Nebius's $40 Billion in Contracts Funds a New AI Factory in Pennsylvania

The publicly traded spinout from Yandex is funding a $20 billion capex plan with customer prepayments, betting on a new model for hyperscale AI infrastructure.

About Nebius

Published

The most expensive bet in AI infrastructure is not being made by a cloud giant. It is being funded by them. Nebius, the publicly traded AI-infrastructure company, is building its future on a foundation of customer prepayments, a model that has secured more than $40 billion in contracted revenue from investment-grade clients like Microsoft and Meta [Nebius, July 2026]. This capital, committed upfront, is financing a planned $16 to $20 billion in capital expenditures for 2026 alone, a sum that includes a new, owned AI factory in Pennsylvania with up to 1.2 gigawatts of power [Reuters, March 2026] [SEC filing, 2026]. The company’s proposition is starkly specific: a GPU-dense, purpose-built cloud for AI training and inference, offered not as a general-purpose utility but as a dedicated, high-performance supply chain for the world's largest model builders.

The Capital-Intensive Wedge

Nebius’s strategy is a departure from the traditional cloud playbook. Instead of building capacity on speculation, the company is using long-term supply agreements with fixed, upfront payments to fund its hypergrowth. Reuters reported the company expects to fund 60% of its growth through these customer prepayments, with the remainder coming from equity and debt [Reuters, March 2026]. This model provides a rare degree of financial visibility in a notoriously capital-intensive sector. The remaining performance obligations,essentially contracted future revenue,stood at around $21 billion at the end of 2025, with Microsoft’s first contract alone worth about $20 billion [The Motley Fool, October 2026]. Critically, these agreements often require customers to pay fees whether or not they use the capacity, with Microsoft’s deal including around $7 billion in upfront payments [us.headtopics.com].

A Full-Stack, Owned Infrastructure

The capital is being deployed to construct what Nebius calls a vertically integrated AI cloud stack. This means moving beyond leasing third-party data center space to owning and operating the entire pipeline, from power and land to the networking fabric between GPUs. The company’s platform provides accelerated compute, from single-node instances to thousand-GPU clusters, connected by NVIDIA’s Quantum-2 InfiniBand fabric [nebius.com]. It has also moved to bolster its software layer, acquiring the agentic search company Tavily for $275 million to integrate those capabilities directly into its cloud platform [Bloomberg, February 2026]. The technical roadmap is focused on peak performance for training and inference, exemplified by being the first cloud to adopt NVIDIA’s Groq 3 LPX to boost token generation rates in its inference service [NVIDIA Newsroom, August 2026].

The Founder's Second Act

The company is led by Arkady Volozh, the co-founder of Russian search giant Yandex. His involvement brings a complex legacy and formidable operational experience to the venture. Volozh was sanctioned by the EU in 2022 following Russia’s invasion of Ukraine, leading to his resignation from Yandex [TechCrunch, June 2022]. After condemning the war and severing ties with Russia, he was removed from the sanctions list in March 2024 and later renounced his Russian citizenship in February 2026 [The New York Times, March 2024] [Bloomberg, February 2026]. At Nebius, he is surrounded by a seasoned leadership team with deep experience in scaling technology businesses, including Executive Chairman John Boynton and Chief Revenue Officer Marc Boroditsky [nebius.com].

The company’s aggressive financing history underscores the scale of its ambition.

Dec 2024 Equity Round | 700 | M USD
Mar 2026 Convertible Debt | 4340 | M USD
Mar 2026 NVIDIA Investment | 2000 | M USD
Jul 2026 Secured Debt | 775 | M USD

The Concentration Risk

For all its forward momentum, Nebius’s bet carries significant, identifiable risks. The business model creates profound customer concentration. The company’s future is inextricably linked to the continued AI spending of a very small number of hyperscalers. While the contracts provide stability, they also create a potential single point of failure if a major partner’s strategy shifts. Furthermore, the capital requirements are staggering, requiring continuous access to debt and equity markets even with customer prepayments. The company must execute flawlessly on its buildout to meet its contractual obligations, as the agreements with Microsoft and Meta reportedly allow the customers a main way out only in the event of Nebius’s own failure to perform [us.headtopics.com].

The Next Twelve Months

The immediate focus is on deploying the raised capital. The construction and commissioning of the Pennsylvania AI factory will be a critical milestone, demonstrating the company’s ability to deliver on its owned-infrastructure promise. Technically, observers will watch for broader adoption of its inference-optimized Token Factory and the integration of the Tavily acquisition. Financially, the market will scrutinize the company’s ability to convert its remaining performance obligations into recognized revenue and to secure additional anchor tenants beyond its current marquee partners to diversify its revenue base.

For the AI researchers and engineering teams at large enterprises, the standard of care for training frontier models has been a patchwork of leased cloud capacity, often juggled across multiple providers to secure scarce GPU inventory. It is a process fraught with complexity, inconsistent performance, and unpredictable costs. Nebius is betting that for the specific, high-stakes population of companies building the largest AI models, the answer is not another general-purpose cloud option, but a dedicated, performance-guaranteed supply chain built from the ground up for their workload. The $40 billion in commitments suggest some of the most demanding customers in the world are willing to pay upfront for that certainty.

Sources

  1. [Nebius, December 2024] Nebius announces oversubscribed strategic equity financing of USD 700 million | https://nebius.com/newsroom/nebius-announces-oversubscribed-strategic-equity-financing-of-usd-700-million-to-accelerate-roll-out-of-full-stack-ai-infrastructure
  2. [Reuters, March 2026] Nebius says ‘well-funded’ for AI race after closing $4.3 billion debt raise | https://www.reuters.com/technology/nebius-says-well-funded-ai-race-after-closing-43-billion-debt-raise-2026-03-23/
  3. [Nebius, July 2026] Nebius raises $775 million in first secured debt financing to accelerate global buildout | https://nebius.com/newsroom/nebius-raises-775-million-in-first-secured-debt-financing-to-accelerate-global-buildout
  4. [SEC filing, 2026] Nebius reports first quarter 2026 financial results | https://www.sec.gov/Archives/edgar/data/1513845/000110465926059872/tm2614392d1_ex99-1.htm
  5. [The Motley Fool, October 2026] Nebius Group: A Deep Dive Into Its AI Infrastructure Business | https://www.fool.com/investing/2026/10/15/nebius-group-a-deep-dive-into-its-ai-infrastructu/
  6. [us.headtopics.com] Analysis of Nebius contracts with Microsoft and Meta | https://us.headtopics.com
  7. [Bloomberg, February 2026] Nebius acquires Tavily | https://www.bloomberg.com
  8. [NVIDIA Newsroom, August 2026] Nebius Token Factory adopts NVIDIA Groq 3 LPX | https://nvidianews.nvidia.com
  9. [TechCrunch, June 2022] Russian tech giant Yandex removes national borders from Maps app | https://techcrunch.com/2022/06/09/yandex-maps-no-borders/amp
  10. [The New York Times, March 2024] E.U. Lifts Sanctions on Yandex Co-Founder | https://www.nytimes.com
  11. [nebius.com] Company website and leadership bios | https://nebius.com/

Read on Startuply.vc