Nebius

Most teams can build an AI agent. But building one that’s reliable, observable, and cost-effective in production remains a real challenge. On Tuesday, August 4th, we're closing that gap live in a webinar, and you’re invited. Join Nebius, LangChain, and Tavily as we build a compliance audit agent from a real-world case study, first deployed using the Nebius Agents Blueprint, then rebuilt with LangChain Deep Agents on NVIDIA Nemotron 3 Ultra. The result is frontier-level quality at roughly one-tenth of the cost, with no model fine-tuning required. You’ll leave with: - A measured comparison of open-model and frontier-model agent architectures showing the real quality and cost trade-offs, on the same task - Grounding and retrieval with Tavily and Pinecone, observability with LangSmith, simulation testing with Snowglobe - The complete architecture and code, so you can reproduce the implementation after the session Leading the build: Devang Sachdev and Tikhon Roshchupkin (Nebius), Srimanth Tangedipalli (LangChain), and Lakshya Prakash Agarwal (Tavily, part of Nebius Group). Can't join live? Register anyway! Everyone who signs up gets the recording. https://lnkd.in/edktgARW

What does building sustainably look like from the inside?   Today, we published our second annual Sustainability Report, covering how we build our AI cloud to deliver more high-performance compute while using fewer resources.   Our Head of Sustainability, Daria Mukhortova, sat down with the leaders involved to discuss what it takes to make this possible: the hardware choices, the community conversations, and why we build for efficiency from day one.   Hear the conversation: https://lnkd.in/dcZuJb6n

Higgsfield created a 95-minute feature film in 14 days, with a $500,000 budget and a team of 15. It was built on Nebius. And while Hell Grind premiered in about as classic a context as you could imagine, at the Cannes Marché du Film, the production context was anything but traditional. Higgsfield's Soul Cinema and Cinema Studio gave the team cinematic control over AI-generated video: 300–500 variations per scene, video prompts running 3,000–4,000 words, and character consistency held across the full 95 minutes. It cost about 100× less than traditional movie workflows require at this length. What made it viable is inference economics. Hell Grind's overnight batch generation meant spinning up hundreds of NVIDIA Blackwell GPUs, then scaling them back down by morning. Nebius delivered a custom configuration that improved Higgsfield's inference performance by over 20%. Today, Higgsfield runs around 30 billion tokens per day through Nebius AI Cloud. When inference economics hits the right inflection point, a 15-person team can ship a feature film in two weeks. Higgsfield plans to open-source Hell Grind so other creators can see exactly how it was done. Watch the behind-the-scenes story: https://lnkd.in/eSEP73Vv

The Physical AI Summit is back. Applications for the 2026 Nebius Physical AI Awards are now open. On November 3, the Nebius Physical AI Summit returns to San Francisco, bringing together founders, developers, researchers, investors, and industry leaders shaping the future of intelligent machines. As part of this year's Summit, we're inviting Physical AI startups to apply for one of the industry's premier global competitions recognizing innovation in Physical AI. Each winning startup will receive $150,000 in Nebius AI Cloud compute credits accelerated by NVIDIA AI infrastructure. Finalists will be invited to the Summit, where they will present their work. Winners will be announced live during the Awards Ceremony. Awards will be presented across five categories: • Physical AI Models • Perception & Spatial Intelligence • Simulation & Synthetic Data • Systems & Deployment • Software, Tooling & Orchestration Entries will be evaluated by a distinguished jury including leaders from Nebius, NVIDIA, and other top global technology companies and academic institutions. Applications are open through September 30. Apply today: https://lnkd.in/eyNX6gWt

Kimi K3 is now available on Token Factory. We’re excited to announce that Nebius Token Factory is an official Day 0 partner for Kimi (Moonshot AI)'s Kimi K3. Kimi K3 is the first open-weight model to reach frontier-level performance, a major step forward for open models. It is built for long-horizon coding, knowledge work and reasoning, with native vision and up to 1M tokens of context. Artificial Analysis scores it at 57 on its Intelligence Index, just two points behind GPT-5.6 Sol (max). That puts Kimi K3 at the top of the open-weight field and firmly among today’s frontier models. Developers can access K3 through Token Factory’s OpenAI-compatible API and console today. Give K3 the hard problem. Build with Kimi K3: https://lnkd.in/eyqrDFhD

Openness keeps AI competitive, adaptable, and available to everyone. We are proud to stand alongside NVIDIA and 70+ others who believe the same.

Nebius is co-hosting an evening at The Garage in Riyadh with Tabby | تابي, JetBrains, and NVIDIA. This evening is for ML engineers, AI architects, and technical decision-makers building production AI systems. Date: Tuesday, 29 July, from 4:30 PM AST Venue: The Garage, Riyadh, Saudi Arabia Free to attend — register at luma.com/ydyrw5xc Speakers: Yazeed Al Qarni, Sr. AI Inference Manager, NVIDIA Ibrahim Imran, Solutions Engineer, JetBrains Yahya Aloyoni, CIO, Tabby Gleb Berjoskin, Senior ML Solutions Architect, Nebius On the agenda: AI infrastructure for regulated industries — why the rules are different Governing agents as AI autonomy increases — how teams maintain control Self-hosted LLMs at scale — deploying, monitoring, and benchmarking on bare metal Building a production-grade AI system — architecture decisions, tradeoffs, what breaks first Panel: running AI agents in production — infrastructure, deployment, and governance See you in Riyadh.

Data center ASMR. Real Nebius sounds. Part five, the last of the series. The first four were Nebius people working on machines. This time: the code that keeps those machines running, typed on a particularly satisfying mechanical keyboard by @Mohammad Gufran. This one's for the developers. Headphones recommended. https://lnkd.in/gmmjdPX8 That's the full series. Back next Friday with new Nebius weekend reads and listens.

This is our first full NVIDIA Vera Rubin NVL72 rack, photographed in our Finland data center. The NVIDIA Spectrum-6 switches we took delivery of last week is the scale-out fabric NVIDIA built for Vera Rubin. Compute and network are now being brought up together, in the same data center, by the same teams. The close co-engineering relationship we enjoy with NVIDIA means we work together from the earliest stage of every new platform, shortening the path from first rack to production workloads. After bring-up and validation comes software polishing and production-grade testing, so when Vera Rubin reaches our customers, it delivers its full potential from day one.

Bringing a new drug to market takes about 10 years. NYB.AI has compressed the front end of that timeline from months to hours with Nebius. The Singapore startup screens billions of compounds to find the few that bind to a disease target. That search rests on molecular docking: modeling how a molecule fits a target protein in 3D. The process is compute-intensive at scale and the accelerated timeline requires rock-solid reliability. That's where Nebius AI Cloud comes in. NYB.AI runs its docking and molecular discovery workloads on Nebius, scaling out to run far more campaigns in parallel than before and leaning on the platform’s stability and reliability to keep long runs moving. “We are very impressed with the cloud platform Nebius has, especially with the scale. We could reduce the compute time from months to a few hours, so we could have feedback loops with scientists in a much faster way.” — Duy Trieu, CTO, NYB.AI We’re proud to support NYB.AI as they speed up the search for new medicines. Watch the full story: https://lnkd.in/eNAHvFXX