Nebius: From Yandex Asset Spin-Off to European AI Cloud Computing and Autonomous Driving Infrastructure Giant
Founded: Arkady Volozh · Nebius Group N.V.
Key Fields
FIELD STAMPSOrigin
Following the outbreak of the Russia-Ukraine war and subsequent Western sanctions, Yandex was forced to split its original Russian assets from its international assets. Founder Arkady Volozh was forced to sell the Russian search and ride-hailing businesses while retaining the international portion, hoping to keep the technical team operational while leveraging legacy hardware and data center resources to transition into an AI infrastructure provider. Consequently, Nebius no longer serves Russian clients and has rebuilt its commercial identity from scratch for Western markets.
Milestones
Turning Points
- The 2022 Russia-Ukraine war directly destroyed Yandex's original domestic Russian commercial foundation, forcing the company to split into the Russian version of Yandex and the overseas version of Nebius.
- Founder Volozh's resignation as CEO after being sanctioned by the EU actually provided a legitimate window for him to pivot toward an international compliance structure and cut geopolitical risks.
- The 2024 Nasdaq relisting and the introduction of Nvidia as a strategic shareholder provided Nebius with its first large-scale growth capital independent of Russian assets.
- The landing of Microsoft's long-term contract transformed Nebius from a cash-burning GPU speculator into a compute supplier with stable cash flow expectations.
Failures & Pitfalls
- Yandex originally derived about 70% of its revenue from Russian advertising and search businesses; after the spin-off, Nebius could no longer use Russian data centers and payment networks, almost entirely losing its existing revenue streams.
- During the restructuring period from 2022 to 2024, the overseas team experienced a massive wave of core engineers leaving, and Nebius was temporarily unable to independently sustain the development of a full search engine and cloud platform, forcing it to abandon the search engine main line.
- Early attempts to serve Middle Eastern and Asian-Pacific clients using medium-sized European data centers failed, as geopolitical and data sovereignty requirements led to the loss of multiple potential orders.
- During the expansion of the Finnish data center, lower-than-expected efficiency in electricity pricing and local approval processes caused the launch dates of some GPU clusters to be delayed, affecting external delivery momentum in the first half of 2024.
关键成功要素
- Leveraged Yandex's legacy distributed systems and search engine infrastructure experience to pivot into a GPU cluster scheduling platform tailored for AI training and inference.
- Upgraded Nvidia from a GPU supplier to an equity strategic partner, locking in the supply of scarce chips such as H100/A100 and jointly defining the next-generation AI-native cloud architecture.
- Decisively cut off Russian assets and clients to secure compliance eligibility for entering Western markets, resuming Nasdaq listing, and attracting sovereign wealth funds.
- Proactively spun off or sold the autonomous driving business Avride, concentrating all organizational attention on AI compute density and energy efficiency.
Lessons
- Geopolitical shocks can instantly cause a company to lose its domestic market, but the previously accumulated engineer network and data center assets can serve as raw materials for a second growth curve.
- A founder being personally sanctioned is not necessarily a purely negative event; it can sometimes accelerate a company's de-politicized asset restructuring, thereby winning the trust of international capital.
- In the compute arms race, binding with Nvidia helps startup cloud providers establish procurement rankings and financing narratives much more effectively than simply purchasing GPUs.
- Long-term contracts are not the ultimate answer; the real barrier lies in whether proprietary rack, cooling, and scheduling systems can drive total cost of ownership per unit of compute below that of hyperscalers.
Core Data
- Market capitalization on the first day of Nasdaq resumption in October 2024:Approx. $14.6 billion (based on public sources, independent verification unverified)
- Size of the engineer team disclosed in July 2024:Approx. 1,300 people (based on public sources, independent verification unverified)
- Planned GPU procurement investment amount:Over $1 billion from 2024 to 2025 (based on public sources, independent verification unverified)
- Size of the financing round with Nvidia's participation:$715 million (based on public sources, independent verification unverified)
- Equivalent amount of Microsoft's multi-year compute agreement:Approx. 140 billion RMB (based on public sources, independent verification unverified)
- Planned total smart compute capacity:2.5 gigawatts (based on public sources, independent verification unverified)
Competitors / Peers
Nebius directly benchmarks against GPU-native clouds like CoreWeave and Lambda in the European AI-native cloud and compute infrastructure track, while also facing pressure from the three hyperscalers—AWS, Azure, and GCP—in the AI training market. Compared to CoreWeave, which relies more heavily on US data centers and a single source from Nvidia, Nebius possesses self-built data centers in Finland and legacy distributed system capabilities from Russia, though its brand and customer density in the US market still lag behind.
- https://www.36kr.com/p/3592573836607746
- https://techcrunch.com/2024/07/21/from-yandexs-ashes-comes-nebius-a-startup-with-plans-to-be-a-european-ai-compute-leader/
- https://www.iaiol.com/cong-e-luo-si-hu-lian-wang-ju-tou-dao-ai-ji-jian-hei-ma-nebius-ru-he-yong-2-5gw-suan-li-gai-xie-quan-qiu-ai-jing-sai-ge-ju
- https://36kr.com/p/3719493224035716
- https://wallstreetcn.com/articles/3774510