How Nebius Group Makes its Money: Revenue Breakdown

How Does Nebius Group Make its Money?

Nebius Group is an AI infrastructure company that was formerly part of Yandex, the Russian search giant. After Yandex divested its Russian operations, the remaining international businesses were rebranded as Nebius Group. The company provides GPU-powered cloud computing infrastructure optimized for AI workloads, offering AI training and inference compute to developers and enterprises. Nebius also operates several AI-related businesses including Toloka (data labeling and AI evaluation), Avride (autonomous driving technology), and TripleTen (tech education). The company has built large GPU clusters using NVIDIA hardware in data centers across Finland and other locations, positioning itself as a pure-play AI infrastructure provider competing with hyperscale cloud providers for AI compute customers.

Nebius Group (NBIS) Business Model

Nebius Group Competitors

Nebius Group’s key competitors and comparable public companies in the technology sector include Nvidia, Cloudflare, Equinix, and Super Micro Computer. Each of these companies competes for market share, investor attention, and revenue in overlapping segments. See how Nebius Group stacks up by comparing their revenue breakdown, margins, and growth metrics.

Revenue Breakdown

AI Infrastructure (GPU Cloud Computing) — 57% of Revenue

AI Infrastructure is Nebius’s core business and growth engine, generating $200 million in revenue with an extraordinary 300% year-over-year growth rate. The company operates large GPU clusters — primarily built with NVIDIA H100 and A100 accelerators — housed in data centers in Finland, with expansion planned across Europe and the Middle East. Nebius sells compute capacity to AI developers and enterprises who need GPU hours for training and running machine learning models but don’t want to build their own infrastructure or compete for allocations from AWS, Azure, or Google Cloud.

Nebius occupies an interesting niche in the AI infrastructure stack: it’s large enough to negotiate favorable GPU supply agreements with NVIDIA (a critical bottleneck) but nimble enough to offer pricing and configuration flexibility that hyperscalers can’t match. For mid-market AI companies that need hundreds or thousands of GPUs (rather than the tens of thousands that OpenAI or Meta require), Nebius provides a purpose-built alternative with full-stack software optimization for AI workloads. The Finnish data center locations offer advantages in cooling costs (critical for power-dense GPU clusters) and access to competitively priced Nordic electricity. The 300% growth rate reflects the explosive demand for AI compute capacity globally, though Nebius is starting from a small base and spending heavily on GPU procurement and data center buildout.

Toloka (Data Labeling & AI Services) — 14% of Revenue

Toloka generated $50 million, growing 25%, as a crowdsourced data labeling and AI evaluation platform. Machine learning models require massive amounts of labeled training data — images tagged with objects, text classified by sentiment, audio transcribed and annotated — and Toloka connects AI companies with a global workforce of human labelers through its platform. The business has evolved beyond simple labeling into AI model evaluation and red-teaming, where human annotators assess model outputs for quality, safety, and accuracy.

Toloka competes with Scale AI, Labelbox, and other data labeling platforms, though it differentiates by offering access to a multilingual global crowd and integrating closely with the Nebius AI infrastructure stack. As AI models grow more sophisticated, the demand for high-quality human feedback data (for techniques like RLHF — reinforcement learning from human feedback) continues to increase. Toloka’s growth, while solid, is more modest than the infrastructure business because data labeling is a more competitive and lower-margin market.

Other (Avride, TripleTen, Marketplace) — 29% of Revenue

The “Other” segment ($100 million, +25%) encompasses three distinct businesses inherited from the Yandex restructuring. Avride is an autonomous driving technology platform that operates robotaxis and delivery robots, having been spun out from Yandex’s self-driving car division; it has active testing programs in Austin, Texas and partnerships with food delivery companies. TripleTen is an online tech education platform offering coding bootcamps and data science programs, primarily serving the U.S. market. The marketplace business provides e-commerce technology. While these businesses are collectively growing, they represent strategic optionality rather than core drivers — management may eventually spin off or sell some of these units to focus resources entirely on the AI infrastructure opportunity. The combined $100 million in revenue demonstrates that these aren’t trivial businesses, but their connection to the core AI infrastructure thesis is tenuous.

Nebius Group (NBIS) Income Statement

All values in millions USD unless otherwise stated.

Financial data sourced from Nebius Group SEC Filings.

Nebius Group (NBIS) Key Financial Metrics

• Gross Margin: 14.3%
• Operating Margin: -100.0%
• Revenue Growth: 105.9%

Is Nebius Group Profitable?

No, Nebius Group is deeply unprofitable, posting a $300 million net loss on $350 million in revenue. The -100% operating margin reflects the enormous upfront capital expenditure required to build GPU data centers — purchasing thousands of NVIDIA GPUs at $25,000–40,000 each, building out power and cooling infrastructure, and hiring specialized AI engineering talent. The 14.3% gross margin is razor-thin because GPU depreciation and electricity costs dominate the cost structure before the infrastructure is fully utilized. Nebius is in a classic “invest now, monetize later” phase where it must spend billions to build capacity that will take years to fill and generate returns. The company had approximately $2.5 billion in cash following capital raises (including a major investment from NVIDIA itself), providing significant runway. Profitability depends on filling GPU capacity at attractive utilization rates, improving gross margins as fixed infrastructure costs are spread across more revenue, and demonstrating that purpose-built AI clouds can compete profitably against hyperscale alternatives.

Nebius Group (NBIS): What to Watch

1. AI infrastructure revenue growth and GPU utilization rates — 300% growth is impressive but starts from a tiny base; sustaining triple-digit growth as the revenue base scales will require consistently filling new GPU capacity with paying customers, and any slowdown in AI training demand would leave expensive infrastructure underutilized
2. Competition with hyperscale clouds — AWS, Azure, and GCP are spending $150+ billion annually on AI infrastructure; Nebius must demonstrate that being a specialized, purpose-built AI cloud offers sufficient differentiation in price, performance, or flexibility to attract customers away from the hyperscalers
3. NVIDIA GPU supply and partnership — NVIDIA invested directly in Nebius, providing both capital and preferential access to scarce H100/B100 GPU supply; the depth of this relationship is crucial because the entire business depends on continuously procuring next-generation chips
4. Gross margin expansion trajectory — 14.3% gross margin is unsustainably low; as data centers fill and infrastructure reaches higher utilization, gross margins should expand toward 30-50%, and investors need to see this progression quarter-over-quarter
5. Portfolio simplification and strategic focus — Avride, TripleTen, and the marketplace business dilute the AI infrastructure narrative; any moves to spin off, sell, or wind down non-core businesses would sharpen the investment thesis and potentially unlock value

Nebius Group (NBIS) Financial Summary

Nebius Group is a pure-play AI infrastructure company built from Yandex’s international technology assets, growing revenue at 106% annually on the back of explosive demand for GPU cloud computing. With $350 million in revenue, a $300 million net loss, and a $30 billion market cap, the stock trades at a significant premium to current fundamentals, pricing in the expectation that Nebius can scale its GPU cloud into a multi-billion-dollar business competing alongside hyperscalers. The NVIDIA strategic investment provides both credibility and GPU access, while the Finnish data center locations offer cost advantages. The key risk is execution: building and filling large-scale GPU data centers profitably is capital-intensive, technically complex, and highly competitive. Investors are paying for the possibility that the AI infrastructure market is so large and growing so fast that multiple specialized players can succeed alongside the hyperscale giants — a thesis that will be tested as Nebius scales past $1 billion in annualized AI infrastructure revenue.