Nebius
Amsterdam-based full-stack AI cloud carved out of Yandex; NVIDIA-backed neocloud (NASDAQ:NBIS) anchored by ~$46B Meta + Microsoft contracts.
1. Core Product / Service
Nebius Group operates three businesses, with the cloud now dominating:
- Nebius AI Cloud — large-scale H100 / H200 / GB200 GPU clusters with InfiniBand fabric, inference-optimized stacks (vLLM, SGLang, TensorRT-LLM), per-second billing, and developer tooling (AI Studio, Slurm-style schedulers). Pricing on A100 ~$1.00-$1.20/hr, undercutting AWS and roughly matching runpod [1].
- Avride — autonomous driving and delivery robotics spin-out, descended from Yandex Self-Driving Group; partnerships with Hyundai and Uber Eats for sidewalk delivery bots.
- Toloka — human-in-the-loop data labeling / RLHF platform for frontier labs (training data, evaluation, red-teaming). Sold a majority stake to Bezos Expeditions + Shopify in 2025.
- TripleTen — re-skilling / bootcamp edtech (smaller, non-core).
Acquired AI search company tavily for $275M (2026) [2] and announced acquisition of Eigen AI ($643M, May 2026) for inference / model optimization [3].
2. Target Users & Pain Points
- Frontier labs and hyperscaler customers (Meta, Microsoft) needing committed multi-GW capacity outside the Big-3 cloud queue.
- AI-native startups / inference shops who want H100/H200 cheaper than AWS/Azure and don't need full enterprise lock-in.
- Autonomous / robotics firms (Avride) — last-mile delivery, food bots.
- Pain points addressed: GPU scarcity at hyperscalers, opaque enterprise pricing, lack of EU-resident AI compute, and lack of integrated data + autonomy stack.
3. Competitive Landscape
| Provider | Tier (SemiAnalysis ClusterMAX 2.0) | Position vs Nebius |
|---|---|---|
| coreweave | Platinum (top) | Largest pure-play neocloud (~$59B market cap); deeper Microsoft/OpenAI ties |
| Nebius | Gold | EU-domiciled, NVIDIA-aligned, fastest revenue ramp |
| Oracle / Azure | Gold | Hyperscaler scale, less neocloud agility |
| lambda-labs | Silver | Strong DL ergonomics, smaller capacity |
| together-ai | Silver | Inference-API focus, not raw GPU |
| runpod | (community/spot) | Cheap on-demand pods; no hyperscaler contracts |
| vast-ai | (marketplace) | Aggregated supply, lowest cost, no SLA |
Differentiation: vertically integrated full-stack (cluster + cloud OS + AI Studio + data via Toloka + robotics via Avride), backed by direct NVIDIA equity and early access to GB200 / Rubin generation [4].
4. Unique Observations
- Yandex pedigree angle — Nebius is the rebranded non-Russia carve-out of Yandex N.V. after Volozh's 2024 separation from the Russia entity. The engineering DNA (search, ML training, autonomy, data labeling) is the real moat — most neoclouds are colo-flippers; Nebius shipped Yandex's distributed-systems team intact. This is why NVIDIA describes the partnership in terms of "engineering depth across the full AI technology stack" [5].
- NVIDIA's $700M (Dec 2024) → $2B (Mar 2026) escalation is one of the largest direct equity stakes Jensen has ever taken in a customer; it locks in priority access to next-gen accelerators and signals NVIDIA's bet that neoclouds (not hyperscalers) will absorb the marginal H200/GB200 unit [5][6].
- Backlog as moat — ~$46B of legally binding Meta + Microsoft commitments [4] makes Nebius supply-constrained, not demand-constrained — the inverse of most neocloud risk profiles. Compare to coreweave, whose Q1 2026 print exposed customer-concentration risk on Microsoft.
- Acquisitions of tavily (search) and Eigen AI (inference optimization) suggest a move up-stack from raw GPU into agent infra — competing with together-ai and Fireworks rather than just CoreWeave. Relevant to ai-inference-engines.
5. Financials / Funding
- Listing: NASDAQ:NBIS (resumed trading Oct 2024 after Yandex separation).
- 2025 full-year revenue: ~$530M; exited 2025 at ~$1.25B ARR [4].
- 2026 revenue guidance: $3.0B-$3.4B (~6x YoY) [4][7].
- Q1 2026 (reported Apr 29, 2026): revenue ~$375-389M, ~+580% YoY; loss per share ~$0.77 [7].
- Backlog:
$46-50B contracted, anchored by Meta ($27B multi-year) and Microsoft (~$19.4B) [4]. - Capital raised:
- Dec 2024: $700M strategic equity (NVIDIA, Accel, Orbis) at $21/share [6].
- Mar 2026: $2B strategic from NVIDIA tied to 5GW capacity by 2030 [5].
- 2026: $4.34B convertible debt round [4].
- Stock: +492% trailing 12 months as of Apr 2026; +87% YTD into Q1 print [7].
6. People & Relationships
- CEO / Founder: Arkady Volozh — co-founder and former CEO of Yandex; engineered the non-Russia carve-out and rebrand to Nebius.
- Strategic investor: NVIDIA (~$2.7B cumulative direct equity + capacity partnership).
- Other investors: Accel, Orbis Investments.
- Anchor customers: Meta Platforms (
$27B), Microsoft ($19.4B). - Acquired: tavily (AI search, $275M, 2026), Eigen AI (inference optimization, $643M, May 2026).
- Subsidiaries: Avride (autonomous driving), Toloka (data labeling, majority sold), TripleTen (edtech).
- Competitors: coreweave, lambda-labs, together-ai, runpod, vast-ai.
Sources
[1] local: daily-log-2026-04-13.md; https://newsletter.semianalysis.com/p/clustermax-20-the-industry-standard (2026-05-09) [2] local: 2026-04-01-diary.md; local: ai-agent-platforms.md [3] https://www.foreignpolicyjournal.com/2026/04/30/nebius-group-nasdaq-nbis-stock-price-rises-4-2-amid-q1-earnings-and-ai-infrastructure-momentum/ (2026-05-09) [4] https://www.fool.com/investing/2026/04/19/nebius-has-landed-46-billion-in-ai-cloud-deals-cou/ (2026-05-09) [5] https://nvidianews.nvidia.com/news/nvidia-and-nebius-partner-to-scale-full-stack-ai-cloud (2026-05-09) [6] https://techcrunch.com/2024/12/02/european-ai-infrastructure-company-nebius-nabs-700m-from-nvidia-accel-others/ (2026-05-09); https://nebius.com/newsroom/nebius-announces-oversubscribed-strategic-equity-financing-of-usd-700-million-to-accelerate-roll-out-of-full-stack-ai-infrastructure (2026-05-09) [7] https://www.foreignpolicyjournal.com/2026/04/30/nebius-group-nasdaq-nbis-stock-price-rises-4-2-amid-q1-earnings-and-ai-infrastructure-momentum/ (2026-05-09)