Nebius

Your weekend listening: Roman Chernin, Nebius Co-Founder and CBO, on 20VC with Harry Stebbings, talking about the hottest topics for AI infrastructure right now: is it a bubble, who wins as inference displaces training, why cheaper AI creates more demand (not less), and why companies building on AI today are just the beginning of the demand curve. Find it on YouTube or wherever you get your podcasts: https://lnkd.in/daU8KAzQ

We're introducing the Nebius Agents Blueprint: an open reference architecture for building, operating, and continuously improving #AIAgents in production. Most agents fail in production not because of the model, but because of the infrastructure around it: wrong retrieval, stale answers, no observability, untested logic, and inference costs that don't survive scale. The Nebius Agents Blueprint addresses the most common production failure modes by combining best-of-breed components across every layer of the #AgenticAI stack: LangChain for orchestration, Pinecone Nexus for retrieval, Tavily for live grounding, LangSmith for observability, and Snowglobe by Guardrails AI for pre-deployment simulation, all running on Nebius Token Factory. We validated it by running a regulatory compliance auditor through four configurations on the same task. Open models on owned inference cut costs by 72% and improved precision by 20% with no fine-tuning required. Read the announcement here: https://lnkd.in/eT2ABaRi

Nebius has announced an investment of £1.7 billion to build out capacity in the UK with three new deployments of NVIDIA infrastructure. The three new sites will deploy the latest generations of NVIDIA’s full-stack, end-to-end AI factory platform technology and combined will reach 65 MW when fully ramped up in 2027. In addition to the capacity investment, Nebius is expanding its commercial and AI R&D hub in London and establishing local partnerships with universities and research institutions through Nebius Academy. Nebius's build-out in the UK aligns with the UK Government's AI Opportunities Action Plan and adds domestic compute capacity so British enterprises, researchers, and public services can build and deploy AI at scale. UK AI Minister Kanishka Narayan MP said: “Nebius’s investment brings significant AI compute into the UK, giving companies what they need to train, test and run advanced systems here at home – helping drive productivity by rolling AI out widely across the economy.” Arkady Volozh, founder and CEO of Nebius, said: "The UK is one of the places where AI is being built, deployed, and adopted at the same time — by startups, by enterprises, and by the public sector. The work is happening here and the demand is here. And we are also here for the long run." Read the full press release: https://lnkd.in/eE9Pwu7Q

"The model was never the hard part." If you've worked on physical AI, you know that the real challenge is...well, everything else. So we built the Nebius Physical AI Workbench for everything that isn’t the model: simulation, synthetic data, training, eval, deployment — NVIDIA Cosmos 3, Isaac Lab, Isaac GR00T, and other physical AI tools, pre-wired and agent-driven. Spend your time on the research instead of the infrastructure. Open source. On GitHub now. Learn more over on the Nebius blog → https://lnkd.in/ehV-PQAb

Nemotron 3 Ultra is live on Nebius Token Factory. This is a clear signal for where open models are headed: long-running agents that can handle serious coding, research and enterprise workflows. With the model side moving fast, the next hard problem is production. Teams don’t just need access to stronger open models. They need to serve them with speed, reliability, predictable economics and the operational controls enterprises expect. That’s the layer we’re building with Nebius Token Factory: production inference for the models teams actually want to use. Excited to see @NVIDIA keep pushing open agentic models forward and to keep building together. Run it today: https://lnkd.in/efsNN2Hu

Congrats to the Recraft team — top-3 on Artificial Analysis Text-to-Image Leaderboard, and top-1 among independent labs. Recraft trains on Nebius, and results like this show what's possible when great models meet great infrastructure.

NVIDIA's Build Your Own Transaction Foundation Model developer example gives financial institutions a starting point for training transformer embeddings on proprietary transaction data. Nebius AI Cloud is where it runs in production, from GPU-accelerated data prep through multi-node training to managed inference on Token Factory. Revolut is already there. Their PRAGMA model family was trained on 40B+ events across 25M users on Nebius AI Cloud. 3x faster pre-training. 21% precision improvement on fraud detection. The developer example is available with 1-click deployment from the Nebius console. Discover more over on the blog -> https://lnkd.in/eEMnNibf

Merci M le Président. Nous sommes fiers de construire l'avenir de l'IA en France !

BuilderShip is an AI hackathon running now through June 12, co-hosted by Nebius, Composio, and Tavily. The top 30 builders earn a spot at the finals on an 89-foot yacht on the San Francisco Bay and a full day of 1:1s with the engineers and investors judging their work. The application is a public post. Show something you've built — an agent, a demo, a repo — tag Nebius, Composio, and Tavily, and you're in the running. Every builder gets access to the sponsor stack from day one: Nebius GPU credits, Token Factory inference keys, Composio integrations, Tavily search, and OpenClaw runtime. The winner takes home $50,000 in cloud credits and an NVIDIA DGX Spark. Finals are June 14 on the San Francisco Bay. Submit by June 12 → ship.builders

Congrats to the JetBrains team on the Mellum2 launch! Nebius was part of the training infrastructure for this model. Excited to see what teams build with it.