Why Nebius Group (NBIS) Is My Largest Position: A Deep Dive

Nebius Group (NBIS) is my largest position — and it’s finally starting to catch the attention of retail investors.

I’ve been talking about it for months, so it's time to condense everything in a detailed article.

Here’s why I believe NBIS is one of the best opportunities in the market.

Origins: From Yandex to Nebius Group

The story of Nebius Group begins inside one of the most iconic tech companies to emerge from Eastern Europe: Yandex. Often dubbed the “Google of Russia”, Yandex was a digital powerhouse, dominating search, maps, ride-hailing, e-commerce, and AI in Russia and surrounding markets. At its peak, it was a $30B company and one of the most successful tech stories in the region.

Then, everything changed.

In early 2022, Russia’s invasion of Ukraine set off a geopolitical and moral reckoning for thousands of Yandex employees. Around 2,000 engineers, product leaders, and researchers — along with key members of Yandex’s founding team — made a bold decision: they would not be complicit. They chose to walk away from the Russian business, even if it meant leaving behind their homes, careers, and in some cases, their families.

This moral stand set in motion one of the most complex corporate restructurings in recent memory. Over the following two years, Yandex navigated sanctions, shareholder pressure, and mounting political scrutiny to ultimately divest all of its Russia-based assets. By mid-2024, a formal separation was completed.

And that’s how NBIS was born.

Today, NBIS is a completely independent company — composed of Nebius (AI cloud), Avride (autonomous driving), Toloka (data labeling), TripleTen (edtech), and a 28% stake in ClickHouse, the fast-growing open-source database platform. The group is legally and operationally severed from Yandex’s Russian operations. Its leadership and board have acquired Dutch or Israeli citizenship, ensuring full compliance with international sanctions and signaling a clean break from its origins.

While NBIS inherits Yandex’s deep engineering DNA — particularly in AI infrastructure, distributed systems, and cloud computing — it is now pursuing a fundamentally different mission: to become a global leader in AI cloud services.

But with AI dominating headlines and investor attention, how is NBIS still flying under the radar?

The answer lies in its unconventional path to the public markets. Due to its roots within Yandex, NBIS bypassed the traditional IPO process entirely. There was no roadshow, no investor marketing, and virtually no institutional coverage.

In fact, according to the company’s Founder and CEO, they were caught off guard by the listing. On a Friday, they received a call from Nasdaq informing them that Yandex’s legacy listing would transition to Nebius on Monday — just three days later. The team had to scramble to meet compliance requirements and prepare investor materials with almost no advance notice.

As a result, the stock debuted with minimal visibility. Analyst coverage is still limited, and a large portion of the float remains in the hands of retail investors.

This disconnect — between the quality of the business and its lack of market visibility — is precisely what makes NBIS one of the most compelling and asymmetric opportunities in tech today.

As the company continues to execute on its ambitious expansion plans, it's likely only a matter of time before the broader market takes notice.

Core Business Explained

The AI revolution is accelerating, pushing the boundaries of what’s technologically possible — but also exposing the severe limitations of today’s compute infrastructure. As demand for AI capabilities explodes, the need for purpose-built, scalable, and efficient compute infrastructure has become one of the most urgent bottlenecks in the tech industry. Nebius Group exists to help solve that.

Positioned at the cutting edge of the global AI infrastructure market, Nebius is building the backbone of tomorrow’s AI economy. Its mission is to deliver the infrastructure, tools, and services required to support AI innovation at scale. With ambitions to scale its operations to thousands of megawatts of GPU compute capacity, Nebius is enabling startups, enterprises, and researchers alike to build, train, and deploy cutting-edge AI models — all on a single integrated platform.

At its core, Nebius is a next-generation AI infrastructure company, often referred to as a “neocloud.” Unlike traditional cloud providers that retrofitted their platforms for AI, Nebius was purpose-built from the ground up for AI workloads. It combines deep expertise in hardware and software development with large-scale GPU deployments to deliver a full-stack solution designed specifically for the demands of modern AI development.

A Three-Layered Architecture: Infrastructure, Platform, and Applications

Nebius operates across three primary layers — infrastructure, platform, and applications — creating an end-to-end ecosystem that addresses every step of the AI development lifecycle.

1. Comprehensive AI Infrastructure

Unlike many cloud providers that rely on off-the-shelf hardware and outsourced services, Nebius controls the entire value chain of its infrastructure. This full-stack control translates into both performance gains and economic efficiency:

• Custom Data Centers: Engineered for energy efficiency and high-density compute, Nebius’ data centers allow for better unit economics and scalability.

• In-House Server Design: Servers are custom-built to optimize GPU utilization and deployment speed — beyond the GPU itself — giving Nebius a competitive edge in cost and performance.

• End-to-End Stack: From hardware manufacturing to cloud orchestration, Nebius owns every layer, enabling tight integration, faster innovation cycles, and better cost controls.

• Managed Services: Tools like Apache Spark, MLflow, and others are seamlessly integrated, letting users focus on development rather than infrastructure management.

2. AI-Centric Cloud Platform

At the platform level, Nebius has developed an AI-native cloud computing environment tailored to the needs of ML and DL practitioners. The platform integrates large-scale GPU clusters, scalable object storage, and managed tools into a cohesive offering:

• AI-Optimized Compute: Support for training, fine-tuning, and inference on some of the most advanced GPUs available, including NVIDIA H100s, H200s, and the upcoming Blackwells.

• Elastic Scalability: Whether it’s a single experiment or a massive training run, users can scale their compute resources up or down with ease.

• Low Latency & High Reliability: Proprietary cloud software and in-house hardware design ensure minimal downtime and consistent performance under load.

This cloud environment gives AI developers everything they need to build and deploy models in one place — with significantly less friction compared to general-purpose cloud services.

3. AI Studio: Inference-as-a-Service

On top of its platform and infrastructure stack, Nebius offers AI Studio, a SaaS environment that simplifies access to powerful open-source AI models via APIs. Designed for both researchers and commercial users, the AI Studio enables fast, cost-efficient deployment of foundational models across a range of use cases:

• Plug-and-Play AI: Integration with popular models like ChatGPT, Gemini, LLama, Mistral, Qwen, DeepSeek, and others.

• Wide Model Coverage: From text generation and image synthesis to embedding models for Retrieval-Augmented Generation (RAG) systems, AI Studio supports a broad spectrum of AI applications.

• Fast Model Onboarding: Nebius can integrate trending models within days or even hours — exemplified by its rapid deployment of DeepSeek models.

• User-Friendly Interface: Pre-configured environments drastically reduce setup time.

• Market-Leading Cost Efficiency: One of the lowest price-per-token (if not the lowest) offerings for inference currently available.

This platform is particularly well-suited for customers who don’t want to manage complex infrastructure but still want to experiment, prototype, or deploy AI applications at scale.

Today, Nebius is one of the largest non-U.S. providers of alternative AI infrastructure. The global market for GPUs is intensely competitive: while the majority of AI chips are used internally by model developers like OpenAI, Google, Microsoft, and Meta, only a fraction — around 10–20% — are made available to public cloud users via hyperscalers such as AWS or Oracle.

The remaining 30–40% of GPU supply is fragmented across dozens of alternative players — but only a handful, including CoreWeave, Lambda Labs, Together.ai, Deep Infra, and Nebius, possess the technical and financial capabilities to deploy large-scale infrastructure and serve global demand. Among these, Nebius stands out for its ability to serve a wide spectrum of clients, from startups and research labs to enterprise customers and AI product builders.

As Nebius CEO Arkady Volozh puts it:

“We are one of the few alternatives capable of serving the core needs of major players, supplying GPUs to startups and corporate clients looking to purchase infrastructure, and supporting smaller clients who use models deployed by us. We are giving customers the freedom to choose.”

Competitive Advantages

One of the most common questions customers ask Nebius is: “How are you different from other cloud providers?” The answer lies in a combination of deep engineering talent, reliability, customization, and an obsessive focus on customer experience.

From day one, Nebius set out to rebuild the cloud infrastructure stack from the ground up — and did so in just 12 months. This rapid execution was made possible by its world-class engineering team, many of whom previously worked at Yandex and followed the founding team to help realize a bold new vision: build an AI-native cloud platform that goes far beyond traditional compute reselling.

Here’s what sets it apart in the booming AI infrastructure market.

1. Engineering-Led Execution Enables Vertical Integration

Nebius is built by engineers, for engineers. With nearly 400 AI/ML and cloud infrastructure engineers with over a decade of experience in the field (and a broader tech team of ~850 professionals), the company brings a level of technical depth rarely found outside the top-tier hyperscalers. This allows Nebius to innovate rapidly, adapt to new hardware generations, and maintain full control over its stack — from server design to orchestration software.

While many cloud providers depend on third-party OEMs like Dell or Supermicro for servers, Nebius designs its own racks and servers in collaboration with NVIDIA. These custom designs are forward-compatible with next-generation GPUs, reducing hardware obsolescence risk and improving performance per watt. Sourcing hardware from Taiwanese ODMs, Nebius bypasses traditional markups and optimizes its data centers for both speed and efficiency.

The result:

• Better thermal management and liquid cooling

• Faster deployments of new GPU models

• Lower downtime and failure rates

• Structural cost advantages — some of which are passed on to customers

This technical advantage is not just theoretical. It translates directly into a more stable, responsive cloud environment that supports high-stakes workloads like large-scale model training and real-time inference.

2. Reliability at Scale

In AI development, reliability is everything. A single node failure in a cluster training a large model can delay progress by hours — or worse, corrupt the run entirely. Nebius has made reliability a central design principle.

To mitigate failure at every layer, Nebius builds "auto-healing" infrastructure with fault tolerance baked into every component:

• Hardware: Custom-built servers tested in-house

• Monitoring: Real-time anomaly detection and predictive failure alerts

• Compute management: Fine-grained resource orchestration

• Kubernetes orchestration: Automated recovery at the container level

• Inference services: Built-in failover to ensure continuity

This robustness enables customers to train and deploy models with minimal interruptions — a critical factor for workloads that can run for days or weeks at a time.

The deeper customers go into the Nebius stack, the more seamless and “auto-magical” the experience becomes. This reliability is a key differentiator, particularly as inference workloads begin to scale rapidly across industries.

3. A Customer-Obsessed Cloud

Another factor that truly sets Nebius apart is its customer-first mindset.

Rather than building a general-purpose cloud for generic workloads, Nebius has always focused on the needs of AI developers. The company works closely with customers to understand their goals — whether training a foundation model, scaling multi-modal inference, or optimizing latency-sensitive apps — and delivers infrastructure that fits those use cases.

Crucially, this level of tailored service is a major differentiator. It’s exactly one of the advantages that makes Nebius so attractive — not only to small, AI-native startups that need deep customization from day one, but also to medium and large companies seeking flexible, high-performance infrastructure.

This philosophy extends into product design. Nebius doesn’t just sell GPU time, it builds features customers didn’t know they needed — like its own LLM, which continuously tests the platform and reports back optimization opportunities. This in-house AI ensures that infrastructure keeps evolving in line with user needs.

“We’re a very customer-centric cloud. The cloud experience matters to us and is a top priority almost every day. We want to build a whole AI stack. We try to determine what products our customers will need next — and deliver them first.”

This mindset drives Nebius’ roadmap and underpins new offerings like its AI Studio, which simplifies access to open-source models and offers one of the lowest cost-per-token inference platforms on the market.

4. Industry-Leading Cost and Energy Efficiency

Despite offering high-performance infrastructure, Nebius maintains pricing that’s often 20–25% lower than the average GPU provider — a result of its vertical integration and optimized operations. Key drivers of this efficiency include:

• Full-stack ownership: Controlling everything from server design to deployment

• ODM partnerships: Avoiding OEM markups and customizing hardware for AI workloads

• Data center efficiency: Operating with a Power Usage Effectiveness (PUE) of ~1.13 (same class as Microsoft and Google, and even better than Oracle, Alibaba, IBM, etc)

• Top 5% supercomputing efficiency globally, based on energy use per unit of compute

• In-house software: Avoiding costly third-party licenses and improving automation

These cost savings don’t just benefit Nebius — they’re passed on to customers, making high-performance compute accessible to more startups, labs, and enterprises globally.

Importantly, all of this positions the company among the most energy-efficient AI cloud providers in the world.

How Are Nebius' Data Centers So Efficient?

NBIS’ data centers can achieve optimized cooling and heat recovery to enhance both cost-effectiveness and environmental sustainability.

Optimized Cooling

Its Finland-based data center employs free cooling, eliminating the need for traditional chillers, water, and refrigerants. This approach not only reduces costs but also minimizes the environmental footprint.

• Higher Operating Temperatures: The data center operates at a maximum temperature of approximately 40°C, which is about 10°C higher than the typical limit set by standard hardware. This higher threshold eliminates the need for inlet air subcooling and enables slower, less energy-intensive airflows compared to conventional designs.

• Broader Workload Range: The center functions effectively under 100% workload within a temperature range of 18°C to 40°C, in contrast to most data centers that aim to stay below 27°C due to server architecture constraints. This capability eliminates the energy demands of subcooling, significantly improving energy efficiency.

• Energy Savings: By operating without subcooling requirements, the data center achieves substantial energy savings, aligning with NBIS’ focus on cost efficiency and sustainability.

Heat Recovery

In addition to optimized cooling, Nebius has implemented an advanced heat recovery system that repurposes waste heat for municipal heating, creating additional value for the surrounding community.

• Regional Innovation: The Finland data center is a pioneer in the region, using server-generated heat to meet local heating needs.

• Energy Reuse: Between 2020 and 2023, the center reused over 80,000 MWh of server heat for municipal heating, equivalent to the energy consumed by around 2,500 Finnish households for heating over four years.

• Meeting Heating Needs: More than 50% of the annual heating requirements of the nearby town were covered by this heat recovery system.

• Cost Savings for Households: The system contributed to household heating cost reductions of up to 12%, accounting for approximately 30% of the data center’s electricity costs.

The company plans to replicate these best practices as it expands its data center capacity, positioning it as a leader in sustainable technology infrastructure. These initiatives not only generate cost savings but also create long-term value for both NBIS and the communities it serves.

In a market that’s moving fast and expanding faster, Nebius isn’t just keeping up — it’s setting the standard for what AI-native infrastructure should look like. This blend of performance, reliability, cost-efficiency, and long-term vision makes Nebius one of the most compelling players in the global AI infrastructure race.

Competitive Advantages – Customer Stories

The most practical way to understand Nebius' competitive edge is by listening directly to its customers. Their experiences offer firsthand insight into what makes the platform stand out — not just on paper, but in real-world deployment.

Interview with Higgsfield AI

Higgsfield AI, one of the top AI video startups, chose Nebius after evaluating several GPU cloud providers. According to its Founder:

• Nebius is extremely start-up friendly. Developers often express frustration with larger cloud providers’ lack of clarity around pricing and usage terms. Nebius, on the other hand, offers full transparency and fair, predictable pricing — especially valuable for early-stage companies with dynamic workloads.

• Flexible consumption, no rigid quotas. Unlike hyperscalers that demand yearly GPU commitments, Nebius offers spend-based discounts. This enables start-ups to scale without overcommitting.

• Enterprise-grade infrastructure. Despite being a newer player, Nebius delivers infrastructure and uptime reliability on par with Google Cloud and AWS. Importantly, hyperscalers typically aren't interested in start-ups — they prefer higher-volume customers.

• Token-based pricing and a powerful AI Studio. Developers can access and run a wide range of models — including ChatGPT and DeepSeek — all within a streamlined platform.

• Exceptional support. “The Nebius team is always available. Compared to other providers, their responsiveness is just on another level.”

• Simple, short contracts. While many providers require dense 200-page legal documents, Nebius keeps contracts brief and transparent — sometimes with just 3-5 pages.

Nebius provides a level of customization and flexibility that no other cloud provider offers.

Interview with Dylan Patel, Founder of SemiAnalysis

As an industry analyst, Dylan Patel offers a high-level view of where Nebius fits in the cloud ecosystem:

• The cloud market is fragmenting. Today’s landscape includes hyperscalers, neoclouds, and specialized compute providers. Each has its own stack and engagement model — bare metal, Slurm, Kubernetes, APIs, inference services, and more.

• Nebius supports the full spectrum. While many providers focus on just one layer of the stack, Nebius is built to serve across use cases — from researchers to production teams — making it uniquely adaptable.

“Where Nebius stands out is in its ability to offer that full spectrum. Their answer is always: ‘Let’s do it.’”

• Engineering talent is a core advantage. Nebius designs its own racks, partners directly with NVIDIA, and builds systems with next-gen hardware in mind — avoiding obsolescence even as GPU architectures evolve rapidly.

• Custom infrastructure = better economics. By sourcing from Taiwanese ODMs and optimizing cooling systems in-house, Nebius achieves higher performance at lower cost — savings it can pass to customers.

Interview with Lynx Analytics

Lynx Analytics is a global AI firm working across life sciences. With customers like Genentech and AstraZeneca, their workloads demand both precision and compliance.

• Nebius enables fast and scalable GPU clusters. This was essential for accelerating matrix-heavy computations used in clinical and R&D workflows.

• Better reliability. Lynx suffered multiple data losses with a prior provider — an issue eliminated with Nebius.

“We lost our data twice in seven months with our previous provider. We needed something more stable, scalable, and GPU-compatible. The ability to run Spark on GPU clusters and connect CPUs and GPUs in the same environment was crucial. Nebius checked all those boxes.”

• GDPR-compliant infrastructure. Nebius' European clusters ensured data privacy and compliance for sensitive experiments.

• Better user experience:

“In my experience, AWS works — but I dread opening the UI. You always have to edit some JSON file just to get basic things going. Nebius is different. It was designed from the ground up for AI workloads. It’s simpler, more intuitive, and overall just more enjoyable to use.”

Interview with CentML

CentML is focused on cost- and performance-optimized infrastructure for large-scale AI and computer vision workloads.

• Bare-metal access. CentML runs its own stack directly on Nebius hardware, unlocking maximum efficiency.

• Consistent performance. Many newer providers fail to deliver reliable GPU uptime. Nebius stood out for its stability and responsiveness.

• Early access to NVIDIA hardware. Nebius customers gain access to cutting-edge GPUs like the H100, H200, and B200 before they’re widely available — allowing teams like CentML to stay ahead of the curve.

• Flexible contracts. No need to lock into year-long reservations. CentML appreciated Nebius’ short-term and on-demand pricing options that didn’t sacrifice affordability.

“Nebius is the best performance-per-dollar solution currently available in the market.”

Interview with KissanAI

KissanAI is reshaping agriculture through AI, helping agribusinesses better serve farmers with language-specific and literacy-aware models.

• Reliable, accessible compute. As a bootstrapped startup, KissanAI needed an infrastructure partner that could offer enterprise-grade resources without high upfront costs. Nebius delivered.

• Training and inference on one platform. This simplified operations and removed the need to split workloads across providers, saving the team both time and resources.

• Intuitive and user-friendly experience. Even non-technical users could deploy and manage resources easily.

• Best-in-class pricing, along with added services and features that typically come at a premium with other providers. Combined with fast, responsive support, Nebius became a critical enabler of KissanAI’s mission (unlike other providers, where support is often slow or unresponsive).

Interview with Captions

Captions develops AI tools for creative video storytelling. Its flagship foundation model, Mirage, required stable, large-scale GPU infrastructure to train effectively.

• Significantly improved reliability. Nebius enabled smoother model training, removing the debugging headaches that plagued earlier runs.

• Proactive support. The team’s responsiveness was crucial — especially during scaling and testing.

• Global reliability. With a team based in New York, having around-the-clock support ensured uptime and peace of mind.

“We know we can trust them. Having a partner we can truly rely on has simplified our lives. We know that if something crashes in the middle of the night, their team is there to quickly recover it.”

Other Customer Stories

Beyond these core interviews, many more customers are benefiting from Nebius’ AI-first infrastructure:

• Chatfuel: Leverages Nebius’ infrastructure and Llama-405B models to improve chatbot performance, optimizing training costs and real-time efficiency while enabling quick deployment.

• London Institute for Mathematical Sciences (LIMS): Uses Nebius’ scalable, reliable infrastructure for advanced research on LLMs, enhancing model training and data processing capabilities.

• Positronic Robotics: Uses Nebius' virtual machines with NVIDIA H100 GPUs to train AI models for robotic control systems, aiding in the development of intelligent cleaning robots.

• SynthLabs: Partners with TractoAI (owned by Nebius) to simplify training infrastructure, accelerating the release of the Big Math dataset and improving model training using high-end GPUs.

• Krisp: Reduced model training time by 50–80% by switching to Nebius’ NVIDIA H100 GPUs, enhancing AI models for noise cancellation and speech recognition.

• Dubformer: Relies on Nebius for AI dubbing and localization, handling vast audio datasets and ensuring continuous 24/7 model training for improved efficiency.

• Unum: Streamlined the training of multimodal models and successfully open-sourced several models, advancing research in compact AI models.

• TheStage AI: Optimizes inference performance with Nebius’ GPU instances, ensuring scalability and reliability for model evaluation and deployment.

• Recraft: Uses Nebius to train a generative AI model with 20 billion parameters, overcoming challenges and achieving benchmark-breaking performance in AI design.

• Converge Bio: Uses Nebius to accelerate AI-driven drug discovery by training LLMs on proprietary biomedical datasets. What used to take weeks is now done in days.

Nebius is clearly a standout choice for developers, and it’s easy to understand why. Their focus on flexibility, reliability, cost-efficiency, customer experience, and top-notch customer support sets them apart in a crowded field. By building their platform from the ground up specifically for AI, they avoid the pitfalls of retrofitting older cloud technologies. This approach allows them to deliver exactly what AI developers need, whether startups looking for fair terms and transparency or enterprises seeking cutting-edge infrastructure.

Their early adoption of NVIDIA’s latest innovations, combined with their commitment to engineering excellence, ensures developers always have access to the best tools available. Plus, their customer-centric mindset means they’re not just selling compute power — they’re partnering with developers to help them succeed.

Nebius proves time and again that they’re one of the go-to choices for anyone serious about AI development.

NVIDIA: Client, Partner, and Shareholder

Few companies in the AI infrastructure space can claim NVIDIA as a client, partner — and shareholder. Nebius is one of them.

According to NVIDIA’s most recent 13F filing, the company holds ~1.2M shares of NBIS, underscoring a strategic relationship that extends far beyond a typical vendor-customer dynamic.

A Longstanding Strategic Partnership

The roots of the relationship go back over a decade. Yandex — Nebius’ predecessor — was historically NVIDIA’s largest customer outside the U.S. and China, laying the foundation for deep technical and operational integration. That collaboration has only strengthened post-spinout.

Today, NVIDIA and Nebius are strategically aligned across three fronts:

1. Product Collaboration

Nebius is one of the first cloud providers globally — and the first in Europe — to gain early access to NVIDIA’s most advanced hardware, including the Blackwell and Blackwell Ultra GPU platforms. This access allows Nebius to build systems tailored for the next generation of agentic and reasoning-based AI.

2. Platform Integration

The Nebius platform is built from the ground up using NVIDIA’s accelerated computing stack — from A100 and H100 to H200 and B200 — giving customers full-stack reliability for both training and inference.

3. Joint Go-to-Market and Ecosystem Participation

In June 2025, Nebius was named a Reference Platform NVIDIA Cloud Partner, joining an elite group of infrastructure providers offering validated, regionally impactful AI cloud services. This designation brings tighter integration across NVIDIA’s hardware, software, and deployment playbooks, with benefits including:

- Access to pre-validated reference architectures,

- Seamless deployment of NVIDIA-optimized clusters,

- Enterprise-grade reliability and consistency across global data centers.

Next-Gen Hardware Access

Thanks to its NVIDIA partnership, Nebius is able to move faster than most competitors when it comes to integrating cutting-edge hardware:

• Blackwell Platform (2025): Nebius will be the first European cloud provider to offer NVIDIA’s energy-efficient Blackwell GPUs to customers, enabling higher performance per watt and lower total cost of ownership for large-scale AI workloads.

• Blackwell Ultra & GB300 NVL72: Nebius will also be among the early adopters of the NVIDIA GB300 NVL72 — a 72-GPU superchip system announced by Jensen Huang during his keynote at the last GTC conference. These instances are purpose-built for AI agents, multi-modal reasoning, and physical simulation workloads, and will be available to Nebius customers by Q4 2025.

To support the rollout of these next-gen systems, Nebius is making significant infrastructure investments:

• New Jersey Data Center: The company’s largest data center (under construction), with capacity up to 300 MW and fully dedicated to Blackwell’s architecture.

• Kansas City Cluster Expansion: Originally built with thousands of Hopper GPUs, this flagship U.S. cluster will be upgraded with NVIDIA HGX B200 systems to support high-throughput AI workloads with minimal latency.

Once complete, these two sites will position Nebius as a major AI infrastructure provider in North America.

NVIDIA Dynamo & the Software Stack

Beyond hardware, Nebius is also working with NVIDIA on the software layer. The company recently joined the ecosystem of partners for NVIDIA Dynamo, an open-source inference framework built to streamline and scale GenAI deployment across distributed clusters.

This move allows Nebius to:

• Offer low-latency inference with dynamic model scaling

• Simplify model deployment pipelines for customers

• Drive better performance at the orchestration and runtime level

These types of collaborations are key to differentiating Nebius not just as a hardware provider, but as a full-stack AI platform.

All in all, the relationship with NVIDIA gives Nebius:

• Hardware priority (including pre-release access to chips),

• Engineering collaboration (rack design, thermals, reliability),

• Ecosystem inclusion (Dynamo, Partner Network),

• And most importantly: credibility in a hyper-competitive space.

Nebius vs. CoreWeave

In recent months, anticipation around CoreWeave’s IPO initially served as a potential catalyst for NBIS, given the companies’ shared focus on AI-focused cloud infrastructure. That narrative, however, took a sharp turn as market conditions changed.

CoreWeave had initially aimed to capitalize on the AI frenzy by targeting a $35B valuation. But as macro sentiment weakened, the company was forced to revise its valuation down to $23B — and even then, the IPO was undersubscribed by institutional investors. The lackluster reception raised broader concerns for comparable players, including Nebius, and sparked investor questions about the near-term demand outlook for neoclouds.

Further compounding skepticism were emerging concerns about CoreWeave’s capital structure — specifically, its heavy debt load — and its significant revenue concentration, with Microsoft accounting for over 60% of 2024 revenue. While these concerns are legitimate, they do not apply to Nebius.

Below is a breakdown of the key strategic differences between these two companies — differences that will likely define their respective long-term trajectories.

1. Engineering Resources: In-House Hardware Design

As we discussed before, a core differentiator for Nebius is its deep bench of engineering talent and full-stack control of its infrastructure. While both companies invest in GPU clusters, Nebius designs its own hardware, including racks purpose-built in collaboration with NVIDIA for next-gen chips.

Engineering capability is often underestimated, but it is the cornerstone of durable infrastructure efficiency. The quality of Nebius’ hardware design will matter more over time as generational leaps in GPU performance compress replacement cycles and make poor hardware decisions more costly.

2. Infrastructure Model: Reseller vs. Vertically Integrated Operator

CoreWeave doesn’t own its infrastructure assets apart from GPUs, it focuses on renting colocation space (using third party data centers) to scale rapidly. While this accelerates go-to-market speed, it comes at a cost: greater reliance on third parties, limited hardware control, and increased exposure to power, cooling, and connectivity constraints in shared data centers.

Nebius, by contrast, pursues vertical integration:

• It builds and owns most of its infrastructure.

• It designs hyperscale data centers from the ground up, with AI workloads in mind.

• It optimizes rack-to-chip performance and power utilization holistically.

This control over infrastructure allows Nebius to scale incrementally and efficiently while mitigating external risks. By owning the stack, it also avoids the rent escalations and logistical bottlenecks that often plague GPU resellers.

The result: better uptime, lower latency, and reduced cost per unit of compute — benefits Nebius can either retain as margin or pass to customers.

3. Growth Strategy: Disciplined Scaling vs. Debt-Fueled Expansion

By designing and building its own data centers, Nebius can fully leverage its hardware optimizations. More importantly, its modular architecture allows it to scale capacity gradually, based on demand — a stark contrast to CoreWeave’s aggressive expansion.

CoreWeave rushed to build capacity as fast as possible to secure large customers like Microsoft. While this strategy enabled rapid growth, it also left the company burdened with over $8B in debt. This gets even more riskier when that debt is secured against its GPUs, a fairly fast depreciating asset that brings high financing costs.

In contrast, Nebius follows a pragmatic and financially disciplined approach, avoiding overbuilding infrastructure that could remain underutilized. With $2.4B in cash and no debt, the company is in a far stronger financial position. While Nebius does plan to raise additional capital to support its expansion, the exact approach remains uncertain. The company could opt for share dilution, potentially through a private placement with strategic investors, as it did with Nvidia in December. Alternatively, it could raise capital via debt financing or by selling a stake in one of its subsidiaries. Regardless of the method, any new funding is expected to be strategic and beneficial for long-term growth, especially as it would mean that demand continues to boom (otherwise, Nebius wouldn’t need more than its current cash pile).

This approach reduces downside risk, especially in a capital-intensive sector where missteps can be existential. CoreWeave, on the other hand, is far more vulnerable to market disruptions.

4. Customer Base and Revenue Concentration

CoreWeave relies heavily on large-scale AI clients, with Microsoft accounting for 62% of its revenue in 2024. This dependence on a single customer introduces significant risks, especially as Microsoft ramps up its own infrastructure investments.

Nebius, on the other hand, has a more diversified approach, targeting a broad range of customers, from small startups and developers to large enterprises. This gives the company a more stable revenue base, as it’s not reliant on one anchor client. For smaller businesses, Nebius offers flexible, scalable AI cloud solutions, enabling them to access powerful compute resources without large upfront costs. At the same time, Nebius also provides customized, high-performance infrastructure for larger enterprises that need tailored AI solutions.

Additionally, and as explained before, Nebius focuses on developer experience, with tools like Nebius AI Studio simplifying AI workload deployment. This customer-centric approach makes it appealing to both emerging companies and established businesses, strengthening its long-term market position while reducing exposure to risks from any single client. In contrast, CoreWeave’s focus on GPU-centric infrastructure for high-end workloads limits its customer pool to a narrower segment, mainly large enterprises with substantial budgets and specific needs.

Both Nebius and CoreWeave are positioned to benefit from the AI infrastructure boom, but they are executing fundamentally different playbooks.

Current Numbers and Expansion Plans

As I said, NBIS is entering its explosive growth phase with a rock-solid balance sheet: over $2B in cash and zero debt. This financial strength gives the company the flexibility to aggressively scale while maintaining operational stability. Still, with large-scale expansion on the horizon, additional capital raises are likely — though any such moves are expected to be strategic.

Explosive ARR Growth

Nebius has delivered exceptional growth in Annual Recurring Revenue (ARR), skyrocketing from $21M at the end of 2023 to an expected $220M+ as of March (earnings next week) — a 10x increase in just fifteen months. This performance has been driven by:

• Customer Growth: Active clients surged from around 10 to over 40 (as of December), with continued momentum as the company expands into new verticals.

• Massive Capacity Ramp-Up: GPU capacity scaled from ~2,000 to over 35,000 units, unlocking significantly more workload throughput.

• Customer Scaling: Existing customers are rapidly increasing their usage, indicating strong product-market fit and growing compute needs.

2025 Outlook

Nebius aims to maintain its momentum with the following 2025 targets:

• ARR: $750M to $1B (I believe they might even achieve $1B+, considering that its GPU capacity has been sold out and their expansion plans point to a capacity large enough to surpass that guidance)

• Revenue: $500–700M, with Adj. EBITDA turning positive

• CapEx: $600M to $1.5B, primarily allocated to NVIDIA GB200 GPUs and to building new data centers across owned, colocation, and greenfield sites in Europe and the U.S.

Medium-Term Vision

Beyond 2025, Nebius is positioning to generate multibillion-dollar annual revenue. Key growth drivers include:

• AI Cloud & GPUaaS Market Expansion: Capturing more share as demand surges for scalable AI compute.

• Continued Infrastructure Investment: Ongoing data center builds and GPU procurement to support future workloads.

• Customer Base Diversification: Growing across enterprise, mid-market, and developer segments.

• Software Product Innovation: Expanding high-margin services like API access to open-source models and turnkey AI tools. So far, these value-added services have been given for free to its customers, helping to attract demand. However, as the company deepens its relationship with customers, its software solutions are expected to become a high-margin revenue source.

Global Expansion Strategy

In the U.S., Nebius is shifting into high gear. According to its CEO Arkady Volozh:

"The majority of AI consumption is happening in the U.S., which is why we’re building our sales and marketing structure there. We’ll continue adding hundreds of megawatts in Europe, but we’ll be expanding much more aggressively in the U.S."

To that end:

• New Jersey Data Center: A custom-designed facility scalable to 300 MW, with the first phase online by summer 2025. This alone will exceed Nebius' initial 100 MW U.S. capacity target for the year.

• Kansas City: A second deployment phase will be delivered by Q2 2025.

• Iceland: A new colocation site in Keflavik, powered by geothermal energy, went live during Q1 2025.

In Europe, investments are equally ambitious:

• $1B+ Infrastructure Program: Includes the tripling of the flagship Finnish facility (to 75 MW, up to 60,000 GPUs), new greenfield developments, and colocation builds like the Paris GPU cluster — one of the first in Europe to offer NVIDIA H200s and, in the future, the new Blackwell platform.

And in Southeast Asia, Singapore is emerging as a strategic hub, where Nebius will seek to build a new data center as well.

“Singapore’s location, business-friendly climate, and AI-forward policies give us a strong base to help local ecosystems grow. We’re partnering with governments and enterprises to support AI adoption and unlock regional opportunities.”

Nebius is executing one of the most aggressive and well-funded expansion plans in the AI infrastructure space — with disciplined capital allocation, strong demand signals, and global momentum.

Just a few months ago, its medium-term power capacity target was 240+ MW. Now, the company has already secured over 400 MW — and it’s not planning to stop there:

“With New Jersey, we now have secured expansion capacity to over 400 MW. And we are actively reviewing options to extend this pipeline further as we seek to grow aggressively to multiples of where we are today.”

Its ability to scale both compute and revenue, while entering new markets and maintaining high performance, suggests the company is on track to become a top-tier global hyperscaler in the AI era.

Market Opportunity: A Large And Rapidly Expanding TAM

The total addressable market for AI infrastructure is witnessing explosive growth, and NBIS is well-positioned to capitalize on this trend. According to its internal estimates:

• The TAM is projected to grow from $33B in 2023 to over $260B by 2030, representing a CAGR of 35%. This exponential growth is fueled by the increasing adoption of AI across industries and the rising demand for compute-intensive solutions tailored to AI workloads.

• A key driver of this demand will be inference workloads, which are expected to constitute 64% of AI server spending by 2027, up from 34% in 2023. As AI applications transition from development (training) to deployment (inference), NBIS’ comprehensive AI infrastructure is uniquely suited to meet these evolving needs.

External factors contributing to this opportunity include the rapid expansion of GPU-as-a-Service and AI cloud markets, which are anticipated to grow eightfold over the next seven years. This creates an immense runway for growth for infrastructure providers like NBIS.

Internally, Nebius is leveraging its competitive advantages to capture this market opportunity. As the generative AI market grows, NBIS is not just keeping pace but actively shaping the ecosystem with its focus on high-performance, cost-efficient, and sustainable AI infrastructure.

IMO, the company is primed to capture a significant share of this booming market.

Bear Cases Addressed

The DeepSeek Panic: Is Demand for AI Infrastructure Slowing?

When DeepSeek was released, NBIS plunged nearly 40% in a single session. The market panicked, believing DeepSeek’s surprisingly low development costs signaled a drastic decline in future infrastructure needs. The fear? That building AI models had suddenly become cheap — and the world wouldn’t need nearly as much GPU compute or cloud infrastructure.

Every AI infrastructure stock took a hit. But less than three weeks later, NBIS more than doubled, hitting an all-time high above $50/share. Why?

First, the initial concerns were overblown. DeepSeek’s claims about development costs were soon challenged. As Big Tech earnings rolled in, they reaffirmed multi-year CapEx plans for AI infrastructure, highlighting a long-term trend — not a short-term bubble.

Second, even if compute becomes more efficient, demand may rise because of that — not in spite of it. This is a classic example of Jevons’ Paradox: when a resource becomes cheaper or more accessible, consumption tends to increase. In other words:

Cheaper infrastructure = faster AI adoption

That’s bullish for Nebius. Their platform caters especially well to smaller AI developers, and they’ve already integrated DeepSeek and every major open-source LLM released since. Developers can choose the model they want and run it seamlessly.

Bottom line? The DeepSeek panic was a massive overreaction.

At NVIDIA GTC 2025, Jensen Huang projected that global data center CapEx would exceed $1T by 2028, doubling from 2024 levels. He and Sam Altman both emphasized that new models require ~100x more GPU power than previous ones. That’s not a slowdown — that’s an inflection point.

“CapEx from hyperscalers might be slowing — will that hurt Nebius?”

This is a common concern. But it’s important to separate hardware spend from compute demand. Even if global CapEx moderates in the short term, the underlying demand for compute continues to grow. AI infrastructure isn’t like traditional IT — it’s not cyclical in the same way. Nebius expands based on real, observable demand, giving it flexibility to pace growth without overextending.

Meanwhile, competitors like CoreWeave are much more vulnerable due to their high debt loads and customer concentration. Even small shifts in interest rates or AI spending could put serious pressure on their business model.

“But what about the competition coming hyperscalers?”

AWS, Azure, and Google Cloud will always be giants — but they often serve differente needs. Hyperscalers prioritize scale and broad applicability. Neoclouds like Nebius serve the high-performance, high-compute, AI-native workloads that hyperscalers can’t always optimize for.

Developers often choose neoclouds for several reasons, such as the ones outlined in the section “Competitive Advantages – Customer Stories”.

And it’s not just startups. Even enterprise customers are being underserved by hyperscalers. During Amazon’s most recent earnings call, CEO Andy Jassy admitted that AWS' AI cloud business could be growing faster — if not for capacity constraints. That’s telling. If the world's largest cloud provider can’t meet demand, there's a massive opportunity for neoclouds to step in.

Importantly, most hyperscalers build infrastructure primarily for internal use. That leaves a wide-open market for companies like Nebius.

So far, Nebius serves mostly AI-native startups. But that’s about to change. Over the next 12–24 months, a wave of enterprise adoption is expected, and Nebius is already adapting its platform to support those broader use cases.

“CapEx becomes obsolete as chips evolve — isn’t that risky?”

This is a common myth. GPUs don’t suddenly become obsolete when new versions are released. For instance, Nebius still uses H100s for inference — and they were sold out as recently as March.

More importantly, Nebius’ vertical integration helps future-proof its hardware. As I explained before, the company works directly with NVIDIA to design next-gen compatible racks, reducing the long-term risk of costly upgrades.

That level of engineering depth is a key advantage over other neoclouds.

“Nebius isn’t profitable — isn’t that a problem?”

The current payback period for Hopper GPUs is around 2.5 to 3 years — which is in line with industry norms.

And keep in mind, Nebius has been offering software and platform services for free to accelerate customer onboarding and usage. As monetization ramps up and software revenue kicks in, payback periods should shorten meaningfully.

The company expects to eventually generate EBIT margins of 30%.

“Isn’t AI infrastructure just going to be commoditized?”

Many investors believe this, but I strongly disagree — at least for the foreseeable future. The industry is still dealing with a chronic supply-demand imbalance. Just ask OpenAI, NVIDIA, Amazon, or any other big tech company — demand continues to outpace supply, and that gap isn’t closing anytime soon.

This isn’t a commodity market. Not when:

• Supply chains are still constrained

• New models demand 100x more compute

• Enterprises are only just beginning to adopt AI at scale

The winners will be those who can deliver specialized, reliable, and high-performance infrastructure at scale — and that’s exactly what Nebius is building.

Subsidiaries: Avride

Avride is Nebius' autonomous mobility subsidiary and one of the most capital-efficient players in the self-driving ecosystem. Spun out from Yandex’s self-driving division — originally established in 2016 — Avride represents nearly a decade of R&D in AI-powered transportation. The company is advancing both autonomous vehicles and sidewalk delivery robots using a shared tech stack, giving it a unique dual-focus advantage in passenger mobility and last-mile logistics.

Core Capabilities and Technology

Avride is a U.S.-based company that operates globally with R&D hubs in the U.S., Israel, Serbia, and South Korea, supported by a team of 200+ engineers and developers focused on autonomous systems. Its capabilities include:

• Self-Driving Vehicles: Avride has developed fully autonomous cars for ride-hailing, food delivery, and logistics, with testing conducted across the U.S., South Korea, and EMEA. Its robotaxi fleet has already completed 47,000+ rides, logging over 22 million autonomous kilometers with zero serious accidents.

• Delivery Robots: Purpose-built for urban sidewalks and indoor spaces, these autonomous rovers can travel at up to 8 km/h and cover 55 km per charge. To date, they have delivered 200,000+ client orders globally, proving their viability across food delivery, grocery, and small parcel logistics.

Note: This data is from a few months ago, so the updated numbers are even better.

Key Differentiators

• Efficiency & Safety: Avride boasts a better utilization rate than peers like Waymo or Cruise, with a superior safety record across varied terrain and weather conditions.

• Capital Efficiency: Avride has operated with remarkable discipline, spending just $310M to date — a fraction of its competitors — with an estimated $150M more needed to reach profitability.

• Scalable Manufacturing: Avride designs its robots in-house and partners with a manufacturer in Taiwan, allowing rapid, cost-efficient scaling as demand rises.

Strategic Partnerships and Rollout Plans

Avride is strategically expanding in controlled and high-demand environments, supported by notable partnerships:

• Uber: Collaboration for both ride-hailing and autonomous deliveries, including robotaxi launches in Dallas (using Hyundai IONIQ 5s) and delivery pilots already happening in multiple U.S. cities.

• Grubhub: Deployed 100+ robots at The Ohio State University, averaging over 1,000 deliveries/day. Expansion is planned across Grubhub’s network of 360 campuses, reaching 4.5M+ students.

• Hyundai: Joint plan to deploy up to 100 autonomous IONIQ 5s in 2025, with initial services launching in the U.S. later this year.

• Rakuten: Partnership for robot deliveries in Tokyo, with future plans to integrate Avride’s tech into Rakuten Ichiba, Japan’s major e-commerce platform.

Geographic Expansion

Avride has already made significant inroads across Asia, Europe, and the U.S.:

• In South Korea, it became the first company approved to test autonomous vehicles on all public roads nationwide.

• In Japan, robots have received regulatory approval and started deliveries in central Tokyo with Rakuten.

• In the U.S., Avride is now live in Ohio, New Jersey, and Kansas City, with more deployments expected in 2025.

Growth Outlook

Avride’s near- and mid-term roadmap includes:

• 2024: Deploy 10–20 autonomous vehicles and over 100 delivery robots for R&D and pilot programs.

• 2025: Scale to 100+ self-driving cars and 1,000+ robots, targeting contribution profit breakeven.

• 2026+: Begin unsupervised deployments in multiple cities, aiming for double-digit million-dollar revenues and a fleet of 200+ cars and 3,000+ robots.

Valuation Considerations

Valuing Avride is inherently complex due to its early monetization stage and dual focus on both autonomous vehicles and delivery robots. However, a few peer benchmarks help frame the opportunity. Waymo — the most mature player in the space — was last valued at $45B. While Avride is earlier in its commercialization journey, the technology gap is narrowing, especially considering Avride’s capital efficiency and growing real-world deployments.

A more relevant comparison is Motional, valued at $4.1B following Hyundai’s increased stake in 2023. Notably, Avride was founded four years earlier, has completed significantly more autonomous rides, and is diversified across both ride-hailing and sidewalk delivery — an area Motional has not entered. These factors suggest Avride could warrant a higher valuation.

That said, using Motional's $4.1B valuation as a baseline seems a reasonable approach. With strong partnerships, low capital burn, and a growing commercial footprint, Avride has the potential to reach a multi-billion-dollar valuation — possibly into the tens of billions — over the next 5–10 years.

According to Nebius’ recent 20-F filling, the company is “actively seeking third-party investment into Avride”.

Subsidiaries: Toloka

Toloka is Nebius’ AI data solutions subsidiary, specializing in high-quality training data for the development and scaling of AI — particularly in the era of Generative AI. In 2024, Toloka successfully pivoted its business model to focus more sharply on serving foundational model producers and GenAI companies, which has already yielded impressive results.

Revenue grew 140% YoY in FY2024, driven by both customer expansion and deeper wallet share across product lines. In Q4, Toloka added several of the world’s largest foundational model producers to its client portfolio. The company projects to more than double its revenue in 2025, reaching $50–70M, and continues to diversify across both its Classic and Evolved GenAI solutions.

Toloka’s customer base includes big tech firms like Microsoft and ServiceNow, along with top-tier AI startups. Its reputation is built on deep ML expertise, robust research partnerships, and an ability to consistently deliver the high-quality data required to train cutting-edge AI models.

Recently, Toloka secured a $72M strategic investment led by Bezos Expeditions (the investment arm of Jeff Bezos) and Mikhail Parakhin, CTO of Shopify. Although the valuation of this round wasn’t disclosed, the investment marked a significant shift in Toloka’s governance structure. As part of the deal, Nebius voluntarily gave up majority voting power, meaning Toloka will now be deconsolidated from $NBIS’s financial statements. However, Nebius remains a major shareholder and retains a strong economic interest in Toloka’s future growth.

Toloka’s CEO noted that another funding round is expected in the near future — a potential catalyst that could further elevate the company’s visibility and valuation.

Valuation Considerations

While private companies in this space, such as Scale AI, often command revenue multiples between 10–20x, I took a conservative approach and applied a 5x P/S multiple. Using the midpoint of Toloka’s 2025 revenue guidance ($60M), this would imply a valuation of approximately $300M.

Granted, using a P/S multiple has its limitations, especially when applied to early-stage companies with high growth potential. However, in the absence of disclosed deal terms, it provides a reasonable baseline.

If that seems high, it’s worth revisiting the scale and importance of the AI data infrastructure market — and Toloka’s growing role within it.

Subsidiaries: TripleTen

TripleTen is a leading EdTech platform based in the United States, recognized for its high job placement rates, strong student satisfaction, and consistently positive graduate outcomes. The platform delivers an AI-powered e-learning experience that combines affordability for learners with operational scalability for the business. As it continues to grow, TripleTen is expanding across both B2C and B2B markets — with a particular focus on the U.S. and Latin America — unlocking new revenue streams and long-term growth opportunities.

In Q4 2024, TripleTen reported a 100% YoY increase in student enrollment, driven by strong performance across both its U.S. and LATAM operations. The company’s tuition rates remain among the most affordable in the market for comparable bootcamp-style programs, giving it a compelling value proposition in the competitive online education landscape.

Key Differentiators

• Affordable Pricing: TripleTen leverages AI-driven automation and diversified trainer sourcing to keep training costs low, allowing it to offer courses at highly competitive price points.

• Top-Rated in the U.S.: 87% of graduates find employment within six months — well above industry averages. High levels of student satisfaction continue to fuel strong word-of-mouth growth.

• Proprietary Technology: The company has developed a robust tech stack that enables fast, low-cost course launches and localization, facilitating rapid market expansion.

• Personalized Learning Support: Students receive expert tutoring and hands-on projects that mirror real-world job challenges, boosting skill acquisition and job readiness.

Growth Levers

• B2C Expansion: TripleTen is launching new programs in high-demand areas such as Cybersecurity Analysis and UI/UX Design, targeting emerging job market needs.

• Geographic Scaling: LATAM is a major growth frontier. Deep localization and tailored course offerings are enabling TripleTen to rapidly scale in the region.

• B2B Product Development: The company is creating corporate bootcamps in both English and Spanish, along with role-specific assessments for data and development positions.

• Alumni Network Utilization: TripleTen is strengthening its alumni community to drive enrollment efficiency and lower customer acquisition costs.

• Financial Efficiency: The business operates with a highly efficient cost structure where CAC is fully recouped via initial student payments, ensuring strong unit economics.

Valuation Considerations

While TripleTen is likely the least significant of Nebius’ three main subsidiaries in terms of near-term valuation impact, it remains a valuable strategic asset — especially given its potential to feed skilled talent into Nebius' ecosystem.

TripleTen is projected to more than double its revenue in 2025, with estimated top-line reaching $40–60M. Comparable education technology firms in the private market typically trade at 1–2x revenue multiples, though they often grow at much slower rates.

To stay conservative, we apply a 1.5x multiple to the midpoint of revenue guidance ($50M), arriving at a valuation estimate of $75M.

Beyond its financial contribution, TripleTen offers potential strategic synergies with Nebius’ broader AI and infrastructure ecosystem. As the company trains thousands of job-ready developers and data professionals each year, it could emerge as a talent pipeline — not just for NBIS itself, but also for its enterprise clients.

28% Stake in ClickHouse

Nebius holds a 28% equity stake in ClickHouse, one of the fastest-growing players in the database management systems (DBMS) space. ClickHouse is an open-source, high-performance columnar database optimized for real-time analytics at scale. Its architecture allows users to query massive datasets at lightning speed, making it a powerful tool for big data workloads, business intelligence, log analytics, and AI/ML pipelines.

Engineered for efficiency, scalability, and cost-effectiveness, ClickHouse stands out in a crowded market dominated by legacy systems. Its ability to handle billions of rows per second while maintaining low latency has made it the go-to solution for modern data-intensive applications.

The platform is now widely used in AI, finance, observability, and cybersecurity, and has become a core analytics layer for many high-performance infrastructure stacks. Importantly, it has attracted a blue-chip client base that includes: Meta, Microsoft, Spotify, Sony, Cloudflare, HubSpot, Shopee, IBM, ServiceNow, and more.

This growing adoption reflects ClickHouse’s ability to deliver enterprise-grade performance and reliability while maintaining the agility of an open-source solution. Its flexibility and speed are particularly well-suited for emerging AI and real-time decision-making use cases, where traditional databases fall short.

Valuation Update and Implications

On May 9, reports surfaced that ClickHouse is in advanced talks to raise hundreds of millions in a new funding round, led by Khosla Ventures. The round is expected to triple the company’s valuation to $6B, a massive step up from its last valuation in 2021.

This development is highly significant for $NBIS:

• A $6B valuation implies that Nebius’ 28% stake is now worth $1.68B (pre-dilution).

• That figure represents over 20% of Nebius’ entire current market cap, highlighting the enormous hidden value in this single investment.

• Importantly, this new funding round is acting as a major re-rating catalyst for $NBIS, as investors reassess the value of its strategic assets.

ClickHouse is no longer just an interesting side bet — it’s becoming one of Nebius’ most valuable holdings. As the world’s data infrastructure shifts toward real-time analytics and AI-native architectures, ClickHouse is positioned to be a foundational layer in that transition.

Note: To account for the dilution associated with this new valuation round, I’ll apply an 80% weighting to the $1.68B figure in the Sum-of-the-Parts analysis.

Nebius Group Valuation: Sum-of-the-Parts Approach

When you invest in NBIS, you’re not just buying a fast-growing AI cloud provider — you're getting exposure to three high-potential subsidiaries and a 28% stake in ClickHouse, one of the hottest companies in data infrastructure.

The best way to assess the company’s value is through a Sum-of-the-Parts framework that captures both the core business upside and the hidden value in its strategic holdings.

Core Business

My first valuation model was based on the company’s medium-term target of 240 MW of deployed capacity. But following management’s latest updates, it’s clear that the bar has been raised substantially.

CEO Arkady Volozh recently confirmed that Nebius is now aiming for 1 GW of capacity by 2026 — a dramatic increase from the 60–100 MW initially projected for 2025.

For context, NBIS’ proprietary Finnish data center, once fully expanded to 75 MW, is expected to generate $1B in ARR at full capacity. Extrapolating that performance across 400 MW+ suggests multi-billion-dollar revenue potential in the short term.

To remain conservative, I assume:

• 400 MW capacity by 2027

• 80% utilization

• $12.5M in ARR per MW (based on Finnish data center)

• → ~$4B in ARR by 2027

From a profitability standpoint, I applied a 30% normalized EBITDA margin, which is still below long-term guidance (30% EBIT margin) and peers like CoreWeave (64% adj. EBITDA margin in 2024).

This yields $1B in EBITDA by 2027.

To value this EBITDA stream, I used a 20x EV/EBITDA multiple, which is very reasonable when compared to CoreWeave’s current implied multiple of ~40x (and lower than its 22.5x IPO multiple). That gives us a $20B valuation for the core business by the end of 2027.

Applying a 15% discount rate to account for execution risks, this implies a $15.1B valuation by the end of 2025 — excluding any value from its subsidiaries or strategic investments.

It’s important to note this:

“In March, we were fully sold out… We got additional GPUs and are selling the additional capacity well.”

Given the strong demand Nebius is seeing, I wouldn’t be surprised if my assumptions end up being significantly outperformed. If that happens, the upside could be enormous.

Sum-of-the-Parts Valuation

Now that we have estimates for each part of Nebius Group, let’s sum everything up to arrive at a price target for the end of this year:

$15.1B + $4.1B + $300M + $75M + $1.34B = $20.915B

Again, since the company is expected to spend its $2.4B cash pile fairly quickly, I’m excluding it from this valuation. Additionally, given that the company plans to raise more cash, but it’s uncertain whether it will be dilutive, I’ll assume a 10% increase in shares outstanding from the end of 2024 to the end of 2025 (from 235.75M to 259.3M).

As such, our final valuation for Nebius Group is:

$21.255B / 259.3M = $80.66/share

This represents an upside of over 120% compared to its price as of writing this.

IMO, some of my assumptions were fairly conservative — but even if you tweak them, the upside potential remains substantial.

Catalysts That Could Trigger a Rerating

While NBIS is already executing on a multi-billion-dollar opportunity in AI infrastructure, the stock remains under the radar for most investors — trading at a sharp discount to its intrinsic value. However, several catalysts are on the horizon that could drive a substantial revaluation in the next 6–18 months. Here's what to watch:

1) Initiation of Institutional Coverage

Due to the unconventional method by which Nebius entered public markets, the company currently has virtually no sell-side coverage. This has led to limited institutional awareness and little visibility across financial media and research platforms.

As the company continues to report quarterly earnings, demonstrates execution against aggressive growth targets, and closes high-profile contracts, analyst coverage is inevitable. Once institutions begin modeling the ARR trajectory and appreciating the scale of Nebius' infrastructure build-out, a rerating becomes almost guaranteed.

2) Avride Funding Round

Avride, Nebius’ autonomous vehicle subsidiary, could soon become a high-value funding lever for the core AI cloud business. Management has made it clear that the team is actively exploring strategic partnerships or capital injections to fund Avride’s future independently:

“We are one of the last independent AV companies... We aim to participate in the upside in the short to medium term. In the long term, we could see Avride as a source of capital for AI expansion.”

A deal with a large automaker, Tier 1 supplier, or tech company would instantly:

• Validate Avride’s technology

• Provide a benchmark valuation

• Unlock capital for Nebius’ expansion

• Remove uncertainty around funding needs

Considering that Avride’s true value is likely far higher than what the market is currently implying, this could serve as a major catalyst.

3) Large Contract Announcements

Nebius is actively engaging with larger customers looking for long-term GPU cloud contracts. Management has already stated that the company feels very comfortable hitting its $750M–$1B ARR target by December 2025, and has more demand than it can currently supply.

“We had many promising discussions that may generate further revenues... We’re prepared to scale our New Jersey facility if a large customer comes on board.”

These longer-duration contracts with enterprise or hyperscaler clients could provide:

• Revenue visibility

• Utilization guarantees

• A signal of trust and validation to the broader market

Any such announcement — especially with a blue-chip name — could drive a powerful sentiment shift and increased investor interest.

4) The European AI Infrastructure Boom

There’s a tectonic shift happening in Europe. After years of lagging behind the U.S., the EU is now heavily investing in AI infrastructure — and seeking to reduce dependence on U.S. tech giants like AWS, Azure, and Google Cloud.

Recent developments include:

• Dutch parliament motions to develop a national cloud platform

• Calls for “Buy European” mandates in public tenders

• €200B InvestAI initiative by the European Commission

• France’s €109B AI roadmap, spearheaded by President Macron

As Europe’s most energy-efficient and cost-effective AI cloud provider, headquartered in the Netherlands, Nebius is perfectly positioned to benefit.

At the same time, Nebius is expanding even more aggressively in the U.S. with GPU clusters in Kansas City and a new proprietary facility in New Jersey — showing that it can scale both sides of the Atlantic. If Europe accelerates its AI sovereignty agenda, NBIS becomes a geopolitical and technological beneficiary — potentially a cornerstone of European AI independence.

5) Clarification of Some Perceived Risks

One of the biggest lingering overhangs for Nebius is legacy perception risk tied to its origins. Despite being headquartered in the Netherlands and having zero operational or ownership exposure to Russia, many platforms still mistakenly associate NBIS with its Yandex roots.

A clear example: Several outlets — including Bloomberg terminals — mistakenly linked Nebius to Yandex's Q1 earnings date, creating confusion among investors.

This misinformation has likely kept risk-averse investors on the sidelines. But with every quarterly report, every customer win, and every expansion update, NBIS is proving it’s a fully independent, Western-facing company — with a leadership team that has already built and scaled one of the most sophisticated cloud stacks in the world.

As sentiment catches up to reality, we expect this perception gap to close rapidly — removing a key barrier for institutional capital.

6) Operating Leverage

Nebius is still in the heavy capex and growth investment phase, which is typical for a hyper-growth company in its early stages. As a result, some investors are hesitant due to negative near-term cash flows.

However, the business model is inherently high-margin once infrastructure is in place. Management has already guided to 30% EBIT margins long-term, putting NBIS in line with top-tier cloud providers.

“Revenues from additional services will be growing... In the long term, we aim to achieve 30% EBIT margins like AWS.”

As new capacity comes online, the company will begin to demonstrate strong operating leverage — which is often the turning point when investors pile into the name.

Once that happens, the current valuation will look like a steal in hindsight.

All in all, the clock is ticking. Between a vastly underappreciated core business, three valuable subsidiaries, and a stake in one of the fastest-growing data infrastructure companies (ClickHouse), and multiple near-term catalysts, Nebius is set up for a major rerating.

Each catalyst on its own could drive material upside. Together, they paint a picture of a company that’s misunderstood, mispriced, and massively underfollowed — but not for much longer.

Arkady Volozh: Background and Alignment

The Founder and CEO of Nebius, Arkady Volozh, is not your average tech executive. He’s a visionary with a decades-long track record of building and scaling successful technology ventures — and now, he’s bringing that expertise to one of the most ambitious AI infrastructure companies in the world.

Background

Arkady began his entrepreneurial journey long before most of today’s tech giants existed. Born in 1964 in Kazakhstan, he studied applied mathematics in Moscow and quickly gravitated toward computing and software. In the late 1980s, he co-founded a series of ventures — including Magister, CompTek, and eventually Arkadia Company — that gave him a front-row seat to the birth of the digital age in the post-Soviet world.

By the early 1990s, Arkady had turned his focus to search technologies, laying the groundwork for what would eventually become Yandex, which he co-founded in 1997. Under his leadership, Yandex evolved from a niche Russian-language search engine into a sprawling tech ecosystem worth over $30B at its peak. From AI and cloud to autonomous driving, ride-hailing, e-commerce, navigation, smart devices, and education platforms — Yandex became one of Europe’s most formidable tech companies.

Alignment

Today, Arkady owns ~15% of NBIS and holds the majority of voting power, allowing him to guide the company with a long-term vision. What’s more telling is that ~90% of his personal net worth is tied up in NBIS stock, signaling an extraordinary level of financial alignment with fellow shareholders.

He isn’t just invested — he’s essentially all in.

This alignment is critical. In a sector where short-term pressure often distorts long-term decision-making, having a founder who is both emotionally and financially committed ensures that strategy isn’t dictated by quarterly noise.

A Leader with Principles

Now a dual citizen of the Netherlands and Israel, Arkady has firmly distanced himself from Russia and the geopolitical baggage that might once have shadowed his legacy. He resigned from Yandex’s Russian operations in 2022 and played a key role in orchestrating the company’s restructuring and international pivot.

In 2023, he publicly condemned the Russian invasion of Ukraine, calling it “barbaric” and expressing deep empathy for its victims. This wasn’t a performative move — it came after 18 months of silence, which he used to quietly relocate 2,000 Yandex employees fleeing the war. This action speaks volumes about his values and character. It also reinforces his desire to build something ethical, global, and enduring.

More Than Just a Startup

What Arkady is building with Nebius is unlike any other early-stage company. This isn’t some overfunded startup run by first-timers chasing hype. It’s a mission-driven organization led by an elite team of seasoned operators — many of whom followed Arkady from Yandex — now working without the limitations of their previous environment.

There are very few opportunities in public markets to invest alongside a founder of this caliber, especially one with this level of financial and emotional buy-in. Arkady has spent more than 30 years building cutting-edge technologies and scaling platforms that serve millions. Now, he’s doing it again — only this time, with full independence, fresh capital, and a global canvas.

If you’re looking for founder-led companies where leadership, mission, and shareholder interests are fully aligned, NBIS checks every box.

Final Thoughts

At its core, this isn’t just another tech stock. Nebius Group (NBIS) represents something rare in public markets — a misunderstood, undercovered company building critical infrastructure at the heart of the AI revolution, led by one of the most accomplished tech entrepreneurs of our time.

Despite operating in one of the most exciting sectors globally, NBIS still flies under the radar. Most investors don't realize what they're looking at — a vertically integrated AI cloud provider with a path to multibillion-dollar revenue, three high-potential subsidiaries, a strategic stake in ClickHouse, and one of the best founder alignments you’ll find in public markets.

Yes, there are risks. Yes, the stock is volatile. But when you zoom out, it becomes clear: this is an asymmetric opportunity with the potential to yield outsized returns in the short, medium, and long term.

If the company keeps executing — and I believe it will — today’s share price will look like a rounding error in hindsight.

For all these reasons and more, NBIS remains my largest position — and I can’t wait to see how this story unfolds.

Disclaimer: As of this writing, M. V. Cunha holds a position in Nebius Group (NBIS) at $25.67/share.

That's it! Thank you so much for reading.

Don’t forget to subscribe for more content like this.