Introduction
Artificial intelligence has transitioned from a niche academic pursuit to a foundational technology reshaping diverse sectors such as healthcare, finance, defence and public services. This transformation has increased demand for high-performance computing (HPC) and specialised AI infrastructure capable of supporting the training and deployment of increasingly large and complex models. Within this context, Nebius AI has positioned itself as an influential provider of full-stack AI infrastructure.
This paper addresses several key questions: What is the historical genesis of Nebius? How has the company developed its technology and organisational strategy? What are the distinguishing features of Nebius’s infrastructure offerings? What partnerships and deployments have defined its global reach? Finally, what broader implications does Nebius’s evolution have for AI infrastructure provision and research?
Origins and Corporate Formation
Nebius AI has its roots in the broader restructuring of technology assets historically associated with Yandex, a major Russian technology group. In the complex geopolitical and economic context following Russia’s 2022 invasion of Ukraine, Yandex divested key assets, leading to the formation of a reconstituted entity that later adopted the name Nebius. This realignment facilitated the company’s strategic focus on AI infrastructure and cloud services, marking a departure from its previous portfolio to a specialised infrastructure-centric business model.
Founded under the leadership of Arkady Volozh, a veteran technology entrepreneur, Nebius Group N.V. was publicly launched with an explicit ambition to build a leading European AI infrastructure provider. The company’s early strategic communications emphasised the need to bridge the emerging AI infrastructure deficit - a critical challenge as industry demand for AI compute began to outstrip available supply. Nebius articulated a vision of investing in proprietary technology, data centre capacity and full-stack offerings capable of servicing the needs of AI development at scale.
Nebius was also characterised early on by an extensive investment in research and development, boasting an engineering backbone of hundreds of professionals focused on cloud, software, hardware and machine learning disciplines. This foundational investment was intended to differentiate Nebius from traditional cloud providers, many of which repurposed general-purpose infrastructure rather than building AI-native systems.
Financing and Strategic Expansion
The expansion of Nebius has been fuelled by substantial strategic financing. In late 2024 Nebius secured an oversubscribed equity financing round of USD 700 million, with backing from institutional investors including Accel and NVIDIA. This capital was earmarked for accelerating the rollout of full-stack AI infrastructure, signalling investor confidence in the company’s proposition at a time when global demand for AI compute was rapidly rising.
This financing enabled Nebius to pursue multiple strategic objectives: build-to-suit data centres, expand into key regional markets, deploy advanced GPU clusters and support workflow services for training, fine-tuning and inference of large models. The scale of financial commitment also reflected a broader recognition that AI infrastructure would be a strategic bottleneck unless addressed with dedicated investment. The liquidity provided by this round positioned Nebius to compete in an environment increasingly dominated by hyper-scale cloud providers like Amazon Web Services (AWS), Microsoft Azure and Google Cloud.
AI-Native Cloud Architecture
A defining element of Nebius’s offering is its AI-native cloud platform. Unlike traditional cloud services, which were originally optimised for generalised workloads (web hosting, database operations, data storage), Nebius’s platform was designed from the ground up for AI workloads. This includes support for distributed training, fine-tuning and inference, integrated into a seamless user experience with self-service access to GPU-accelerated compute clusters.
Nebius’s infrastructure integrates NVIDIA accelerated computing systems - notably GPUs such as Nvidia H100 and H200 Tensor Core models, as well as the latest Blackwell Ultra series - along with high-performance networking (InfiniBand) architectures that reduce communication latencies between nodes and optimise large-scale parallel processing.
The platform emphasises bare-metal performance, meaning that hardware is exposed directly to workloads to minimise virtualisation overhead. This approach maximises Model FLOPS Utilisation (MFU) and ensures that computational resources are used with high efficiency, a critical requirement for training models with billions or trillions of parameters.
Software Stack and Workflow Optimisation
Beyond hardware, Nebius has invested significantly in its own cloud software architecture, including tooling for environment management, job monitoring, automated fault recovery and optimisation of distributed workloads. Virtual machines come pre-configured with essential AI libraries and drivers, reducing setup time for developers and researchers.
By designing both hardware and software in-house, Nebius seeks to offer a unified system that can be fine-tuned to meet the specific needs of AI practitioners. This contrasts with repurposing generic cloud infrastructure and highlights the company’s focus on reducing operational friction for AI workflows.
Global Deployments and Regional Expansion
Nebius has pursued an aggressive global build-out strategy, establishing AI infrastructure deployments across Europe, North America and the Middle East. Notable examples include GPU cluster rollouts in Paris as part of a broader $1 billion European investment and the expansion of AI infrastructure in Israel featuring one of the country’s first NVIDIA Blackwell GPU deployments.
In 2025 Nebius formally announced deployment of advanced AI infrastructure in the United Kingdom, including state-of-the-art facilities with Nvidia Blackwell Ultra GPUs coupled with NVIDIA Quantum-X800 InfiniBand networking. This infrastructure was described as among the UK’s most advanced AI supercomputing platforms, enhancing domestic compute capacity for local businesses, researchers and public services.
To comply with stringent enterprise requirements, the Nebius AI Cloud 3.0 “Aether” platform delivered enterprise-grade security features such as SOC2 Type II certification, GDPR compliance and end-to-end encryption, features increasingly essential for mission-critical AI deployments in regulated sectors such as healthcare and finance.
Nebius has also pursued expansion in the United States, including plans to build an AI data centre in Vineland New Jersey, which was tied to multibillion-dollar agreements with major technology firms. These deployments signified an important strategic move into North America, intending to supply GPU-based compute capacity at scale to hyperscalers and large enterprises.
Strategic Commercial Partnerships
A key indicator of Nebius’s industry position has been its strategic commercial agreements with major technology companies. Perhaps most prominent was the reported multi-billion-dollar agreement with Microsoft to supply dedicated GPU infrastructure over a multi-year period. Estimates of the contract’s value ranged from approximately USD 17.4 billion to potentially nearly USD 19.4 billion through 2031, reflecting long-term commitments to support large-scale AI workloads.
Similarly, Nebius secured a reported USD 3 billion, five-year agreement with Meta to provide AI infrastructure, demonstrating trust from major hyper-scale AI developers in Nebius’s infrastructure capabilities. These partnerships underscored the broader industry trend of outsourcing specialised compute capacity to dedicated infrastructure providers in response to surging demand and capacity constraints faced by traditional cloud providers.
These agreements did not merely serve commercial ends: they also signalled Nebius’s role as a critical node in the global AI computing supply chain. By committing capacity to leading AI consumers, Nebius helped to alleviate bottlenecks in AI training and inference workloads that would otherwise strain existing public cloud services.
Position in the Evolving AI Infrastructure Landscape
The emergence of “neocloud” providers such as Nebius reflects a broader evolution in the AI infrastructure landscape. Traditional cloud giants have historically dominated compute provisioning, but the exponential growth in demand for GPU-accelerated workloads has outpaced their capacity. This has opened space for specialised firms that combine engineered hardware systems with AI-optimised platforms. Nebius’s approach illustrates several key trends:
1. Specialisation over Generalisation: By focusing on AI workloads exclusively, Nebius can tailor both hardware and software layers, from GPU selection to orchestration frameworks, achieving performance and usability benefits over general cloud offerings.
2. Partnership-Driven Growth: Nebius’s strategic contracts with major technology companies reflect a collaborative model in which specialised infrastructure suppliers augment rather than replace the capacity of hyper-scale providers.
3. Regional Sovereignty of Compute: Deployments in regions such as the UK and Israel have policy implications, aligning with government initiatives to build domestic AI capacity and reduce dependence on foreign cloud infrastructure.
4. Operational Efficiency: Bare-metal performance, pre-configured environments and integrated lifecycle tooling reduce barriers to entry for researchers and enterprises, democratising access to high-performance compute.
Challenges and Constraints
While Nebius’s growth has been rapid, it has not been without challenges. Reports indicate that high capital expenditures and ongoing operating losses have accompanied expansion efforts, raising questions about long-term sustainability and market dynamics. The intense competition for GPU resources, rising energy costs and financing requirements for new data centres constitute significant operational risks.
Furthermore, as the broader AI industry evolves, alternative architectures, including custom accelerators and edge AI deployments, may require further innovation beyond GPU-centric models. Nebius, alongside other infrastructure firms, will need to adapt to heterogeneous computing environments and evolving model architectures.
Conclusion
Nebius AI’s trajectory illustrates the dynamic evolution of AI infrastructure in the early-to-mid 2020s. From its origins as a reconstituted technology entity, the company has grown into a significant provider of AI-optimised infrastructure, distinguished by its technological focus, strategic financing, global deployments and high-profile commercial agreements with industry leaders. Its role in expanding compute capacity, supporting sovereign infrastructure initiatives and enabling next-generation AI workloads exemplifies the critical importance of specialised infrastructure in an era defined by rapid increases in model complexity and data intensity.
The Nebius case provides valuable insights into the interplay between technology innovation, capital markets and the structural demands of AI workflows. As the industry continues to evolve, further research is warranted into the economic models, technological architectures and policy frameworks governing AI infrastructure providers. Nebius’s growth and strategic choices will likely remain of interest to scholars, industry practitioners and policymakers seeking to understand and shape the future of AI computing ecosystems.