Nebius in its benchmark of compute stability and availability through its platform.
Original source · 2026-07-16 2026-10-25 London Institute for Mathematical Sciences: An August 1, 2024 researcher article said Mikhail Burtsev hoped to use GPU time awarded by Nebius to advance knowledge graphs, improve data capture and develop mathematical AI models that learn complex numeric patterns and dependencies.
Original source · 2024-08-01 Atlas update 2026-10-25 EUrouter: In its 23 October 2026 changelog, EUrouter said it stopped routing to Nebius because Nebius no longer committed to EU processing on public endpoints. Shared models continue through other EU providers. EUrouter also retired the Nebius-only Hermes 4 405B route, named DeepSeek V4.1 Flash as its replacement and disabled adding Nebius BYOK keys.
Original source 2026-10-25 azena: Azena announced that it became an official Nebius partner. Its Model Garden uses Nebius Token Factory to provide access to open models, with customer-specific usage accounting and spending limits.
Original source · 2026-10-21 2026-10-25 Revolut: Nebius’s undated customer story says it supports Revolut’s cross-sell optimization, text-to-image card customization and retraining of lighter open-source models. It also reports that money-laundering investigations went from up to six human touchpoints to close to one.
Original source 2026-10-25 Prosus: A 22 October 2026 article quotes Prosus chief AI officer Euro Beinat reporting doubled throughput with Nebius Token Factory and autoscaling that handles peaks without holding GPU capacity in reserve.
Original source · 2026-10-22 2026-10-25 Lynx Analytics: Nebius’s customer story describes Lynx Analytics building an automated theorem-proving framework in LynxKite 2000:MM. A graph neural network trained in LynxKite and powered by Nebius AI Cloud served as its graph-based retriever.
Original source 2026-10-25 JetBrains: A Nebius job posting names JetBrains Mellum as a TractoAI client.
Original source 2026-10-25 Black Forest Labs: Nebius’s October 21, 2026 announcement says Black Forest Labs used its NVIDIA B200 infrastructure for Self-Flow research foundational to FLUX 3. Black Forest Labs’ engineering head also described Nebius as dependable infrastructure for sustained visual-model training.
Original source · 2026-10-21 2026-10-24 TD SYNNEX: TheEnergyMag reported on October 21, 2026, that an agreement announced October 20 gives partners access to Nebius AI Cloud across Europe for training and inference, with reserved-capacity options and technical support.
Original source · 2026-10-21 2026-10-23 Higgsfield: Trained SOL 2.0 on Nebius Blackwell GPUs.
Original source · 2026-07-07 2026-10-23 World Labs: World Labs said it signed a definitive agreement to join AMD. The transaction is expected to close by the end of 2026, subject to regulatory approvals and customary closing conditions.
Original source · 2026-09-28 2026-10-23 Higgsfield: Higgsfield announced a $400 million Series B led by DST Global at a $5.4 billion valuation. The company also said its annualized revenue had reached $700 million.
Original source · 2026-08-17 2026-10-23 Recraft: An ACM SIGGRAPH article dated July 7, 2026, reports that Recraft V4 had served more than 350 million images and won over 50% preference on PartiPrompts. A June 2, 2026 Nebius post ranks Recraft top three on Artificial Analysis’s Text-to-Image Leaderboard and first among independent labs.
Original source · 2026-07-07 2026-10-22 Datadog: Nebius AI services and H200 monitoring validation
Original source 2026-10-22 Pulsar Platform: Nebius model APIs for its core platform
Original source 2026-10-22 Aegis: Token Factory inference for Rust code repair
Original source 2026-10-22 bunq: Runs Finn’s open-weight models on Nebius
Original source · 2026-10-21 2026-10-22 Bellboy Robotics: Uses Nebius Physical AI Workbench for physical AI deployment
Original source · 2026-10-21 2026-10-22 Moon Surgical: Uses Nebius Physical AI Workbench
Original source · 2026-10-21 2026-10-22 bunq: Benjamin Kleppe said bunq now runs Finn’s open models on dedicated NVIDIA GPUs in the EU. He reported a week-long Kimi K3 test in support conversations with quality within 1% of the best proprietary model, a 1.9-second median response time, and six days from the first dedicated endpoint to live traffic. He said GLM 5.3 is in a live test sized for Finn’s full workload on half the GPUs.
Original source · 2026-10-21 2026-10-22 Voxel51: Porsche, Volkswagen Group brands and CARIAD announced a shared autonomous driving data system with NVIDIA, Microsoft and Voxel51. Voxel51 supplies data search, labeling, curation and validation workflows, with infrastructure running on Azure.
Original source · 2026-10-21 2026-10-22 TD SYNNEX: TD SYNNEX announced a pan-European agreement giving its partners access to Nebius AI Cloud for AI training and inference. Partners can obtain NVIDIA GPU capacity through a consumption model without upfront infrastructure investment.
Original source · 2026-10-20 