Is NBIS the Next Hyperscaler? Inside the Native AI Cloud Play

There is a company growing its revenue at 547% year over year, sitting on $46 billion in contracted backlog, and counting Microsoft and Meta as its biggest customers. Its name is Nebius Group. Its ticker is NBIS. And it may be the most important infrastructure story in markets right now that most investors have not fully priced in.

The question is not whether Nebius is growing. The numbers answer that. The real question is whether this company can do something no one has done since Amazon built AWS: become a genuine hyperscaler from scratch.

What Nebius Actually Is

Nebius Group is not a startup. It was carved out of Yandex, the Russian technology conglomerate, in 2024 and relisted on NASDAQ under the ticker NBIS. The company was built from the ground up as a purpose-designed AI cloud, not a general-purpose cloud that retrofitted AI capabilities on top of existing infrastructure.

That distinction matters more than it might seem. AWS, Microsoft Azure, and Google Cloud were all built to serve general enterprise workloads. Storage, databases, web hosting, virtual machines. They added AI infrastructure on top of that foundation, which means they are managing legacy architecture alongside cutting-edge GPU clusters. Nebius has none of that baggage. Every data center it has built, every system it has designed, has been optimized for one thing: running AI workloads at scale.

The company operates what it calls a full-stack AI platform. That means it provides the physical GPU clusters, the software layer for model training and inference, developer tools, and increasingly, its own AI services. The Token Factory product handles production-scale inference and post-training. The Tavily acquisition in early 2026 added agentic search capabilities. The AI Cloud 3.1 "Aether" release brought enterprise-grade security and faster performance. Nebius is not just renting GPUs. It is building the operating system for AI development.

The Numbers That Matter

Full-year 2025 revenue came in at $529.8 million, up 350% from $118 million in 2024. The fourth quarter alone generated $227.7 million, a 547% increase year over year. The core AI cloud business grew even faster, up 802% in Q4 to $214 million. The annualized run rate at year-end 2025 was $1.25 billion.

For 2026, management is guiding to an annualized run rate of $7 billion to $9 billion. That is roughly a 7x increase in one year.

The company also turned EBITDA-positive for the first time in Q4 2025, posting a 24% adjusted EBITDA margin on the AI infrastructure business. Cash on hand was $3.7 billion at year-end, including $834 million in positive operating cash flow in the fourth quarter alone.

These are not the numbers of a company that is still figuring out its business model. They are the numbers of a company in the middle of a hypergrowth inflection.

The $46 Billion Backlog

The most important data point for understanding Nebius is not its current revenue. It is its contracted backlog.

In September 2025, Microsoft signed a five-year agreement to purchase $17.4 billion worth of dedicated GPU infrastructure capacity from Nebius, with an option for an additional $2 billion. In December 2025, Meta signed a $3 billion five-year deal. Then in March 2026, Meta expanded its commitment dramatically: $12 billion in committed capacity, with an option for an additional $15 billion. That brings the total potential backlog to approximately $46 billion.

Read that again. Microsoft and Meta, two of the largest technology companies in the world, companies that collectively spend hundreds of billions of dollars on their own infrastructure, are outsourcing AI compute capacity to Nebius. That is not a ringing endorsement. That is a structural signal. Even the hyperscalers cannot build fast enough to meet their own AI demand. They are buying from a company that built faster and smarter.

Nvidia has also made a strategic investment in Nebius, with a stated goal of helping the company reach 5,000 megawatts of data center capacity by 2030. When Nvidia invests in your infrastructure business, it is not just a financial bet. It is a supply chain guarantee.

The Native AI Advantage

The legacy hyperscalers face a structural problem. Their data centers were designed for general-purpose compute. Retrofitting them for high-density AI workloads requires enormous capital expenditure and time. The cooling systems, power infrastructure, and network architecture that AI training clusters require are fundamentally different from what a traditional cloud data center needs.

Nebius designed its facilities for AI from day one. That means better energy efficiency, higher compute density per square foot, and lower unit economics over time. It also means faster deployment. The company ended 2025 with 170 megawatts of connected capacity and is guiding to exit 2026 with 800 to 1,000 megawatts. That is a roughly 5x increase in a single year, which is only possible because the company has been pre-positioning infrastructure, securing power contracts, and building out sites in advance.

The US expansion is accelerating. Data centers are operational or under development in Kansas City, Minneapolis, New Jersey, Alabama, and Oklahoma. The Minneapolis facility alone is a 31-megawatt sovereign AI hub designed for Blackwell-generation GPUs.

The Risks Are Real

None of this means NBIS is a risk-free investment. The risks are significant and worth understanding clearly.

The company is still burning cash at the group level. The gross margin metrics look ugly at the consolidated level because infrastructure buildout costs are front-loaded. The company is spending aggressively now to capture a market that it believes will be enormous. That bet could be wrong.

Customer concentration is a serious concern. Microsoft and Meta together represent the overwhelming majority of the contracted backlog. If either company pulls back on AI spending, renegotiates terms, or decides to build more capacity in-house, the revenue trajectory changes dramatically.

Execution risk is also real. Scaling from 170 megawatts to 800 to 1,000 megawatts in a single year requires flawless project management, reliable power procurement, and no major supply chain disruptions. Any one of those factors could cause delays.

Finally, the valuation is demanding. At roughly $40 billion in market cap and $530 million in trailing revenue, the stock is trading at approximately 75 times trailing sales. That multiple is pricing in years of hypergrowth and near-perfect execution. There is very little margin for error.

The Bigger Picture

The legacy hyperscalers spent a combined $416 billion on capital expenditures in 2025. That number is projected to grow to $700 billion or more in 2026. And yet, even with that level of spending, the demand for AI compute is outpacing supply. That gap is exactly what Nebius was built to fill.

The company is not trying to replace AWS or Azure. It is not competing for the enterprise software customer who wants managed databases and serverless functions. It is competing for the AI developer, the research lab, the foundation model company, and increasingly, the hyperscaler itself that needs more GPU capacity than it can build on its own timeline.

If that market continues to grow the way the capital flows suggest it will, Nebius does not need to become the next Amazon Web Services. It just needs to become the infrastructure layer that AI runs on. That is a smaller target, but it is still a very large one.

Whether NBIS is the next hyperscaler is the wrong question. The better question is whether it becomes the infrastructure backbone of the AI era. The $46 billion backlog suggests that answer may already be yes.

This article is for informational purposes only and does not constitute financial advice. Always do your own research before making investment decisions.