Research Finder
Find by Keyword
Can a "Neo-Cloud" Actually Defy the Hyperscale Gravity?
Nebius misses on top-line estimates while doubling down on a $20 billion infrastructure bet and securing 3GW of power for 2026.
02/17/2026
By the numbers:
- Revenue: $227.7 million (up 547% YoY)
- Annualized Revenue Run-rate (ARR): $1.25 billion
- EPS: -$0.99 (missing -$0.42 estimate)
- 2026 Revenue Guidance: $3.0 billion to $3.4 billion
- 2026 CapEx Plan: $16 billion to $20 billion
Key Highlights
- Management is dramatically scaling its footprint with nine new data centers across the U.S. and Europe to capture sustained GPU demand.
- The acquisition of Tavily aims to deliver agentic search capabilities directly within the AI cloud stack.
- Contracted power guidance was revised upward to over 3GW for 2026, signaling aggressive land-grab tactics for energy.
- The core AI infrastructure business reached a 24% adjusted EBITDA margin, proving that the hardware-heavy model can generate cash flow.
- Execution remains tied to massive capital-intensive cycles with a reliance on new debt and asset-backed financing.
The News
Nebius Group reported its Q4 2025 results, characterized by explosive year-over-year growth and a significant bottom-line miss due to heavy investment and deployment timing. The company announced a major expansion of its data center footprint and a strategic pivot into agentic AI through the acquisition of search provider Tavily. For more details, the full financial release can be found here.
Analyst Take
The latest results from Nebius suggest a company that is comfortably running at full throttle, even if the wheels are occasionally kicking up some financial dust. While the $227.7 million revenue figure missed some of the more bullish targets on Wall Street, the 547% year over year growth tells a story of a business that is essentially being built from scratch in a hyper-growth sector. We see this as a classic "build it, and they will come" scenario; except in this case, the customers are already waiting at the gate with their checkbooks. The $1.25 billion ARR is the real number to watch here, as it indicates the scale at which Nebius is now operating. Our analysis suggests that the revenue miss is less about demand and more about the friction of physical deployment. When you are standing up thousands of H100s and Blackwell units, the bottleneck is rarely the sales team; it is the electrician and the supply chain.
What was Announced
The headline announcement centered on the massive expansion of the Nebius "AI Factory" concept. The company confirmed plans for nine new data centers, including a multi-billion dollar facility in Birmingham and a 240MW site near Lille, France. Architected to handle the power density of next-generation NVIDIA Blackwell Ultra and Vera Rubin systems, these sites are designed to provide the physical foundation for the company's aggressive 2026 targets.
On the software side, Nebius announced the acquisition of Tavily. This move aims to deliver an agentic search infrastructure. Unlike traditional search, this technology is designed to let AI agents query the live web to verify facts in real-time. By integrating this into the Nebius AI Cloud 3.1 platform, the company is attempting to move up the value chain from raw compute to a managed agentic stack. They also launched the Nebius Token Factory, a service designed to optimize inference at production scale, which we see as a direct attempt to improve the stickiness of their cloud for AI-native startups.
The financial strategy has also shifted significantly. To fund the $16 billion to $20 billion in planned 2026 capital expenditures, Nebius is moving away from its previous no-debt stance. The company expects to use asset-backed financing and potentially new equity to bridge the funding gap not covered by operating cash flow. This is a high-stakes transition. It moves Nebius from a growth-stage company into a capital-intensive infrastructure utility. We are observing a company that is willing to leverage its balance sheet to the hilt to ensure it does not lose the land-grab for compute capacity.
We are particularly interested in the shift in contract dynamics. Average contract durations increased significantly this quarter, and the company is seeing customers scale from hundreds to tens of thousands of GPUs. This suggests that enterprise clients are no longer just experimenting; they are architecting their core business processes around Nebius infrastructure. The strategic contracts with major tech players, valued at upwards of $20 billion over five years, provide a floor for revenue, but they also create a dependency on a few massive anchor tenants.
The decision to extend the depreciation schedule for GPUs is a move we have seen from the larger hyperscalers. It is an accounting lever that aims to deliver a smoother EPS profile, but it assumes that the hardware will remain relevant for five years. In a world where chips are upgraded every twelve months, that is a bold assumption. If the pace of innovation stays this fast, Nebius might find itself with warehouses full of older silicon that no one wants to rent at premium rates. We believe this risk is mitigated by their focus on the highest-end clusters, but it remains a point of caution.
Furthermore, the focus on power is the most observable part of this report. Securing 3GW of power by 2026 is an incredibly ambitious goal. In the current market, power is the new currency. By locking in these grid connections now, Nebius is architecting a moat that is very difficult for smaller players to cross. Even the largest hyperscalers are struggling with power availability; so seeing a specialized player like Nebius move this aggressively suggests they are thinking three steps ahead in the logistics game. Recent analysis from Bain & Company indicates that the availability of power will be the primary constraint on AI growth through 2028, and Nebius seems to have taken that lesson to heart.
When you look at the margins, the 24% adjusted EBITDA for the AI business is respectable. It shows that even with the massive overhead of building data centers, the actual act of renting out compute is profitable. However, the net loss remains wide. This is a business that is trading short-term profitability for long-term market share. It is a race. They need to get the hardware in the racks and the power in the cables before the market reaches equilibrium. We are watching for any signs of "GPU indigestion" in the broader market, where supply might finally catch up with demand, but so far, the waitlists suggest that day is still far off.
Looking Ahead
Based on what we are observing, Nebius is betting that the world will continue to have an insatiable hunger for GPU compute well into the end of the decade. The key trend that we are going to be tracking is the company's ability to execute on its massive data center build-out without significant delays.
Based on our analysis of the market, our perspective is that the physical layer, the power, and the concrete will be the ultimate decider of who wins the AI infrastructure race. Going forward, we are going to be looking for how the company performs on its debt management and whether the Tavily acquisition actually leads to higher-margin software services. When you look at the market as a whole, the earnings print confirms that the AI boom has moved into the heavy-industrial phase. HyperFRAME will be closely monitoring how the company does in securing its supply chain for the Vera Rubin systems in future quarters.
Steven Dickens | CEO HyperFRAME Research
Regarded as a luminary at the intersection of technology and business transformation, Steven Dickens is the CEO and Principal Analyst at HyperFRAME Research.
Ranked consistently among the Top 10 Analysts by AR Insights and a contributor to Forbes, Steven's expert perspectives are sought after by tier one media outlets such as The Wall Street Journal and CNBC, and he is a regular on TV networks including the Schwab Network and Bloomberg.