Nebius monthly digest: July 2025

Our July’s digest describes how our customers like Stanford University and Shopify use flexible compute capacity and the steps we are taking to boost performance. Nebius’ first anniversary also took place in July, marked by Nasdaq in Times Square, and there was news from across the ocean as well.

Nebius in the spotlight at Times Square

We are grateful to Nasdaq for celebrating our first anniversary on the iconic Times Square billboard in New York. We’re proud to be a publicly traded company and a part of the Nasdaq family.

Elastic for both researchers and industry giants

• CRISPR-GPT is AI geneediting expert designed at Stanford. Nebius’ flexible infrastructure that helped build CRISPR-GPT allowed to transition seamlessly from prototyping to large-scale model training.
• Shopify’s objective when accessing GPUs is to offer its developers easy access to compute whenever it is needed to make rapid iteration and experimentation possible. This involves the use of Nebius’ large training clusters.

Advancing raw performance, ways to access your cloud and security by design

• We launched Enhanced Object Storage, a new performance-focused class of storage. It’s great for streaming datasets and weights to GPU and for checkpointing. The new class offers up to 10 GB/s write throughput per client and unlimited capacity.
• We entered a strategic partnership with Mithril . This integration enables Mithril customers to access the Nebius AI Cloud regions through the Mithril Console for experimentation, largescale training and production inference workloads.
• Back in June, we announced the Nebius MCP Server, our integration that connects Claude by Anthropic or other AI chatbots to the Nebius AI Cloud infrastructure. Now you can read how this conversational approach transforms your infrastructure management workflow.
• Explore our newly launched trust center to understand how we handle security, compliance and data protection at Nebius. We’ll be adding more documentation and resources soon. Transparency builds trust, and we’re taking another step in this direction.

Visiting 10 Downing Street

We were honoured to attend the first UK Sovereign AI Industry Forum, co-hosted by the Prime Minister’s Advisor on AI Opportunities Matt Clifford , Department for Science, Innovation and Technology and NVIDIA . Our own General Manager for EMEA Rashid Ivaev participated in the event.

Nebius’ 2024 Sustainability Report and tech docs: dive in to the details

• Nebius is building AI infrastructure sustainably and at scale. Published in July, the 2024 Sustainability Report highlights how this approach creates value across the ecosystem — from lowering the total cost of ownership (TCO) to reducing environmental impact and enabling broader access to technology.
• Read this blog post to understand in great detail how to launch production-grade AI agents at scale. From proof-of-concept to production-level systems, our Product Marketing Manager Dylan Bristot breaks down the whole stack.
• Compute just got more powerful — the NVIDIA HGX B200 platform is now available, with updates across related docs. You’ll also find guidance on using preemptible VMs, including compatibility details, creation steps and platform-specific behavior. Another addition is a breakdown of VM lifecycles and statuses.
• Managed Kubernetes network setup made easier — learn about network requirements for clusters, how to fix the “Unable to allocate CIDR from the pool” error and how to use non-default subnets for clusters and node groups.
• Soperator gets more practical — run Slurm batch jobs with ease and test distributed GPU performance by using an NCCL all-reduce benchmark in your cluster.
• New articles in Object Storage show how to transfer data into buckets and manage lifecycle rules for object expiration and transition.
• Third-party tools are now in one place — integrations now live in a new dedicated section. Kick it off with a tutorial on installing dstack, an open-source tool for orchestrating AI workloads.