Nebius Stock Price Forecast - NBIS at $89.46: AI Infra Winner Backed by Microsoft and Meta

Nebius (NASDAQ:NBIS): hyperscale AI infrastructure with hyperscale risk

NBIS snapshot, valuation and trading profile

Nebius Group N.V. (NASDAQ:NBIS) closed at $89.46 on Dec 19, with after-hours trading at $90.69. The stock has moved between $18.31 and $141.10 over the past year, with the last session trading in a $80.16–$90.54 intraday range. At current levels, market cap is about $22.53B, average daily volume is ~20.5M shares, and the quoted P/E is roughly 72x on trailing earnings, while revenue growth data show noisy year-on-year swings (headline Rev Growth (YoY) –106.23%) due to the transition from the legacy Yandex structure to a focused AI infra entity.

From Yandex legacy to focused AI and cloud infrastructure

Nebius originates from Yandex N.V., the Dutch parent of the Russian tech group. In July 2024, the company completed the divestiture of all Russian assets after a 2.5-year Nasdaq suspension, re-listing as a largely clean European and Israeli–based AI infrastructure company. The strategic reset is straightforward: build data centers, fill them with high-end NVIDIA GPUs, sign long-duration contracts with hyperscalers, and turn that into a recurring infrastructure cash flow engine. The old consumer-facing Russian assets are gone; what remains is an AI and cloud infra platform built by a team that has already run one of the world’s major search and cloud stacks.

Core engine: Nebius AI GPU cloud and dedicated capacity

The central business line is the Nebius AI segment, which is now the primary revenue contributor, despite contributing only $0.5M in 2022. Nebius AI sells GPU capacity as a service, in two main formats. First, there is classic GPU cloud: customers pick a GPU class and capacity, pay on a per-hour or contracted basis, and run training and inference workloads. Second, and more important for scale, there are dedicated capacity agreements with hyperscalers, where a large MW block and GPU fleet are reserved exclusively for a single customer under multi-year contracts. The flagship example is the Microsoft deal, centered on a New Jersey data center and structured around 300 MW of active power. The contract is sized at $17.4B over five years, or $3.48B per year, which implies around $11.6M revenue per MW per year on that block alone. That contract sits inside a broader target of 100 MW active usage by end-2025 and a very aggressive ramp to 1 GW active usage by end-2026, with the New Jersey facility acting as one of the anchor sites. In parallel, Nebius has a five-year Meta Platforms agreement worth roughly $3B, with the first phase of deployment starting now and revenue ramping through 2026. Together, these contracts validate NBIS as more than a speculative AI story: the company is becoming a core infrastructure vendor to two of the largest buyers of AI compute in the world.

Reported growth: revenue, ARR and capacity-constrained upside

The numbers from recent quarters show why the market is willing to pay a premium multiple on current earnings. In the latest reported quarter, revenue reached $146.1M, a 355% year-on-year increase, after an even more extreme 625% YoY growth rate in the prior quarter. The quarter technically missed consensus by about $9–13M, but that miss is driven by capacity constraints, not demand shortfall, because Nebius reports that all currently available capacity is sold out. Management originally guided full-year 2025 revenue to $450–630M and has refined that to $500–550M, still implying roughly 350% annual top-line growth from the current base. More important than near-term revenue is the contracted run-rate. Nebius expects 2025 year-end ARR of $900M–$1.1B, and 2026 year-end ARR of $7–9B, which is consistent with 1 GW active usage and the economics of the Microsoft and Meta deals. The true bottleneck in the model is not demand but how fast Nebius can secure GPUs, build data centers, and energize power.

Secondary growth levers: Toloka AI, TripleTen and Avride

Beyond the GPU cloud core, NBIS owns three notable side businesses that add optionality but also weigh on current profitability. Toloka AI provides data labeling and AI/LLM training services. Revenue grew from $10.5M to $11.1M to $26.4M between 2022 and 2024, but adjusted EBITDA losses widened to –$40.1M in 2023 from –$22.3M a year earlier. After outside capital—linked to Jeff Bezos—entered, Nebius no longer controls a majority voting stake, and Toloka now appears under equity method or discontinued operations in the latest disclosures. The business is strategically relevant but currently an economic drag. TripleTen, the coding bootcamp unit, is more encouraging financially. Revenue climbed from $2.2M in 2022 to $28.8M in 2024, about 23.2% of consolidated revenue, with quarterly commentary pointing to roughly 100% YoY growth, approximately 6,000 new students per quarter, and rising average fees per student. Adjusted EBITDA losses narrowed to –$30.8M from –$35.7M, suggesting visible operating leverage as the cohort base scales. With tech and AI job demand still strong, TripleTen has a realistic route to profitability if enrollment and pricing continue to climb. Avride, the autonomous driving and robotics unit, is the most speculative. It is tied into Uber, with a plan to make Avride vehicles available in Dallas via the Uber app by the end of 2025, and sits within a broader industry context where Uber is targeting up to 20,000 autonomous vehicles in San Francisco by 2026 alongside partnerships with Waymo and WeRide. Financially, Avride is negligible on revenue, with $0.3M reported in 2024 and an adjusted EBITDA loss of –$67M, meaning it is effectively a long-dated option rather than a current P&L driver.

Capacity roadmap: from sub-GW to multi-GW AI power

Nebius is running an explicitly aggressive capacity expansion strategy. The most recent investor materials indicate a target of more than 2.5 GW of contracted power by year-end 2026, and up to 1 GW of connected power by the same date. This is a meaningful step-up from the prior >1 GW contracted goal. Management also talks about reaching 800–1,000 MW of active power by year-end 2027, which, at revenue densities similar to the Microsoft deal (around $11–12M per MW per year), implies potential ARR in the mid- to high-teens billions and, in an optimistic case, ~$24B if utilization and pricing hold. Under the hood, this implies a GPU fleet measured in hundreds of thousands of units. A reasonable modeling scenario assumes a 60%/40% mix between H200 and B200 GPUs by 2026, driving a total GPU count of roughly 377,000–380,000 units. Per-unit operating costs, including power, networking, storage, facilities, software, and personnel, are modeled in the $2,000–2,500 per GPU per year range, with power draw assumptions of 0.7 kW for H200 and 1.0 kW for B200. Nebius has already indicated plans to deploy more than 20,000 B200 GPUs, and that is likely only the first wave if guidance is to be believed.

GPU pricing, depreciation and the structural margin squeeze

The main economic challenge for any AI infra provider, including NBIS, is the interplay between falling GPU prices, rapid NVIDIA product cycles, and the useful life assumptions used in the accounts. Historical data on H100 pricing show an on-demand rate decline of around 52% since 2022. Older GPUs like the A100, now about five years old, trade closer to $0.80–$1.00 per hour on cloud platforms, compared with significantly higher initial rates. Nebius is deliberately positioning itself at the low end of the price curve on newer hardware. On H100, the company is cited as the cheapest provider at about $2.95 per hour for on-demand usage, undercutting hyperscalers such as Microsoft Azure and AWS on headline pricing. Customers that lock in multi-year commitments with the larger clouds can receive discounts of 30–50%, and AWS has been cutting prices as performance improves to protect market share. The power component of the cost stack is non-trivial: using an average U.S. electricity price of $0.1807/kWh, an H100 at 700W or a B200 at 1,000W running at roughly 80% utilization generates significant power spend over a year, especially when stacked across hundreds of thousands of GPUs. The industry trend is straightforward: as chips age and NVIDIA releases more capable parts like B200 and beyond, per-hour pricing declines while capital costs are sunk, compressing gross margins unless volume keeps rising. Many hyperscalers assume six-year useful lives in their accounting, but the economic life is shorter because customers migrate to newer architectures. Nebius currently uses four-year useful life assumptions for GPUs, which is more conservative but still exposed to rapid obsolescence. The financial modeling in the source work shows that if Nebius simply freezes GPU deployments, revenue per GPU decays as prices fall, and net income turns negative by year four even in an otherwise healthy utilization scenario. The implication is clear: NBIS must continuously add GPUs and power capacity just to maintain earnings, and more aggressively to grow them.

From heavy losses to scalable margins: the 2026 profitability bridge

On reported 2024 numbers, Nebius still looks loss-making. Core EBITDA was about –$266M, with segment-level adjusted EBITDA losses of –$40.1M for Toloka, –$30.8M for TripleTen, and –$67M for Avride. However, once you switch from historical P&L to forward capacity and contracts, the earnings power changes materially. A central scenario anchored on management guidance assumes 2026 ARR at around $7.9B, the mid-point within the $7–9B range, with 300 MW fully active for Microsoft, additional MW for Meta, and sold-out standard GPU cloud capacity around that. Under that scenario, operating income is projected at ~$3.8B, after operating expenses but before interest and tax. With non-current debt of $4.09B, an assumed 8% cost of debt yields ~$327M in annual interest expense. If you then apply a 25% tax rate to pre-tax income, you arrive at an estimated net income of about $2.6B. At the current $22.53B market capitalization, that implies a forward P/E of roughly 7.8x on 2026 earnings. Even if you stress-test the model by increasing total costs by 50%, which would reduce net income to around $1.55B, NBIS would still trade near 13x stressed forward earnings, which is not demanding for a company compounding revenue at triple-digit rates. The second layer is the reinvestment loop. If Nebius reinvests 50% of operating income—about $1.94B—into incremental B200 GPUs, it could add on the order of 55,000+ new GPUs per year at modeled costs. Using a per-GPU revenue figure of ~$20,900 per year, this block would generate roughly $1.16B of additional annual revenue. With around 49% operating margins and 25% tax, first-year net income on that incremental investment would be about $872M, implying a first-year ROI of ~45%. That is an unusually attractive return profile for an infrastructure business, assuming demand and pricing hold.

Cash, leverage and the cost of building 2.5 GW of AI power

The balance sheet has been scaled up to support this level of capex. After a $4.3B capital raise, Nebius reported $4.79B of cash and equivalents at the end of September, up from $1.68B in June. That liquidity sits against $4.09B in non-current debt, giving an enterprise value of roughly $20.16B versus the ~$22.53B equity value. The cash burn is substantial. Operating cash flow in the last quarter was about –$80.6M, a 131% YoY increase in absolute consumption, but the real weight comes from capex of $955.5M on property, plant, and equipment, up 455% YoY. Combined, free cash flow was approximately –$1.04B for the quarter. Management has been explicit that future expansion will be funded by a mix of corporate debt, asset-backed financings, and equity. So far, share count growth has been limited compared to the scale of the capex plan, indicating a preference for debt funding. However, given the magnitude of capex required to reach 2.5 GW contracted power, further equity dilution is highly likely over the next several years. For investors monitoring governance and incentive alignment, it is crucial to track insider behavior through sources like NBIS insider transactions and the broader NBIS stock profile. Sustained insider accumulation into weakness would support the long thesis; systematic selling into strength would argue for a more cautious stance.

Technical and operational edge: HPC, Soperator and cooling economics

Nebius’s competitive positioning is not just about cheap GPU hours; it is rooted in real technical moats. The management team built multiple supercomputers at Yandex—Chervonenkis, Galushkin, Lyapunov—and now operates ISEG, a supercomputer ranked 13th globally. That experience gives Nebius a tangible advantage in interconnect design, high-speed storage, and reliability engineering. Interconnect is critical as GPU clusters get larger; poorly designed fabrics waste compute and drive up costs. Nebius’s history in running a “Russian Google”–scale search engine means it has deep expertise in storage architectures, which maps directly into AI training workloads that are throughput-sensitive. On the software side, Nebius offers Soperator, a stack that combines Slurm, a proven workload manager for HPC, with Kubernetes, the standard for container orchestration. Slurm optimizes large training jobs by decomposing and scheduling them efficiently; Kubernetes adds auto-scaling and self-healing capabilities. Kubernetes alone is not optimized for large-scale model training; Slurm alone lacks modern cloud-native orchestration. Soperator merges the two, aiming to cut wasted compute cycles and lower effective cost per training run, which matters a lot when you are promising industry-low hourly pricing. On the physical infrastructure side, Nebius favors cold-climate deployments such as sites in Finland, where free cooling via filtered outside air allows for simpler cooling stacks, fewer components, and lower power consumption. The company reports power usage effectiveness (PUE) below the global average, which directly lowers operating costs per MW and strengthens the unit economics of the entire GPU fleet.

Risk profile: funding, politics and cyclicality in AI infrastructure

The key risks for NBIS are straightforward but serious. First, the capex and funding cycle is brutal. Building out toward 2.5 GW contracted power and 1+ GW connected power requires persistent access to cheap capital. If debt costs rise or equity markets turn against high-growth AI names, the company could face a higher cost of capital or be forced into more aggressive dilution. Second, while management and many investors argue that Nebius is structurally de-linked from Russia after the asset divestiture, political and perception risk remains. Regulatory or political scrutiny around the company’s origins could affect capital access or customer willingness, particularly in the U.S. and Europe, even if the balance sheet and operational footprint are now outside Russia. Third, the AI infrastructure cycle may prove more cyclical than current sentiment assumes. The current environment is characterized by “sold-out capacity,” hyperscalers racing to secure GPUs, and fear of missing out on foundation models. It is entirely possible that in the late 2020s, as models mature and software efficiency improves, the growth rate of GPU demand slows or flattens. If that happens while Nebius is heavily levered and still in investment mode, the economics could deteriorate quickly. Finally, technology risk is non-trivial. Nebius is heavily exposed to the NVIDIA roadmap and to assumptions about the economic life of top-end GPUs. Any disruption in the supply chain, a change in the GPU competitive landscape, or a misjudgment about depreciation and pricing trajectories could compress margins and undermine the high-ROI reinvestment loop described above.

Valuation framework and peer context for NASDAQ:NBIS

On simple trailing metrics, Nebius looks expensive. Forward P/S is quoted near the mid-30s, with a last-reported reading around 36–37x, and current P/E is over 70x. However, these multiples are based on an income statement that does not yet reflect the ramp from Microsoft and Meta, nor the full effect of the contracted capacity pipeline. On a 2026 view, using $7.9B ARR and $2.6B net income as a central case, NBIS trades at under 8x forward earnings and around 2.6x 2026 sales, which is far more reasonable. In peer comparisons referenced in the source work, Nebius screens as one of the cheaper high-growth AI infra names on a forward PS basis, with only CoreWeave appearing lower, and that discount is arguably justified because Nebius is more diversified and has a stronger software and HPC stack. Another valuation lens considers a scenario where 2025 revenue climbs from $555M toward $3.45B, driving the forward PS multiple to around 5.5x on that intermediate-year revenue base, which again looks reasonable given the growth rate and expected ARR trajectory into 2027. If the company continues to hit its capacity and ARR targets, it is plausible that the market eventually assigns a forward P/S in the 8–10x range, which, applied to a multi-billion dollar revenue base, yields a market cap range around $27–34B versus ~$22.5B today. The gap between current and potential value represents the embedded upside case that bullish investors are underwriting.

Buy, sell or hold: where Nebius stands now

Pulling all of this together, Nebius (NASDAQ:NBIS) is not a cheap stock on trailing metrics, but the forward math anchored on contracted power and hyperscaler deals points to a company that may be undervalued on 2026–2027 earnings power if management executes. At approximately $89–91 per share, with a $22.53B market cap and a 52-week range of $18.31–$141.10, the risk–reward is asymmetrical: downside comes from execution, funding, and AI cycle risk, while upside is driven by successfully turning 2.5 GW of contracted power and $7–9B ARR into a high-margin, compounding cash machine. Based strictly on the data, the capacity pipeline, and the implied forward multiples, the stock justifies a bullish stance. On that basis, and acknowledging the volatility and funding risk that come with the story, NBIS currently aligns more with a “Buy” than a “Sell” or “Hold, for investors willing to tolerate substantial execution and cycle risk in exchange for exposure to what could become one of the core independent AI infrastructure platforms globally.

That's TradingNEWS