Overview Nebius, a global AI cloud platform, delivers a unified full-stack AI cloud that spans the complete AI journey – from compute capacity to software and services that enable fast and efficient training and inference at scale. Founded around deep in-house technological expertise, Nebius offers a comprehensive and integrated suite of AI and ML cloud solutions, including both hardware and software built in-house. This combination of AI-optimized hardware and software enables us to deliver high-performance GPU compute clusters, storage, managed services, and advanced tools for AI model training and inference at enterprise-scale. Headquartered in Amsterdam and listed on Nasdaq, Nebius Group offers one of the few global, at scale, multi-tenant clouds purpose built for AI, with a significant presence in Europe, the U.S., and other geographies around the world. Nebius Group includes Nebius as well two distinct businesses that operate under separate brands: Avride, a leading developer of autonomous vehicles and delivery robots; and TripleTen, a leading edtech platform reskilling people for careers in tech. Nebius Group also owns significant equity stakes in ClickHouse and Toloka, both of which have been spun out of the group. Nebius: Full-stack AI cloud platform Nebius provides a full-stack AI cloud - from silicon to software - encompassing data centers, in-house-designed infrastructure, and an integrated software layer with advanced tools to support any use case across the AI lifecycle, from data preparation and model training to inferencing and production deployments at scale. Our platform is architected to serve the different types of workloads at any stage of scale or maturity of the underlying business. Nebius offers solutions for: ● IT operations, DevOps and Platform Engineering teams needing access to compute resources in order to set up and manage the organizations’ infrastructure; ● Data Scientists, ML Researchers and ML Engineers that want easily accessible and scalable capacity for key workloads; and ● AI Engineers or AI Product Managers that want to consume AI services in order to fine-tune or serve models and agents, without interacting at all with the underlying infrastructure. The foundation of our cloud is a highly efficient and sustainable hardware infrastructure layer that delivers scalable compute, storage, and networking resources engineered for high-performance AI workloads. We build our infrastructure from the ground up, designing servers and racks in-house, embedding innovation in the design of our data centers and developing software for workload orchestration and optimization. We believe this results in greater maximization of compute performance and lower customer Total Cost of Ownership (TCO). The performance of our hardware is supported by our long-standing partnerships and collaboration with leading chipmakers and OEMs, such as NVIDIA, and a consistent track record of being one of the first-to-deploy the latest generation of NVIDIA GPU chips. Our designs optimize power and cooling efficiency, lower latency, and create seamless integration with our cloud platform. This improves our performance and reliability while also strengthening our value proposition by combining the reliability and user experience of a hyperscaler with the flexibility and efficiency of purpose-built AI infrastructure. Built on top of this robust foundation is our proprietary, purpose-built AI cloud platform, which streamlines and accelerates AI development and deployment. We operate one of the few global, multi-tenant AI-specialized clouds on the market. We can quickly and efficiently provision compute resources on demand from a single node to thousands of nodes with high-performance storage GPU-to-GPU networking, and managed services, including advanced AI and ML tools. This global, multi-tenant architecture provides flexibility and ensures that customers can handle everything from small-scale experiments to large-scale enterprise-grade AI workloads – such as model pre-training, training and data pipelines, post-training/fine-tuning, and inference at enterprise-scale, without over-provisioning, adjusting resources dynamically to meet their evolving needs. Our cloud is designed to serve the needs of organizations of all sizes with built-in enterprise-grade observability, security and compliance. 33 Table of Contents As AI is rapidly becoming a general-purpose technology, we are well positioned to service customers from large enterprises, established software vendors, scaled AI companies and startups to research labs and individual developers building the next generation of AI models, applications and services. We believe they choose our platform for its flexibility, reliability, and comprehensive support for diverse AI workloads of all sizes. Our customers are building transformative applications across a diverse range of industries, such as physical AI, healthcare and life sciences and media and entertainment. The following sections highlight the key layers of the Nebius full-stack platform. Data Centers Our team has decades of experience in developing capacity at scale. For this reason, we build our infrastructure from the ground up, designing servers and racks in-house and embedding innovation in the design of our data centers resulting in optimization of compute performance. The performance of our hardware is supported by our long-standing partnerships and collaboration with leading chipmakers and OEMs, such as NVIDIA, and a consistent track record of being one of the first-to-deploy the latest generation of NVIDIA GPU chips. We leverage our advanced data center design to enhance unit economics by reducing energy overheads, optimizing IT workload allocation, and lowering server maintenance costs. This design also improves utilization and helps to ensure seamless scalability at each site. We select data center sites based on access to power, existing industrial zoning, and alignment with local governments on delivering economic benefits to the region. In 2025, we owned and operated a data center in Finland, signed a build-to-suit location in New Jersey and signed several co-location agreements in Kansas City, the UK, Israel, France and Iceland. In February 2026, we announced the expansion of our data center footprint to include nine additional sites across seven locations in the US (Missouri, Alabama, Oklahoma, and Minnesota) and Europe (France, UK) and the Middle East (Israel), bringing our total contracted power to more than 2 GW. The majority of this capacity will be deployed from data centers that we own and design internally. 34 Table of Contents We operate four types of data centers: Greenfield We own the land and manage the power infrastructure, and our engineers design every aspect of the data center. This approach offers the greatest flexibility for optimizing energy efficiency and performance. Our Missouri and Finland data centers are greenfield facilities. Finland features what we believe to be one of the world’s leading power usage effectiveness (PUE) levels, employs an air-based free-cooling for high-density workloads that does not rely on external water intake. The site also integrates heat recovery, which has historically supplied up to two-thirds of local heating demand. In 2025 this system reused almost 20 GWh of server heat, contributing to an estimated 10% reduction in household heating costs. Similar cooling system designs will be implemented across new greenfield sites, such as Missouri and Alabama (announced in early 2026). Brownfield To accelerate speed to market, we also consider land that has existing assets already in place, including buildings, power facilities, and other structures. These sites provide additional opportunities to utilize our unique designs and generate optimal data center efficiency with faster time to market. Build-to-suit We may also partner with a developer who owns the land and has secured the power. Under such agreements Nebius would still provide custom specifications for the data-center buildout. This allows us to drive energy efficiency and infrastructure optimization within the facility. Our New Jersey facility follows a build-to-suit model. Co-location We lease capacity at existing data centers through third-party providers, enabling us to rapidly deploy compute resources. While we do not own these facilities, we apply rigorous selection criteria to ensure they meet our performance, reliability and scalability standards. Operational efficiencies are achieved through the deployment of our in-house-designed racks, optimizing power consumption. Locations of some of our current co-location sites include France, Iceland, UK, Israel, and Kansas City in the US. Data center footprint We have a broad data center footprint across Europe and the US. We primarily define our data center capacity in three ways: ● Contracted power is capacity that has been secured by land and contracted power commitments. As of February 2026, we had contracted more than 2 GW of power. ● Connected power is capacity that has power connected into data centers; and ● Active power is capacity being consumed by IT equipment and available for revenue generation. As of December 31, 2025, we had approximately 170 MW of active power capacity across the globe and are rapidly expanding our footprint. Our preferred method of acquired capacity is through greenfield data centers, though we remain opportunistic with co-locations as this offers faster time-to-market capacity. We are actively exploring additional sites to significantly expand our capacity. ​ 35 Table of Contents We operate or are in-contract across more than 16 data center locations, including: ​ Europe and Middle East ● Mäntsälä, Finland – a greenfield data center built to our own design specifications to optimize power and hardware for greater efficiency. ● Paris, France – in July 2024 we signed an agreement for our Paris data center, our first co-location facility. ● Keflavik, Iceland – in December 2024, we signed an agreement in connection with adding a cluster of thousands of GPUs at a co-location in Iceland. ● London, UK – in November 2025, we secured the contracts to our new facility featuring NVIDIA Blackwell Ultra GPUs, supporting the goals set out in the UK Government’s AI Opportunities Action Plan. ● Israel – in October 2025, we secured the contracts to our new co-location facility housing one of the country's first publicly available AI deployments. ● Israel – in February 2026, we secured the land and capacity needed to commence operations in two new co-location sites in Israel. ● France – in February 2026, we secured the land and capacity needed to commence operations in two new co-location sites in France. ● UK – in February 2026, we secured the land and capacity needed to commence operations in a new co-location site in the UK. United States ● Kansas City – in November 2024, we secured the contracts for our first co-location data center in the US, located in Kansas City, MO. ● New Jersey – in February 2025, we signed an agreement for our first built-to-suit facility, located in Vineland, NJ. ● Missouri – in February 2026, we secured the land and capacity needed to commence operations in a newly owned data center site located in Missouri. ● Alabama – in February 2026, we secured the land and capacity needed to commence operations in a newly owned data center site located in Alabama. ● Minnesota – in February 2026, we secured the contracts needed to commence operations in a new co-location data center site in Minnesota. ● Oklahoma – in February 2026, we secured the contracts needed to commence operations in a new co-location data center site in Oklahoma. 36 Table of Contents Racks and Servers Designing our servers and racks in-house gives us full control over server prototyping, production and deployment, which is a key factor in reducing operational costs, accelerating time-to-market and scaling AI infrastructure. Our servers are engineered to operate at temperatures up to 40°C (105°F), compared to the ASHRAE standard limit of 27°C (80°F). This makes air cooling sufficient to maintain optimal performance for the current generation of chips, even those with high thermal density. Beyond energy savings, our proprietary server firmware and toolless rack design simplify maintenance and repairs, so components can be replaced within minutes instead of hours, improving reliability and uptime and offering significant TCO gains for our customers. This also reduces staffing requirements, allowing one engineer to manage thousands of servers. Furthermore, our streamlined design enhances workplace safety, reducing risks associated with complex traditional server maintenance. Our hardware stack consists of the following core components: ● Compute – our compute solutions primarily include GPU instances, providing flexibility for diverse workloads. Our strong relationships with NVIDIA and OEM partners support our ability to consistently provide the latest and most advanced GPU technology available in the marketplace. We announced in January 2026 that Nebius will be among the first NVIDIA Cloud Partners to bring the next-generation accelerated computing platform, the NVIDIA Vera Rubin NVL72, to customers in the US and Europe. Our GPU-based servers include NVIDIA GB300 NVL72, GB200 NVL72, HGX B200, HGX B300, HGX H100 and RTX PRO 6000. ● InfiniBand-connected GPU clusters – we use NVIDIA InfiniBand NDR/XDR GPU-to-GPU interconnects, ensuring high-speed, low latency communication. ● Storage – we offer a range of storage solutions to meet diverse customer demands, including block storage, shared file storage and object storage. Our platform combines in-house storage offerings with solutions from leading third-party storage providers, giving customers flexibility to optimize storage based on their individual needs and use case. Infrastructure-as-a-Service (IaaS) Built on top of this robust foundation are our proprietary, purpose-built IaaS solutions for AI-native workloads. We operate one of the few global, multi-tenant AI clouds on the market. We can quickly and efficiently provision compute resources on demand from a single node to thousands. Embedded orchestration across resource usage enables efficient scaling and fewer performance bottlenecks supported by a unified control plane, enabling efficient scaling of distributed training and inference workloads. Our cloud is designed to serve the needs of large organizations by embedding enterprise-grade security and governance functionality. Teams across DevOps, IT Operations and Platform Engineering operate in this layer. Key capabilities include: ● Elastic scaling and resource allocation – The platform supports dynamic scaling of compute resources based on workload demand, enabling efficient utilization of GPU infrastructure across training and inference workloads. ● We offer compute instances in the form of virtual machines and containers. Container orchestration is based on an upstream managed Kubernetes stack, built on open-source solutions and proprietary components, delivering resilient and extensible infrastructure for managing containerized workloads and services at scale. ● AI-optimized storage offering includes object storage, a shared filesystem, with homegrown or 3rd party options. A Data Transfer Service is available for customers to move object storage buckets across different environments. ● Virtual Private Cloud networking offerings such as routing are also supported. This global, multi-tenant, virtual architecture provides flexibility and ensures that customers can have quick and easy access to AI-optimized compute resources and handle everything from small-scale experiments to enterprise-grade AI workloads – such as data preparation, model pre-training, fine-tuning, and inference at enterprise-scale, without over-provisioning, adjusting resources dynamically to meet their evolving needs. 37 Table of Contents This AI-optimized IaaS layer provides the foundation for higher-level software capabilities, enabling customers to move from infrastructure provisioning to model development and deployment without managing the underlying hardware. Machine-learning operations (MLOps) layer On top of this virtual infrastructure, the MLOps functionality streamlines the entire machine-learning lifecycle, from data preparation, pre-training and post-training/fine-tuning, as well as serving models, offering complete choice for users, depending on the requirements of their use case and infrastructure control. By integrating model optimization and deployment workflows, the MLOps layer reduces time-to-production and enables efficient transition from experimentation to live environments. Data Scientists, ML Researchers and ML Engineers can schedule jobs and deploy applications to align with their own workflows and tooling, at various levels of infrastructure expertise. It includes: ● Apps – a comprehensive catalogue of third-party applications easily accessible as managed services or images, such as MLflow, JupyterLab,vLLM and ComyUI. ●Serverless – a suite of features that allows developers and data scientists to schedule training jobs and use inference endpoints, without having to deal with the complexity of setting up and configuring infrastructure. ●Data Ops – data pipeline and data-centric AI workflows, delivered in partnership with Toloka, support dataset creation, labeling, evaluation, and continuous improvement of data used in production systems. ●Orchestration – various options for scheduling AI workloads including our own open-source solution Soperator, and deep integrations with third-party tools such as SkyPilot, Ray/Anyscale and dStack. For example, Soperator, used by many of our customers, is a Slurm-on-Kubernetes AI workload orchestration offering for machine learning and high-performance compute clusters. It enables robust job scheduling, fault-tolerant training, within a simplified user experience. Artificial intelligence operations (AIOps) layer We also natively integrate a set of features that address the needs of AI Engineers and Product Managers that require less infrastructure complexity to easily serve and fine-tune models in production, at scale. This enables easy access to existing open source and commercial models. Our AIOps layer supports the deployment and operation of AI systems in production, abstracting infrastructure complexity while providing performance, reliability, and governance. These layers are natively integrated into our platform to enable customers to move from infrastructure provisioning to model development and into production deployment within a single platform. Such solutions include: ● Nebius Token Factory – Our enterprise-grade managed inference service, released in November 2025, enabling vertical AI companies and digital enterprises to deploy, optimize and fine-tune open-source and custom models at scale with enterprise-grade reliability and control. Token Factory incorporates autoscaling and performance management capabilities, enabling customers to operate high-throughput inference workloads with predictable latency and cost efficiency. Token Factory supports all major open models, including DeepSeek, GPT-OSS by Open AI, Llama, NVIDIA Nemotron and Qwen, and offers customers the option to host their own models. Unlike traditional GPU-per-hour pricing, Nebius Token Factory is monetized through a token-based model, offering customers greater flexibility and cost efficiency. ● Model Hub – Model Hub offers direct access to various AI models from NVIDIA in the form of containerized microservices (NVIDIA NIMs), easily deployed on top of the customers’ environment, providing more control of infrastructure concepts and configurations. ● Agentic Services – In February 2026, we acquired Tavily, an agentic search provider serving large enterprises and AI technology companies. The acquisition brings real-time search infrastructure into Nebius’s cloud platform, and advances Nebius’s strategy towards a unified platform where vertical AI companies and enterprises can build, tune, and run autonomous agents. Furthermore, we announced plans with Toloka to bring Tendem.ai into the Nebius ecosystem, the first platform to embed vetted human experts directly into agentic workflows for human-in-the-loop validation. Expert judgment is callable via the Model Context Protocol (MCP), the emerging standard for AI tool integration. With those two new additions, Nebius is expanding the integrated software stack developers need to assemble and operate enterprise-grade agentic systems. These capabilities expand the platform toward supporting agentic systems, including real-time data integration (Tavily) and human-in-the-loop validation (Tendem) within production workflows. 38 Table of Contents Enterprise Platform Our cloud is designed to serve the needs of any type of organization by embedding enterprise-grade security and governance functionality. Our multi-tenant cloud platform provides a unified control layer that enables customers to securely access, consume, manage, and operate AI workloads at scale. It integrates self-service interfaces, enterprise-grade security and compliance, cost visibility and resource controls, and robust operational capabilities to ensure reliability and performance across the full AI lifecycle. Our enterprise platform includes: ● Access and experience: Self-service provisioning and well-documented interfaces (GUI, API, IaC/Terraform, SDKs), enable developers and enterprises to easily access and consume the platform based on their preferred workflows and tools. ● Security and compliance: Enterprise-grade features that give organizations the trust, control, and simplicity they need to run their most critical AI workloads in production at scale, delivered as part of Nebius AI Cloud 3.0 “Aether” release in the third quarter of 2025: o Fine-grained IAM (SSO, role-based access) for secure organizational access to the platform. o Independently validated security certifications such as SOC 2 Type II including HIPAA, and ISO 27001, and alignment with NIS2, DORA, ISO 27032, ISO 27701, and ISO 27799 regulatory frameworks for security. o Audit logging and secrets management. ● Cost control and billing: cost visibility through FOCUS-compliant data export, along with capacity planning, quotas, and resource controls, provides transparency and efficient management of infrastructure spend. ● Operations, reliability and sustainability: our full-stack control across every step of the hardware and software build, from servers to agents, uniquely enables us to provide end-to-end observability, automated health checks and self-healing. We also report on energy consumption and provide transparency into efficiency metrics, enabling customers to better manage workloads and support their carbon accounting and reporting. Customers and Go-to-Market Strategy Our primary customers today range from large enterprises, established software vendors, scaled AI companies and startups to research labs and individual developers building the next generation of AI models, applications and services. We believe they choose our platform for its flexibility, reliability, and comprehensive support for diverse AI workloads of all sizes. Our customers are building transformative applications across a diverse range of customer segments and industries. With respect to our go-to-market efforts, we have made and continue to make significant investments in our sales and marketing functions to expand our customer base and build our brand recognition. Our direct sales and channel teams continue to scale, supported by growing pre-sales and post-sales teams that seek to ensure customer success from initial engagement through deployment. This includes dedicated system architects who lead proofs-of-concept and accelerate onboarding, as well as robust ongoing technical and engineering support. We plan to strategically focus our sales and go-to-market organization build out targeting a number of key enterprise verticals that represent end markets that are already seeing early traction in AI adoption and have the potential for long-term value creation. We see attractive opportunities in growing our footprint in physical AI, healthcare and life sciences and media and entertainment, as well as retail and e-commerce and financial services. Below we provide more details on our key target customer segments: Enterprise Customers These include mid-market and larger enterprises that plan to use AI to drive efficiencies and optimized results within their organization. Use cases can range from in-house model development and fine-tuning to the deployment and inferencing of open-source AI models. We anticipate that the scale of deployments from enterprise customers and related compute requirements, in particular inference workloads, will grow substantially over time as AI models become more widely available and cost effective to deploy in production systems. 39 Table of Contents Software Vendors These companies include technology firms that are adopting AI to enhance an existing suite of software services, or to develop new products and applications that harness the power of our AI cloud platform to generate new sources of revenue and heightened efficiency. Our business with software vendors has already started to scale rapidly, for both training and inference workloads, as these customers have already started taking the next step from model building to application deployments at enterprise-scale. AI-Native Tech Companies These are generally VC-backed AI-native technology companies that are building AI-specific solutions and need a full-stack AI cloud service that is flexible, scalable, and able to meet their AI workload needs. The AI workloads that these customers run with Nebius include training, fine-tuning and inferencing using both proprietary and open-source models. Typically, these customers make use of a range of different products and services that are available on our platform, including managed services for workload management and orchestration as well as MLOps tools. This customer segment also includes independent developers and researchers who are able to access our AI cloud platform via our self-service offering, which provides instant access to GPUs on demand. AI Labs These companies are at the forefront of AI research and development and require massive, scalable compute infrastructure to support the training, fine-tuning and deployment of large-scale AI models, particularly large language models (LLMs) that utilize hundreds of billions or trillions of parameters. Their workloads are computationally intensive, demanding high-performance GPUs, low-latency networking and distributed storage solutions to process vast datasets efficiently. Hyperscaler Contracts AI hyperscalers represent important large scale customers for AI compute capacity. The purchasing behavior of these industry-leading technology companies (some of which may offer competing solutions to our AI cloud) differs from our primary customer base and may not include our primary AI cloud services. We may engage in such customer contracts when the supply conditions and economic terms are in the best interests of the company. In 2025 and early 2026, we secured significant long-term committed contracts with two large hyperscalers, Microsoft and Meta. These multi-year, multi-billion dollar agreements offered attractive economics and financing terms, which enabled us to invest in and grow our core AI cloud business. Other businesses and equity stakes Avride Avride is a developer of autonomous driving technology for self-driving cars and delivery robots for use-cases across ride-hailing, logistics, e-commerce, food and grocery delivery. The company’s main operations are in Austin, Texas, with additional R&D hubs in Europe, Israel and South Korea. In 2024, Avride signed a multiyear partnership with Uber to deploy its autonomous vehicles and delivery robots for Uber and Uber Eats in the US. As part of this collaboration, Uber Eats launched delivery services utilizing Avride’s sidewalk robots in Austin and Dallas, TX, in 2024, with further expansion to Jersey City, NJ, in February 2025. The partnership also encompasses autonomous vehicle solutions, which launched commercially on the Uber ride-hailing platform in Dallas in December 2025. Avride has also partnered with Grubhub, deploying its sidewalk robots for last-mile deliveries at the Ohio State University campus. Within the first month of deployment, the number of daily deliveries reached approximately 1,200. In September 2025, Avride also launched campus service at The University of Arizona in Tucson. Avride continues to expand its presence beyond campuses and food delivery to include use cases in supermarkets at H-E-B and, in Japan, the company is providing autonomous retail logistics deliveries for Mitsui Fudosan at the country’s largest outlet mall. Core to Avride’s growth and distribution strategy is expanding the partner network. During 2025, Avride onboarded Shake Shack, Wendy’s, and White Castle, and signed commercial agreements with Uber Eats and Grubhub to utilize those distribution platforms. 40 Table of Contents In March 2025, Avride entered a strategic partnership with Hyundai for the joint development of an autonomous driving platform and the expansion of its fleet. As part of this collaboration, Avride will initially deploy 100 Hyundai Ioniq 5 SUVs retrofitted with autonomous driving technology in the near-term, with plans for further fleet expansion. In October 2025, Avride announced a strategic partnership with Uber. Under the agreement, both Uber and Nebius will make strategic investments and other commitments of up to $375 million in Avride. This investment accelerates Avride’s capacity to advance its autonomous technology, continue building its fleet of vehicles, and expand its coverage map. We are actively exploring further third-party investment into Avride, including transactions in which we may cede control. TripleTen TripleTen is an edtech platform focused on reskilling individuals for careers in technology and driving broader AI education. As of December 31, 2025, the company offered seven immersive program tracks – AI / Machine Learning, AI Automation, Data Analytics, Cybersecurity, Quality Assurance, AI Software Engineering, and UX/UI design – principally in the US and Latin America. In September 2025, the company expanded the offerings of Nebius Academy as a B2B solution that helps companies and large organizations educate their workforces. TripleTen operates on a proprietary tech stack and automated platform that enables scalable course development, localization, and expansion at minimal incremental cost. Material Equity Stakes In addition to our core AI cloud and other businesses, we own equity stakes in both ClickHouse and Toloka, two businesses that were both created and developed by our in-house engineering teams and spun out from the group. We hold a significant minority stake in ClickHouse, an open-source, column-oriented, database management system provider that was spun off from the group in September 2021. We also hold a significant equity stake in Toloka, a leading data provider for LLM and GenAI developers, which was spun off from the group in May 2025. Competition The markets that we target are highly competitive and rapidly changing. Given the current pace of innovation and technological advancement, we anticipate continued high levels of competition in the industry. As a full-stack AI cloud provider, we face competition from cloud computing providers that are scaling AI-specific offerings, such as Amazon (AWS), Google (Google Cloud Platform), Microsoft (Azure) and Oracle. Given the breadth of our services, we also face different competitors across the AI cloud stack. For example, on the compute layer, we see other AI-centric providers offering bare-metal GPU clusters and GPU-centric data center operations. This group of competitors includes CoreWeave, Crusoe, and Lambda Labs, among others. Solutions such as Token Factory may face competition from AI-native inference-as-a-service solutions such as Fireworks AI and Together AI. We believe that our core competitive advantages include: ● Our full-stack, AI-native cloud approach from silicon to software, with offerings spanning the entire AI cloud value chain - from data center compute, hardware to software solutions and production-scale token generation; ● Leading team of AI / ML and cloud engineers focused on developing our growing portfolio of tools and services to optimize and accelerate complex AI workloads at the AI cloud and AI Platform and Applications layers; 41 Table of Contents ● Proven ability to engineer, develop, deploy and scale a wide variety of AI-native software and service technology businesses that customers demand; ● A comprehensive suite of AI services – from data preparation to app deployment and token generation– within a single environment, eliminating the need for multiple vendors and resulting in efficiencies, reduced operational complexity and improved ROI; ● Architecture optimized for large-scale AI deployments, enabling customers to scale their infrastructure quickly and easily up or down as needed, with auto-scaling capabilities that automatically adjusts infrastructure to match workload fluctuations, ensuring AI applications run at optimal efficiency; ● Strong track record of planning, building and operating energy- and resource efficient data centers with heavy power loads and high rack densities in a reliable, scalable manner, leading to high utilization rates; ● In-house hardware design, development and production with lower cost of ownership leading to faster time-to-market; ● Workload orchestration and optimization tools that allow customers to achieve scalable, efficient and sustainable outputs; ● Longstanding partnerships with critical AI hardware providers and leading server original equipment manufacturers (OEMs); ● Our Reference Platform NVIDIA Cloud Partner status, one of only a handful of AI cloud providers to meet these requirements globally, underscoring Nebius’s expertise in designing and deploying a full stack of hardware and software infrastructure that meets NVIDIA’s standards for AI and ML workloads; and ● Flexible access to affordable capital with a strong balance sheet. The other businesses within the group also face their own competitors in the markets in which they compete. For example, Avride competes with other major developers of self-driving technologies, including Waymo, Zoox, and others. TripleTen primarily competes with a number of US-based education technology bootcamp providers. Employees and workforce culture As of December 31, 2025, Nebius had approximately 1,500 employees, the majority of whom are engineers. Talent is the foundation of our business, and we have historically built our products around people. Our full-stack AI cloud offering is a direct result of the expertise in our team across domains including data-center construction and operations, hardware engineering and R&D, cloud solutions development, AI and ML engineering, backed by experienced business development and management professionals. Our HR approach is built on our principles of fairness, transparency and compliance with local labor regulations across all our global locations. Nebius offers competitive compensation packages in line with industry standards. We aim to promote professional growth and high performance, and to support employee wellbeing. We offer flexible working arrangements, including remote working options, as well as mental health support services. Our benefits program is designed to support both short-term and long-term employee needs, including meal and transportation allowances, home-office support for remote employees, and healthcare plans. Commitment to sustainability Sustainability is at the core of our business, enhancing efficiency, reliability and affordability while reducing environmental impact. We integrate responsible practices across our operations and technology stack to align growth with resource management with long-term community benefit. Alongside building AI cloud that delivers maximum performance per watt, we prioritize supporting the communities where we operate through a growing set of initiatives across education partnerships, training programs and access to compute resources for academia and research. Our sustainability efforts focus on three key areas: sustainable computing, empowerment through technology, and reliability and security. 42 Table of Contents Sustainable Computing As AI workloads scale and demand for compute grows, we focus on driving efficiency across the entire technology stack – from hardware and data center systems to software tools for workload orchestration and optimization – to optimize resource use. This enhances infrastructure reliability, results in economic benefits for both our business and our customers, and reduces environmental impact. ● Vertical integration – through vertical integration and full-stack control, we reduce energy consumption per workload, increase compute density per megawatt, lower operating overhead, working to secure performance and reliability across system components rather than addressing efficiency at a single layer. ● Hardware – our in-house servers and racks are designed for thermal efficiency and simplified maintenance. Internal tests show our servers consume approximately 20% less power than equivalent third-party hardware and are fully operational at higher temperatures, enabling efficient cooling configurations and minimizing downtime risks. ● Data-center design – facilities we design and own achieve up to four times lower overheads compared to industry averages, as measured by infrastructure overhead metrics such as PUE, and feature advanced cooling architectures, including closed-loop liquid cooling systems for next-generation GPUs that do not rely on water intake. At our Finnish site, cooling systems integrate with heat recovery that captures server heat and donates it to a municipal heating network. In recent years, this heat has covered up to two-thirds of the municipality’s annual heating demand, contributing to lower heating costs for households and reducing carbon emissions associated with conventional heat production. ● Cloud-embedded capabilities – our cloud platform is designed for efficiency. Flexible GPU allocation, workload batching, auto-scaling, auto-healing, and resource-efficient inference tools help maximize cluster utilization, minimize idle capacity and lower client costs. Empowerment through Technology We leverage our full-stack AI infrastructure and expertise to support innovators, researchers and organizations developing next-generation technologies, while helping individuals build the skills required in an AI-driven economy. Our cloud platform enables advanced AI workloads across high-impact domains such as healthcare, life sciences and robotics, supporting applications in areas including genomics, drug discovery, precision diagnostics and autonomous systems. For example, our infrastructure has been used by companies such as Prima Mente, which trained Pleiades, a large-scale epigenetic foundation model, enabling analysis of DNA methylation patterns for early detection of neurodegenerative diseases such as Alzheimer’s. We also support early-stage innovation through targeted programs. In 2025, we launched the AI Discovery Award, providing GPU credits to startups working on areas such as cancer prediction, protein targeting, transcriptomic mapping and precision diagnostics, with the aim of accelerating high-impact research and fostering collaboration between startups and investors. Through Nebius Academy, our AI education platform for researchers and engineers, we provide training, certifications and academic partnerships designed to equip participants with practical skills in data science, machine learning and generative AI. These programs are complemented by initiatives we are developing to expand access to AI resources, including cloud grants and compute access for academic and research institutions. Our approach also extends to the communities where we operate. Throughout 2025, we have been engaging with local stakeholders to understand community priorities, with the objective of delivering sustained, long-term local benefits alongside infrastructure development. Our community engagement plan includes building local partnerships with educational institutions and training providers, offering AI literacy and workforce development programs, and prioritizing local hiring and skills development where possible. Reliability and Security We are committed to maintaining the highest standards of information security and operational resilience. Our services adhere to stringent security certifications, and we work to establish a comprehensive array of safety controls in our pioneering work with autonomous technologies. More details on our sustainability initiatives can be found at nebius.com/sustainability. 43 Table of Contents History and Development of the Company; Organizational Structure Nebius Group N.V. is a Dutch public company with limited liability. The registered office is at Schiphol Boulevard 165, 1118 BG, Schiphol, The Netherlands. Nebius’s Class A ordinary shares are listed on the Nasdaq Global Select Market under the ticker symbol NBIS. In July 2024, we completed the divestment of all our group’s businesses in Russia and related businesses in certain international markets. The divested businesses constituted more than 95% of the group’s consolidated revenues, assets and employees at that time. Following the divestment, the group continues to be headquartered in Amsterdam, with principal operations in Europe, the US and Israel. Trading in our shares resumed on October 21, 2024, following the completion of the divestment. Nebius Group N.V. is the holding company of the group. Our principal operating subsidiaries are Nebius B.V., Nebius Inc. and EdTech Plus B.V.