Nebius

  Executive Briefing

Nebius is an Amsterdam-headquartered, full-stack GPU cloud and neocloud company publicly traded on Nasdaq (NBIS) that builds, operates, and monetizes large-scale AI infrastructure across Europe and the United States. Formally established in August 2024 from the restructured, non-Russian assets of Yandex N.V., it is led by Arkady Volozh — Yandex's co-founder and former CEO — who returned to the helm after European Union sanctions against him were lifted in March 2024 following his public condemnation of Russia's invasion of Ukraine. John Boynton chairs the board. The company employs roughly 1,300 senior engineers, many drawn from Yandex's two-decade history of building large-scale internet and cloud infrastructure, giving Nebius an unusually deep technical bench for a company of its age.¹

Nebius operates vertically across the AI infrastructure stack. At the compute layer it designs proprietary server racks and in-house cooling systems, runs its own InfiniBand-based networking fabric (called Nebius Fabric), and operates owned and colocated data centers from Finland to New Jersey with a target of 800 MW to 1 GW of fully connected capacity by end of 2026 and 5 GW deployed by 2030 under its NVIDIA strategic partnership.² On top of that hardware base sits Token Factory, a managed inference service covering 60-plus open-source models with an OpenAI-compatible API, and an emerging agentic AI stack built through the acquisitions of Tavily (agentic web-search, announced February 2026, reported at up to $400 million)³ and Eigen AI (kernel-level inference optimization, announced May 2026 at a reported ~$643 million in cash and shares, pending antitrust clearance).⁴

The company's financial profile reflects its rapid ascent: 2025 revenue of $529.8 million was followed by a Q1 2026 quarter of $399 million — a 684 percent year-over-year increase — and a 2026 full-year guidance of $3.0 to $3.4 billion at approximately 40 percent adjusted EBITDA margin.⁵ Two anchor hyperscaler contracts de-risk that trajectory: a reported $17.4 billion five-year agreement with Microsoft (announced September 2025)⁶ and an expanded $27 billion five-year agreement with Meta Platforms (announced March 2026, covering NVIDIA Vera Rubin capacity from 2027).⁷ NVIDIA made a $2 billion strategic equity investment in March 2026 cementing the supply-chain relationship.⁸ As of June 2026 Nebius carried a market capitalization of approximately $59 billion and was scheduled for inclusion in the Nasdaq-100 on June 22, 2026 — a striking trajectory for a company that resumed Nasdaq trading less than two years earlier after a multi-year suspension.⁹

Nebius matters in the AI infrastructure landscape for several compounding reasons. It is the only major neocloud with significant owned data-center infrastructure inside Europe, positioning it to capture sovereign-AI and GDPR-sensitive workloads that US-centric rivals cannot serve on comparable terms. Its NVIDIA backing provides privileged access to next-generation silicon — including early deployments of the Blackwell Ultra, GB300 NVL72, and the forthcoming Vera Rubin platforms — ahead of many competitors. And its vertical integration from rack design through managed inference and agentic search creates a differentiated full-stack offering that goes beyond the raw GPU rental model typical of early neoclouds.

  At a Glance

  Origins & Founding

Nebius's lineage runs directly to Yandex, Russia's dominant internet company, which Arkady Volozh co-founded in 1997. Yandex N.V. was incorporated in the Netherlands in 2007 as the Dutch holding company for the group and went public on Nasdaq in May 2011, raising $1.3 billion in one of the largest tech IPOs of that year.¹ At its late-2021 peak Yandex carried a market capitalization of approximately $31 billion.

The AI cloud division that would become Nebius launched internally in 2018 as Yandex.Cloud, building GPU clusters and data-center infrastructure to serve AI and machine-learning workloads for Russian enterprises and the Yandex product portfolio itself. The division accumulated meaningful GPU capacity and engineering expertise over the following four years, giving it a running start when the corporate picture changed dramatically.

Russia's February 2022 invasion of Ukraine triggered international sanctions and an immediate Nasdaq trading halt on Yandex N.V. shares. Volozh was personally listed on EU sanctions registers. He publicly and unequivocally condemned the invasion, renounced connections to Russian operations, and — after a review process — was removed from the EU sanctions list in March 2024.¹ In parallel, Yandex N.V. negotiated the divestiture of all Russian business assets to a consortium of Russian investors, a transaction that closed in July 2024 for approximately $5.4 billion, providing Nebius with a substantial opening balance sheet.¹

The remaining non-Russian assets — the Finnish data center, the AI cloud unit, autonomous-vehicle startup Avride, edtech company TripleTen, and stakes in data-labeling platform Toloka and database company ClickHouse — were reconstituted as Nebius Group N.V. Volozh was formally named CEO in August 2024, and Nasdaq trading resumed on October 21, 2024 under the ticker NBIS.¹ The founding thesis was explicit: leverage Yandex's engineering talent and existing GPU infrastructure to build a European-anchored, NVIDIA-backed, full-stack AI cloud independent of the incumbent hyperscalers.

  History & Timeline

    2018–2021: Yandex.Cloud and GPU infrastructure genesis

Yandex's cloud division began accumulating GPU capacity and data-center expertise to power internal AI research and external cloud customers. The Finnish data center in Mäntsälä — now Nebius's flagship owned facility — was built and operated during this period. By 2021 Yandex.Cloud had established meaningful infrastructure and a technical team capable of competing with European hyperscalers on AI workloads.

    2022–2023: Sanctions, trading halt, and corporate restructuring

Russia's invasion of Ukraine in February 2022 immediately halted Nasdaq trading in Yandex N.V. shares. The company spent the following two years navigating EU sanctions (including those on Volozh personally), a Nasdaq delisting notice issued in March 2023 (successfully appealed by June 2023), and the complex negotiation of the Russian asset divestiture. Engineering work on the AI cloud continued throughout, insulating the technical roadmap from the corporate turbulence.

    2024: Divestiture, rebrand, and Nasdaq return

The Russian asset sale closed in July 2024. Yandex N.V. was renamed Nebius Group N.V. in August 2024, with Volozh as CEO. Nasdaq trading resumed on October 21, 2024, with an initial market cap in the range of $3.5 to $4.75 billion. On December 2, 2024, Nebius closed a $700 million equity placement — oversubscribed, priced at $21 per share — with NVIDIA (receiving approximately a 0.5 percent stake at that time), Accel Partners, and Orbis Investments as investors.¹¹

    2025: Hyperscaler contracts, AI Cloud 3.1, and infrastructure buildout

The pace accelerated sharply in 2025. In January, AI Studio added FLUX.1 Schnell and FLUX.1 Dev text-to-image models from Black Forest Labs. In June, Nebius raised $1 billion in convertible notes to fund data-center expansion.¹² The transformational event came in September 2025 when a $17.4 billion, five-year capacity agreement with Microsoft was announced, sending the stock up roughly 44 percent in a single session.⁶ The Birmingham, Alabama campus — a 300 MW AI factory on a 75-acre former Regions Bank site — was acquired in September 2025 for approximately $90 million. December 2025 brought the launch of Nebius AI Cloud 3.1 with NVIDIA HGX B300 and GB300 NVL72 systems, 800 Gbps InfiniBand, Capacity Blocks, real-time dashboards, and Slurm/Dstack orchestration — plus the initial deployment of GPU capacity for Meta under an initial $2.9 billion contract.¹³ By end of 2025 Nebius had reached approximately 170 MW of connected capacity and annual recurring revenue of approximately $1 billion.

    2026: NVIDIA partnership, Meta expansion, and agentic stack acquisitions

Q1 2026 saw $2.47 billion in capital expenditure in a single quarter. In February 2026, Nebius announced the acquisition of Tavily — an agentic web-search infrastructure company — for a reported $275 million upfront with earn-outs up to approximately $400 million.³ Birmingham AI factory permits were awarded in April 2026. On March 11, NVIDIA announced a $2 billion strategic equity investment in Nebius alongside a partnership to deploy 5 GW of AI compute by 2030.⁸ Five days later, the Meta deal was expanded to a $27 billion five-year agreement covering NVIDIA Vera Rubin NVL72 deployments beginning from 2027.⁷ In May 2026, Nebius announced the acquisition of Eigen AI for a reported ~$643 million in cash and shares (pending antitrust clearance as of the announcement date) to strengthen Token Factory's inference optimization with kernel-level AWQ 4-bit quantization, MoE, and long-context techniques.⁴ Q1 2026 revenue came in at $399 million (+684 percent YoY), with ARR reaching $1.92 billion — a 54 percent quarter-over-quarter increase. The company was scheduled for Nasdaq-100 index inclusion on June 22, 2026.⁹

  What They Offer — Products and Platform

Nebius's commercial offering is organized around four interconnected product lines spanning raw compute, managed inference, and agentic AI services.

GPU-as-a-Service / AI Cloud is the foundational layer: on-demand and reserved NVIDIA GPU clusters — H200, Blackwell B200, Blackwell Ultra, HGX B300, and GB300 NVL72 systems — available in dedicated and shared capacity tiers for training, fine-tuning, and inference at scale. Nebius AI Cloud 3.1 (December 2025) introduced Capacity Blocks for predictable multi-week reservations, real-time capacity dashboards, and native orchestration via Slurm and Dstack.¹³ An "Aether" serverless compute layer (reported under the AI Cloud 3.5 branding in some financial analyses) is building toward a fully managed, autoscaling compute surface, though the precise product name may vary across marketing materials.

Token Factory (successor to the AI Studio inference service) is Nebius's managed inference-as-a-service product: a developer-facing, OpenAI-compatible API serving 60-plus open-source models with sub-second latency targets, a 99.9 percent SLA, autoscaling to up to 200 billion tokens per day, and speculative decoding. Compliance certifications include SOC 2 Type II, HIPAA, and ISO 27001. Data-residency options span EU regions (Finland, France) and US regions. Enterprise customers access Token Factory through volume-committed contracts; developers access it through a pay-as-you-go API.¹⁴

Tavily (acquired February 2026, pending integration) contributes agentic web-search infrastructure, giving AI agents built on Token Factory real-time factual grounding. Tavily's existing customers include IBM, Cohere, and Groq, providing immediate enterprise distribution for the combined agentic stack.³

Eigen AI (acquisition announced May 2026, pending antitrust clearance) will add kernel-level inference optimization — including AWQ 4-bit quantization, mixture-of-experts (MoE) routing optimization, and long-context throughput improvements — directly into the Token Factory inference layer, with the goal of making Nebius's managed inference the highest-performance option for frontier open-weight models.⁴

Supporting platform services include managed Kubernetes, object storage, VPC networking, and RBAC/SSO enterprise identity controls, rounding out a full cloud platform for AI-native workloads.

  Technology & Infrastructure

Nebius's technology differentiation rests on vertical integration: the company designs its own server racks and in-house liquid cooling systems, operates its own high-speed networking fabric, and builds or co-owns its data-center facilities rather than leasing commodity colo capacity at the margin.

    Hardware

The current GPU fleet is built on NVIDIA silicon: H200 clusters are operational across all major regions; Blackwell B200 and Blackwell Ultra (HGX B300, GB300 NVL72) systems were deployed from late 2025 onward with 800 Gbps InfiniBand interconnects.¹³ Nebius has early access to the upcoming NVIDIA Vera Rubin NVL72 platform via the March 2026 strategic partnership, with the first large-scale Vera Rubin deployments contracted to Meta beginning from 2027.⁷ The NVIDIA partnership also provides early access to Vera CPUs and BlueField storage systems. Eigen AI's kernel optimizations, once integrated, will provide additional per-GPU efficiency gains at the software level.⁴

    Networking and Software Stack

Nebius Fabric is the company's proprietary InfiniBand-based networking software, designed to reduce latency overhead for large distributed training runs — a key differentiator vs. providers that run standard RDMA networking on commodity switches. Slurm-based HPC orchestration and Dstack integration are offered for training workloads. Nebius has published MLPerf Training v5.1 benchmark results validating its cluster performance. At the inference layer, Token Factory uses speculative decoding to maximize throughput, with Eigen AI's quantization and MoE optimizations expected to extend that lead further once the acquisition closes.⁴

    Data Centers

Nebius operates or is constructing a multi-continent footprint:

End-2025 connected capacity: approximately 170 MW. Target: 800 MW to 1 GW fully connected by end of 2026; 5 GW deployed by 2030 under the NVIDIA partnership.²

  Pricing and Performance Position

Nebius competes on price-performance rather than pure price, emphasizing throughput speed and European-region availability as primary differentiators alongside competitive token pricing for open-source models.

Token Factory reported pricing (noted as subject to change): Mistral 7B at $0.08 per million input tokens; Mixtral 8x7B at $0.14 per million input tokens; Llama 2 70B at $0.20 per million input tokens.¹⁴ Volume discounts are reported to escalate from 10 percent at 1 million tokens per month to 50 percent at 1 billion tokens per month, with a Batch API offering a further 20 to 30 percent discount for off-peak submissions.

On throughput, Nebius claims DeepSeek R1 (164k context, MIT License) achieves 248 output tokens per second on its infrastructure — stated as up to 3x faster than major hyperscaler equivalents — and 4.5x faster time-to-first-token in European regions versus competitors. The company positions comparable Llama-405B-quality inference as "more than 2.5x cheaper than GPT-4o." European regions (Finland, France) are reportedly priced approximately 5 percent below US-equivalent instances, an unusual inversion explained by lower European energy costs at owned facilities.¹⁴

GPU compute pricing for H100-class capacity follows standard neocloud market rates. Nebius emphasizes performance-per-dollar enabled by Nebius Fabric's reduced latency overhead on large training runs and by its in-house rack design, which can use higher-density power and cooling configurations than standard colo arrangements permit.

  People and Leadership

Arkady Volozh, CEO and co-founder, is the defining figure. He co-founded Yandex in Russia in 1997 and served as its CEO from approximately 2000 until 2022, building it into Russia's dominant search engine and one of the largest technology companies in Europe by market cap. His 2022 departure was forced by EU sanctions; his public condemnation of the Ukraine invasion and subsequent sanctions removal in March 2024 enabled his return as Nebius CEO in August 2024.¹ Volozh's engineering and commercial credibility — and his 20-plus-year record building large-scale infrastructure — is a material asset for investor and customer confidence in a company that is effectively building a hyperscale cloud from scratch.

John Boynton chairs the board. Roman Chernin (Business Director) and Andrei Korolenko (Product and Infrastructure Director) both carry over from the Yandex leadership structure and are responsible for day-to-day operational execution.¹

On the commercial side, Marc Boroditsky joined as Chief Revenue Officer in May 2025 (previously at Cloudflare), signaling a shift from infrastructure build-out toward enterprise sales. Dan Lawrence joined as SVP of Americas in early 2026 (previously at AWS, exact seniority not independently confirmed), reflecting the company's accelerating US market push. Matt Weigand, a partner at Accel, joined the board in December 2024 in connection with the $700 million equity investment.¹¹

Nebius's approximately 1,300-strong engineering team is a notable asset: many carried over from Yandex's infrastructure divisions and hold deep expertise in large-scale distributed systems, GPU cluster management, and low-level networking — competencies that take years to accumulate and cannot be replicated quickly through hiring alone.

  Funding, Ownership and Business

Nebius's financial foundation traces to three sources: the approximately $5.2 billion in proceeds from the 2024 Yandex Russian asset divestiture (net of obligations), a series of capital markets transactions, and rapidly growing customer revenue.

Capital markets activity since the October 2024 Nasdaq return:

• December 2, 2024 — $700 million oversubscribed equity placement at $21 per share (33.3 million shares); investors: NVIDIA (~0.5% stake at the time), Accel Partners, Orbis Investments.¹¹
• June 2025 — $1 billion convertible notes raised for data-center capex.¹²
• 2025–2026 — Reported ~$4.6 billion in convertible bond issuances across the period (third-party estimates; not broken out in official disclosures).
• March 11, 2026 — NVIDIA $2 billion strategic equity investment; exact equity percentage not publicly disclosed.⁸
• 2026 CapEx guidance — $20 to $25 billion, raised to meet pre-committed customer demand (forward-looking).

The business model is anchored by long-term capacity contracts with hyperscaler anchor tenants: the $17.4 billion Microsoft deal (expandable to $19.4 billion)⁶ and the $27 billion Meta deal⁷ together represent over $46 billion in contracted backlog, providing revenue visibility that supports asset-backed financing for data-center construction. On top of this contracted foundation, Nebius serves a growing base of AI-native startups and enterprise developers through Token Factory's pay-as-you-go and volume-committed pricing, diversifying revenue concentration over time.

Nebius is a publicly traded company with no single majority owner following the Yandex divestiture. Financial results: 2025 revenue $529.8 million; Q1 2026 revenue $399 million (+684% YoY); ARR at Q1 2026 $1.92 billion (+54% QoQ); 2026 full-year guidance $3.0–$3.4 billion at ~40% adjusted EBITDA margin (all forward-looking figures subject to execution risk).⁵

  Customers and Partnerships

Nebius's customer and partnership ecosystem spans anchor hyperscalers, AI-native companies, strategic technology investors, and infrastructure alliances.

Anchor hyperscaler customers: Microsoft committed to a reported $17.4 billion five-year agreement (announced September 2025, expandable to $19.4 billion) for dedicated GPU capacity primarily from the Vineland, New Jersey facility.⁶ Meta Platforms signed an initial $2.9 billion contract in December 2025, subsequently expanded in March 2026 to a $27 billion five-year agreement covering NVIDIA Vera Rubin NVL72 capacity from 2027.⁷

AI-native customers: Cursor (AI code editor) uses Token Factory for inference. Black Forest Labs (Germany, creator of the FLUX image models) has its FLUX.1 Schnell and FLUX.1 Dev models integrated into Token Factory. Mistral AI (France) was an early AI-native customer, fitting Nebius's European positioning. IBM, Cohere, and Groq were customers of Tavily prior to the acquisition and represent early agentic-stack relationships.³

Strategic partners: NVIDIA functions as both financial investor and primary hardware supplier. The March 2026 $2 billion NVIDIA investment is structured alongside a commitment to deploy 5 GW of NVIDIA-powered AI compute by 2030.⁸ DDN and WEKA are strategic storage and data-orchestration alliances supporting training workloads. Hugging Face and Amazon are partners of the Toloka subsidiary (data labeling). Uber is partnered with the Avride subsidiary on food-delivery robots in Austin, Texas, with a planned expansion to autonomous vehicles.

Financial investors: Accel Partners (Matt Weigand on board), Orbis Investments, and NVIDIA are the primary equity investors from the December 2024 round.

  Competitive Position

Nebius competes in the neocloud / GPU-cloud segment, primarily against CoreWeave (the largest rival and also an NVIDIA-backed neocloud), Lambda Labs, Parasail, and — at the managed inference layer — against Together AI, Fireworks AI, and Groq. Hyperscalers (AWS, Azure, Google Cloud) are indirect competitors for enterprise GPU capacity.

Nebius's principal points of differentiation:

European-first footprint. Nebius is the only major neocloud with significant owned data-center infrastructure inside the EU, enabling GDPR-compliant, low-latency, and sovereign-AI positioning that US-centric rivals cannot easily replicate. This is a structural advantage for European enterprise and government customers.¹

Vertical integration. In-house rack design, proprietary Nebius Fabric networking, and a managed software stack reduce training-run latency and total cost of ownership relative to commodity GPU lessors. The Fabric software layer, in particular, addresses the InfiniBand congestion and routing inefficiencies that degrade performance at scale.

NVIDIA strategic investor and early silicon access. The March 2026 $2 billion NVIDIA investment provides privileged access to next-generation platforms — Blackwell Ultra, GB300 NVL72, and Vera Rubin — ahead of most competitors. At the scale Nebius is contracting (5 GW by 2030), supply-chain certainty is a material competitive advantage.⁸

Full-stack agentic ambition. Token Factory + Eigen AI + Tavily builds a differentiated offering beyond raw GPU rental: a managed inference layer with kernel-level optimization and real-time search grounding, aimed at the emerging category of production agentic applications.

Anchor hyperscaler contracts. The $46 billion-plus contracted backlog from Microsoft and Meta de-risks revenue, enables asset-backed data-center financing at lower cost of capital, and validates Nebius as an enterprise-grade counterparty — advantages that smaller neoclouds lack.¹⁰

Principal risks include high revenue concentration in two customers (Microsoft, Meta), massive ongoing capex requirements ($20–25 billion in 2026 alone), NVIDIA chip dependency (single primary hardware supplier), CoreWeave's head start in US market share and customer relationships, and the execution complexity of simultaneously ramping AI factories on three continents.

  Notable Events

Several events define Nebius's trajectory and are worth noting beyond the general timeline.

The 2022–2024 sanctions and trading suspension was existential: a two-year period during which the company could not access public capital markets, its CEO was personally sanctioned, and its ultimate corporate form was uncertain. The successful navigation of that period — culminating in a smooth October 2024 Nasdaq resumption — reflects both the legal/diplomatic work involved and the underlying asset quality that made the structure worth preserving.

The September 2025 Microsoft announcement was the company's inflection point on revenue visibility. The stock's 44 percent single-session gain reflected the market's reassessment of Nebius from a speculative infrastructure build to a contracted revenue machine, and precipitated the subsequent Meta and NVIDIA deals.

The Tavily acquisition (February 2026) marked Nebius's first significant move beyond infrastructure into the AI application and agentic stack — a strategic bet that the highest-margin AI cloud revenues will accrue to providers that offer a complete inference-plus-retrieval-plus-orchestration platform, not just raw compute.

The Eigen AI acquisition (announced May 2026, pending antitrust clearance) at a reported ~$643 million — substantial for an inference-optimization startup — signals how seriously Nebius is pursuing Token Factory as a frontier inference product rather than a commodity API layer.⁴

  Outlook and Roadmap

Nebius's stated 2026 targets are: $3.0 to $3.4 billion in revenue, approximately 40 percent adjusted EBITDA margin, $20 to $25 billion in capital expenditure, and 800 MW to 1 GW of fully connected GPU capacity.⁵ The company is guiding to $7 to $9 billion in ARR by end of 2026, implying a roughly fourfold increase from the $1.92 billion ARR at Q1 2026. All forward-looking figures reflect management guidance and are subject to execution risk.

Strategic priorities for the 2026–2030 period, as publicly signaled:

AI factory scaling. The Lappeenranta, Finland (310 MW), Béthune, France (240 MW), Vineland, New Jersey (300 MW+), and Birmingham, Alabama (300 MW) facilities represent the primary capital deployment priorities. Combined with Iceland and operational clusters, Nebius expects to reach 5 GW of deployed capacity by 2030 under the NVIDIA partnership.²

Vera Rubin platform deployment. The Meta contract is structured around NVIDIA Vera Rubin NVL72 systems, with delivery beginning from early 2027. Successfully deploying the first large-scale Vera Rubin installations will validate Nebius's ability to operationalize each successive NVIDIA generation at hyperscale — a critical execution test.

Agentic AI cloud. Post-integration of Eigen AI and Tavily, Nebius aims to offer a production agentic stack — high-performance kernel-optimized inference, real-time web grounding, and full cloud orchestration — under a single commercial relationship. This positions Token Factory not just as a GPU-API alternative to OpenAI but as an end-to-end agentic platform for enterprises building AI agents.

Sovereign AI expansion. Nebius is actively pursuing European government and regulated-industry customers for whom GDPR compliance, EU data residency, and non-hyperscaler independence are procurement requirements. The Finnish and French owned facilities are structurally suited to this market.

Nasdaq-100 inclusion. The scheduled June 22, 2026 addition to the Nasdaq-100 index is expected to broaden the institutional investor base, increase index-fund demand for NBIS shares, and reduce cost of equity capital — supporting the ongoing capex program.

Non-core asset optionality. Avride (autonomous vehicles), TripleTen (edtech), and the ClickHouse stake (size undisclosed) represent potential monetization opportunities that could redeploy capital into the core AI cloud business if strategic conditions warrant.

---

  References

1. Nebius Group — Wikipedia
2. NVIDIA and Nebius partner to scale full-stack AI cloud — nebius.com
3. Nebius to acquire Tavily — nebius.com
4. Nebius to acquire Eigen AI — nebius.com
5. Q1 2026 earnings coverage — The Motley Fool, Apr 2026
6. Microsoft GPU deal — Data Center Dynamics
7. Meta AI infrastructure agreement — nebius.com
8. NVIDIA strategic investment announcement — nvidianews.nvidia.com
9. Nebius deep-dive — markets.financialcontent.com
10. $46B contracted backlog — The Motley Fool, Apr 2026
11. $700M equity round — nebius.com
12. $1B convertible notes — tech.eu
13. Nebius AI Cloud 3.1 launch — nebius.com
14. Token Factory inference service — nebius.com

  References

1. Nebius Group history, Volozh background, and corporate restructuring — Wikipedia: Nebius Group; TechCrunch, Nov 2024. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸

2. NVIDIA partnership and 5 GW deployment target — nebius.com newsroom; nvidianews.nvidia.com. ↩ ↩² ↩³

3. Tavily acquisition announcement — nebius.com. Acquisition value ($275M upfront, up to $400M with earn-outs) reported by third parties; not officially confirmed by Nebius. ↩ ↩² ↩³ ↩⁴

4. Eigen AI acquisition announcement — nebius.com. Reported ~$643M value in cash and shares; pending antitrust clearance as of May 2026. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

5. Financial guidance and Q1 2026 results — The Motley Fool, Apr 2026. All forward-looking guidance figures subject to execution risk. ↩ ↩² ↩³

6. Microsoft deal — Data Center Dynamics; nebius.com. ↩ ↩² ↩³ ↩⁴

7. Meta five-year agreement — nebius.com. Vera Rubin delivery scheduled from early 2027. ↩ ↩² ↩³ ↩⁴ ↩⁵

8. NVIDIA $2B strategic investment — nvidianews.nvidia.com. Exact equity stake from this investment not publicly disclosed. ↩ ↩² ↩³ ↩⁴ ↩⁵

9. Nasdaq-100 inclusion and market cap — markets.financialcontent.com. Market cap figures fluctuate rapidly; $59B reflects June 12, 2026 reported figure. ↩ ↩² ↩³

10. $46B+ contracted backlog figure — The Motley Fool, Apr 2026. ↩ ↩²

11. $700M equity placement — nebius.com. NVIDIA's ~0.5% stake reflects the December 2024 round; the March 2026 $2B investment is a separate, larger transaction with an undisclosed equity percentage. ↩ ↩² ↩³

12. $1B convertible notes — tech.eu, Jun 2025. ↩ ↩²

13. Nebius AI Cloud 3.1 — nebius.com. ↩ ↩² ↩³

14. Token Factory pricing and performance claims — nebius.com/services/token-factory. Pricing subject to change; performance claims are Nebius-reported and reflect company benchmarks. ↩ ↩² ↩³