CtrlK
Nebius is an AI cloud company incorporated in the Netherlands and listed on the Nasdaq stock exchange under the ticker symbol NBIS. The company provides full-stack GPU infrastructure and managed AI services to AI builders worldwide, including large-scale GPU clusters, cloud storage, developer tooling, and a managed inference platform called Nebius Token Factory. It operates data centers across the United States, Finland, France, the United Kingdom, and Iceland, with facilities under development in several additional locations.
Nebius emerged in mid-2024 from the international assets of Yandex N.V., the Dutch holding company of Russian internet giant Yandex. Following Russia's invasion of Ukraine in 2022, Yandex's Russian operations were sold to a consortium of Russian investors and the remaining non-Russian assets were reorganized as Nebius Group N.V. Arkady Volozh, a co-founder of Yandex, returned to lead the newly reconstituted company as CEO and founder.
Nebius is often grouped with other GPU-focused cloud providers known as neoclouds, alongside CoreWeave, Lambda Labs, and Crusoe. Unlike general-purpose hyperscalers such as Amazon Web Services or Microsoft Azure, Nebius focuses exclusively on AI compute workloads. The company has secured multi-year infrastructure supply agreements with Microsoft and Meta Platforms worth tens of billions of dollars collectively, making it one of the largest contracted AI infrastructure providers in the world as of 2026.
Nebius traces its corporate lineage to Yandex, the Russian technology company co-founded by Arkady Volozh and Ilya Segalovich in 1997. Yandex grew into Russia's dominant search engine and expanded into e-commerce, maps, ride-hailing, autonomous vehicles, music streaming, and cloud services. In May 2011, Yandex N.V., the Dutch holding company, completed an IPO on Nasdaq, raising approximately $1.4 billion in what was then the largest technology IPO since Google's 2004 listing. Volozh served as CEO from 2000 until mid-2022.
Russia's full-scale invasion of Ukraine in February 2022 created an existential crisis for Yandex's international standing. On February 28, 2022, trading in Yandex N.V. shares on Nasdaq was suspended. The European Union sanctioned Volozh in June 2022, citing alleged promotion of pro-government narratives, and he resigned from all positions at Yandex that same month. Volozh later said he delayed his public condemnation of the war until people he felt responsible for were able to leave Russia. In August 2023, he publicly condemned Russia's "barbaric" aggression against Ukraine and described himself as a Kazakhstan-born Israeli tech entrepreneur. He had been living in Israel since 2014.
The EU removed Volozh from its sanctions list in March 2024. Volozh officially renounced his Russian citizenship in February 2026 and holds Israeli citizenship.
Negotiations to separate Yandex's Russian and international operations began shortly after the Ukraine invasion and continued for two years. The complexity of the deal required personal approval from President Vladimir Putin, as the Russian government considered Yandex's search and mapping operations strategically important.
In July 2024, Yandex N.V. sold its Russian assets, including the core search engine, Maps, and the Yandex brand in Russia, to a consortium of Russian investors for approximately $5.2 billion. The sale was described at the time as the largest corporate exit from Russia since the invasion. The Dutch holding company retained the non-Russian operations: the AI cloud business, Toloka (an AI data platform), TripleTen (an edtech company), Avride (an autonomous driving startup), and equity stakes in ClickHouse and other businesses.
The company was rebranded Nebius Group N.V. following the sale. Trading in Nebius Class A ordinary shares resumed on Nasdaq on October 21, 2024, under the ticker NBIS. The company launched with approximately 1,000 former Yandex employees and $2.4 billion in cash on its balance sheet, providing significant runway for capital-intensive data center buildout.
Arkady Volozh serves as CEO and co-founder of Nebius. Born in 1964 in Guryev (now Atyrau), Kazakhstan, into a Russian-Jewish family, he studied applied mathematics at Gubkin Russian State University of Oil and Gas in Moscow and co-developed a Russian-language search engine in 1993 before founding Yandex. He has spoken publicly about the difficulty of navigating Yandex's relationship with the Russian government and stated that his decision to build Nebius outside Russia was driven by a desire to operate without political constraints. John Boynton serves as Chairman of the Board.
In December 2024, Nebius raised $700 million in an oversubscribed private placement from institutional investors. The round was led by Accel and included participation from NVIDIA and accounts managed by Orbis Investments. The company issued 33,333,334 Class A shares at $21.00 per share, representing roughly a 3% premium to the volume-weighted average price since trading had resumed. NVIDIA's stake in this round was small, worth approximately $33 million, giving NVIDIA an equity interest of about 0.5%.
Nebius said proceeds would fund data center deployments in the United States and other markets, including greenfield sites and colocation expansions. Matt Weigand from Accel received observer rights and was subsequently nominated as a director at the 2025 Annual General Meeting.
In June 2025, Nebius raised $1 billion through a private placement of senior unsecured convertible notes in two equal tranches: $500 million of notes due 2029 carrying a 2.00% coupon and $500 million of notes due 2031 carrying a 3.00% coupon. Goldman Sachs Bank Europe SE acted as sole placement agent. The proceeds were directed toward continued data center construction and GPU procurement.
Days after announcing its Microsoft deal in September 2025, Nebius launched an additional capital raise that closed with approximately $4.2 billion in combined follow-on equity and convertible debt. The raise was designed to fund the infrastructure commitments implied by its expanding customer contracts.
In March 2026, NVIDIA announced a $2 billion direct equity investment in Nebius via pre-funded warrants exercisable into 21,065,936 Class A ordinary shares at an exercise price of $0.0001 per share. The partnership was framed as a strategic move to target 5 gigawatts of AI infrastructure capacity by 2030. NVIDIA designated Nebius as a preferred partner for next-generation hyperscale AI infrastructure. Nebius shares rose approximately 16% on the announcement.
Nebius AI Cloud is the company's core GPU compute platform. It is designed for intensive AI training and inference workloads and uses a combination of proprietary software and hardware optimized for AI. The platform supports NVIDIA GPU generations including H100, H200, B200, B300, GB200 NVL72, and GB300 NVL72. Compute can be accessed as on-demand instances or reserved capacity in large clusters.
The platform supports both managed Kubernetes and Slurm for job orchestration. It includes InfiniBand interconnects for high-bandwidth intra-cluster communication, topology-aware job scheduling, and fast NVMe-backed storage. Nebius built lightweight virtualization layers that independent benchmarks have found deliver near-bare-metal performance, placing it in the top tier of neocloud providers in at least one major GPU benchmarking study.
Nebius AI Cloud offers on-demand GPU access and dedicated reserved clusters. Pricing as of 2025 included H100 instances at approximately $2.95 per GPU-hour, H200 at approximately $3.50 per GPU-hour, and L40S at approximately $1.55 per GPU-hour. The platform has availability zones in the United States and Europe.
Nebius Token Factory, formerly called Nebius AI Studio, is a managed inference platform for deploying open-source language models at production scale. It launched as AI Studio in late 2024 and was relaunched as Token Factory in October 2025.
Token Factory supports over 60 open-source models, including Llama, Qwen, DeepSeek, and Mistral variants. It offers dedicated model endpoints with autoscaling, speculative decoding, and multi-region routing. Customers can fine-tune models using LoRA or full fine-tuning, then deploy the fine-tuned version on a dedicated endpoint with one-click tooling. The platform carries SOC 2 Type II, HIPAA, ISO 27001, and ISO 27799 certifications and offers zero-retention inference in either EU or US data centers to satisfy data-residency requirements.
In May 2026, Nebius agreed to acquire Eigen AI, a 20-person startup founded by alumni of MIT's HAN Lab, for approximately $643 million in a cash-and-stock deal. Eigen AI specializes in inference optimization and post-training software that determines how quickly and cost-efficiently a trained model generates responses in production. Nebius said it would integrate Eigen's technology directly into Token Factory. Joint implementations of several open-source models by Nebius and Eigen AI had already ranked among the fastest on the Artificial Analysis benchmark at the time of the announcement.
Toloka is an AI data platform that Nebius spun out as a separate company. Founded in 2014 by Olga Megorskaya as an internal Yandex crowdsourcing project for data annotation, Toloka evolved over the following decade into a provider of training data, evaluation pipelines, and human-in-the-loop workflows for large language models and generative AI systems.
Toloka's offerings include reinforcement learning gym environments for model training, data annotation and labeling services, and model evaluation systems. Its customers include Amazon, Microsoft, Anthropic, Shopify, and Poolside.
In May 2025, Jeff Bezos' investment firm Bezos Expeditions led a $72 million funding round into Toloka. Mikhail Parakhin, CTO of Shopify, also participated. Following the close of that transaction, Nebius ceased to hold majority voting power in Toloka and began reporting its stake as an equity method investment rather than a consolidated subsidiary.
In February 2026, Nebius agreed to acquire Tavily, an AI agent search company, for $275 million (rising to $400 million if milestones are achieved). Tavily was founded in late 2024 by Rotem Weiss, Assaf Elovic, and Yuval Rozio, and provides search infrastructure that lets AI agents query the web in real time. Its customers include IBM and AI companies such as Cohere and Groq. Nebius positioned the acquisition as an expansion from raw GPU infrastructure into a platform where enterprises can build autonomous agents with integrated compute and search capabilities. The Tavily team continues to operate under the Tavily brand following acquisition close.
TripleTen is an edtech platform that offers online courses in software development, data science, and business analytics. It targets career changers seeking to enter the technology industry and operates with a bootcamp-style format. TripleTen works with over 40 employer partners to provide job placement services to graduates. It is a consolidated subsidiary of Nebius Group.
Avride is an autonomous vehicle and delivery robot developer that emerged from Yandex's self-driving car program. It develops technology for ride-hailing, last-mile logistics, and food delivery. Avride has a partnership with Uber and planned to launch a robotaxi service through that partnership. It operates as a consolidated subsidiary of Nebius Group.
Nebius holds a 28% stake in ClickHouse, an open-source columnar database company spun out of Yandex. ClickHouse has been valued at approximately $6 billion. In May 2025, Nebius participated in ClickHouse's Series C funding round. Nebius reports ClickHouse as an equity method investment.
On September 8, 2025, Nebius announced a multi-year infrastructure supply agreement with Microsoft. Bloomberg reported the contract value at up to $19.4 billion over five years, with a base commitment of $17.4 billion and an option for Microsoft to purchase an additional $2 billion. Under the agreement, Nebius delivers dedicated GPU-powered compute capacity to Microsoft for AI workloads. Deliveries began in late 2025 from Nebius's new data center in Vineland, New Jersey.
CEO Arkady Volozh described the Microsoft deal as "the first of these contracts" and indicated at the time of signing that more agreements with major technology companies were in progress. Nebius said it planned to finance the project through deal cash flows, debt secured against the contract, and other financing mechanisms. Nebius shares surged approximately 60% in after-hours trading on the day of the announcement.
Microsoft's decision to source capacity from Nebius reflected a broader pattern among hyperscalers of contracting with specialist GPU cloud providers to supplement their own data center capacity during a period of intense AI infrastructure demand.
In November 2025, Nebius signed a five-year infrastructure contract with Meta Platforms valued at $3 billion. The deal was limited to the capacity Nebius had available at the time of signing; the company said demand was overwhelming and it had to cap the agreement size accordingly.
In March 2026, Meta expanded its relationship with Nebius to a deal worth up to $27 billion. The expanded agreement includes $12 billion of dedicated GPU capacity and up to $15 billion of additional available compute across a number of locations over a five-year period. Part of the agreement covers one of the first large-scale deployments of NVIDIA's Vera Rubin GPU architecture. New capacity deliveries under the expanded Meta deal are scheduled to begin from early 2027 as Nebius brings additional data centers online.
Meta shares surged approximately 15% on the announcement of the expanded deal. Nebius's total contracted backlog across all agreements had reached nearly $50 billion by early 2026.
Nebius operates and builds data centers across Europe and the United States, with additional facilities in Iceland and Israel.
Nebius's first and largest European data center is located in Mäntsälä, Finland, approximately 60 kilometers north of Helsinki. The facility expanded to 75 MW of capacity in early 2026 and houses what Nebius describes as Europe's first operational deployment of the NVIDIA GB300 NVL72 platform.
In addition to the Mäntsälä facility, Nebius announced construction of a second Finnish data center in Lappeenranta, near the Russian border, with planned capacity of up to 310 MW. The Lappeenranta AI factory is expected to deliver its first capacity to customers in 2027 and would be one of Europe's largest dedicated AI data centers when fully operational.
Finland offers several advantages for AI infrastructure: abundant hydroelectric power, cool ambient temperatures that reduce cooling costs, and geopolitical stability as a NATO member state. Nebius has committed more than $1 billion to AI infrastructure investment in Europe by mid-2025.
Nebius has expanded aggressively in the United States since 2024. It opened its first US availability zone and has facilities in Kansas City, Missouri and Vineland, New Jersey.
The Vineland, New Jersey facility is the site contracted under the Microsoft agreement and has planned capacity of up to 300 MW across multiple phases. The first capacity phase was expected by summer 2025, with the full 300 MW buildout extending through 2026.
Nebius also announced an AI factory in Birmingham, Alabama (designated BHM01), designed specifically for large-scale AI training workloads. The company has targets of 800 megawatts to 1 gigawatt of fully connected capacity by end of 2026 and a longer-term goal of 2.5 gigawatts of contracted power.
Nebius operates a colocation deployment at Equinix's PA10 campus in the Saint-Denis district of Paris. The Paris cluster was among the first deployments worldwide to use NVIDIA H200 GPUs when they became available.
Nebius launched a UK availability zone in late 2025 with an initial deployment of 4,000 NVIDIA Blackwell Ultra GPUs at a UK colocation facility. Customers at UK launch include Basecamp Research, which uses compute for life sciences applications.
Nebius has a 10 MW compute cluster in Keflavik, Iceland, powered by Iceland's geothermal and hydroelectric grid. The company positions the Keflavik deployment for customers with strict sustainability requirements or carbon-neutral compute targets.
Nebius has R&D operations in Israel. In a separate contract announced around the same time as its broader expansion, Nebius agreed to build and operate a $140 million national supercomputer in Israel equipped with 4,000 NVIDIA Blackwell GPUs, partially funded by the Israeli government's Innovation Authority.
Nebius sources GPU hardware exclusively from NVIDIA, in line with its positioning as an NVIDIA-partnered AI cloud. Its deployed and ordered fleet includes NVIDIA H100 SXM, H200 SXM, B200, B300, GB200 NVL72, and GB300 NVL72 systems.
The GB300 NVL72 systems at the Mäntsälä facility represent what Nebius claims is the first deployment of that architecture in Europe. Nebius was also among the first cloud providers to accept pre-orders for GB200 NVL72 and HGX B200 cluster configurations.
Nebius uses InfiniBand networking for intra-cluster communications in large cluster configurations, which is required for efficient distributed training workloads. Customers can provision clusters ranging from a few GPUs up to thousands within a single fabric.
On storage, Nebius partners with WEKA, providing high-throughput, low-latency storage that can keep GPU utilization high during training runs. The company uses lightweight virtualization rather than full virtual machine overhead, which independent benchmarks have found places Nebius in the same performance tier as bare-metal providers.
Nebius Group N.V. has been publicly traded since May 2011, initially as Yandex N.V. on Nasdaq under the ticker YNDX. Trading was suspended on February 28, 2022, following Russia's invasion of Ukraine.
After the sale of the Russian assets closed in July 2024, the company rebranded as Nebius Group N.V. and resumed trading on Nasdaq on October 21, 2024, under the new ticker symbol NBIS. The resumption of trading was not a traditional IPO but a continuation of the existing listed entity under its new name and strategy.
Nebius shares opened at approximately $10 in October 2024 and rose sharply following the Microsoft deal announcement in September 2025, the Meta deal expansions, and the NVIDIA investment announcement. Over the 12-month period through approximately May 2026, NBIS shares had appreciated roughly 737%. The company's market capitalization reached tens of billions of dollars, with the stock trading at elevated price-to-sales multiples relative to sector norms, reflecting expectations for rapid revenue growth on contracted backlog.
As of December 31, 2025, Nebius had approximately 1,371 employees.
Nebius's core AI cloud infrastructure business grew revenue by more than 600% in 2024. The company reported full-year 2024 revenue from continuing operations with a net loss of $396.9 million and an adjusted EBITDA loss of $266.4 million. Cash and equivalents at year-end 2024 were $2,449.6 million.
For full-year 2025, Nebius reported group revenue of $529.8 million with Q4 2025 revenue of $227.7 million, up 547% year-over-year and 56% sequentially. The company exited 2025 with an annualized run-rate revenue (ARR) of $1.25 billion, above its original guidance range of $750 million to $1 billion. Total assets reached $12.4 billion by end of 2025, driven by large-scale capital expenditure on data center buildout. Cash at year-end 2025 was approximately $3.7 billion.
Nebius guided for 2026 revenue of $3 billion to $3.4 billion and adjusted EBITDA margins approaching 40%, with a targeted year-end 2026 ARR of $7 billion to $9 billion. Capital expenditure for 2025 reached $2 billion per quarter, reflecting the pace of data center construction. The company remained unprofitable on a net income basis, reporting a net loss of approximately $518 million for 2025.
Nebius competes with a group of GPU-focused cloud providers that are collectively referred to as neoclouds. These companies position themselves against general-purpose hyperscalers by offering lower prices, more immediate GPU availability, and closer technical alignment with AI workloads.
| Provider | Founded | GPU focus | Notable customers | Funding / listing |
|---|---|---|---|---|
| CoreWeave | 2017 | NVIDIA-only; bare-metal Kubernetes | Microsoft, OpenAI | IPO March 2025; $23B+ raised |
| Nebius | 2024 (as Nebius) | NVIDIA-only; full-stack AI cloud | Microsoft, Meta | Nasdaq-listed (NBIS); $700M raise + $2B NVIDIA investment |
| Lambda Labs | 2012 | NVIDIA-primary; 1-Click Clusters | Various AI startups | Private; NVIDIA investor |
| Crusoe | 2018 | NVIDIA; sustainability focus | Various enterprises | Private; $4B+ raised |
CoreWeave is the largest neocloud by revenue and was the first to complete a public listing via an IPO in March 2025, reporting nearly $1 billion in quarterly revenue and projecting a $5 billion annual run-rate. CoreWeave has a large existing contract base with Microsoft. Nebius differentiates from CoreWeave partly through its integrated software stack, including the managed inference platform and the Toloka data services ecosystem, and partly through its European data center footprint, which is more developed than CoreWeave's.
Lambda Labs competes primarily on ease of use and rapid cluster provisioning, with its 1-Click Cluster feature that lets customers launch multi-GPU configurations in minutes. Lambda tends to attract AI research teams and startups that want low friction and competitive hourly pricing. NVIDIA also holds a stake in Lambda.
Crusoe built its business around sustainable and low-cost energy, originally using stranded natural gas to power GPU clusters and then expanding to utility-scale renewable deployments. Crusoe secured agreements for 4.5 gigawatts of natural gas capacity to power AI data centers. Nebius's Iceland and Finland facilities target a similar green compute segment, though Nebius's overall strategy is more geographically and contractually diversified.
Nebius's AI cloud customer base includes AI-native startups, research organizations, and enterprises. Named publicly by Nebius include:
Brave Search, which runs large language models on Nebius infrastructure to deliver AI-generated summaries for over 11 million queries per day. The company cites near-100% compute utilization on Nebius GPU clusters.
Basecamp Research, a life sciences AI company, was among the first customers to deploy on Nebius's UK cluster.
CentML, an open-source model inference company, uses Nebius compute as part of its automated compute optimization service.
TheStage AI, which builds inference simulators and deep neural network optimization tools, uses Nebius for compute.
Microsoft and Meta Platforms are Nebius's largest customers by contract value, with multi-year dedicated capacity agreements described above.
Toloka's customers, which include Anthropic, Amazon, Shopify, and Microsoft, sit within the broader Nebius ecosystem even though Toloka operates as a separate entity following Bezos Expeditions' investment.
Nebius's re-emergence as an AI infrastructure company attracted significant attention from technology investors and analysts. The TechCrunch headline from November 2024 called Nebius "the publicly traded AI infrastructure startup," noting the unusual situation of a company that had been listed for 13 years under a prior name effectively relaunching as a venture-stage business.
The Microsoft deal in September 2025 was covered extensively by Bloomberg, CNBC, the Wall Street Journal, and Reuters, with most coverage noting the scale of the contract relative to Nebius's then-current revenue and the implicit validation it provided for the neocloud model.
Analysts at Seeking Alpha, Motley Fool, and institutional research firms noted that the contracted backlog approaching $50 billion compared favorably with Nebius's 2025 revenue of $530 million, implying years of high growth if contracts are fully executed. However, several analyst notes raised concerns about valuation: NBIS traded at forward price-to-sales multiples more than 900% above the sector median as of early 2026, reflecting expectations that may leave little margin for execution shortfalls.
Volozh himself received significant press attention, including a Bloomberg feature in January 2025 and coverage in the Moscow Times in December 2025, covering his transformation from a Russian internet oligarch into the CEO of a Western-listed AI infrastructure company.
Nebius carries several risks that analysts and its own SEC filings have identified.
The company's origins create what some institutional investors describe as perception risk. Although Volozh was removed from EU sanctions in March 2024, renounced Russian citizenship in February 2026, and the company's Russian assets have been fully divested, some ESG-focused funds and institutional allocators remain cautious. This perception may cap the investor base relative to competitors with less complicated histories.
Nebius is deeply capital-intensive. Quarterly capital expenditure reached $2 billion in late 2025, and the company has guided for continued heavy spending through 2026. The company was unprofitable through 2025, with a net loss of approximately $518 million that year. Although management has guided for positive EBITDA margins in 2026, the company must execute a complex multi-geography data center buildout on accelerated timelines to meet contract delivery dates.
Contract concentration is another concern. The Microsoft and Meta deals, which together account for the majority of Nebius's contracted backlog, create significant customer concentration. Any renegotiation, cancellation, or scope reduction would materially affect revenue projections.
GPU availability risk is inherent to the neocloud model. Nebius relies entirely on NVIDIA for its compute hardware, and supply constraints or pricing changes from NVIDIA would affect all neocloud providers. The NVIDIA strategic investment partially addresses this by aligning incentives, but it does not guarantee supply priority.
Data center execution risk is significant given the pace of buildout. Construction delays, permitting issues, utility interconnection timelines, and equipment lead times have all affected GPU cloud providers industry-wide. Nebius's revenue growth depends on bringing new capacity online within the timelines implied by its customer contracts.