Goldman Sachs Reiterates Buy on Nebius: Strong AI Demand, Enterprise Clients Driving High Margins
AI summary card
Goldman Sachs Reiterates Buy on Nebius: Strong AI Demand, Enterprise Clients Driving High Margins
Nebius’s visibility into AI compute demand extends through 2027 over the next two years; GPU demand from non-hyperscaler enterprise clients is oversubscribed by 3–5x, with significantly higher margins than hyperscaler contracts.
- Strong AI compute demand with visibility extending to 2027
- GPU demand from non-hyperscaler clients oversubscribed by 3–5x, with higher margins
- Guidance for contracted power capacity exceeding 4 GW by end-2026, with >3 GW already secured
- Bloom Energy technology helps reduce site permitting complexity and accelerates deployment
- Software stack and multi-tenancy capabilities represent key differentiators
- Maintain Buy rating with 12-month target price of $234
Report interpretation
Overview
This report summarizes the core takeaways from Goldman Sachs’ Nebius Group management roadshow held in the Netherlands. The firm reiterates its 'Buy' rating on Nebius, citing the company’s position within a robust, multi-year AI compute demand cycle with visibility extending through 2027. The central thesis is that Nebius is transitioning from pure bare-metal hyperscale services toward higher-margin enterprise cloud services—and overcoming power deployment bottlenecks via partnerships such as Bloom Energy.
Core views
Demand side: Nebius continues to see strong demand over the next two years, with customers’ appetite for AI compute extending into next year and beyond—providing full visibility through 2027. Management views this as a multi-year AI compute cycle, with demand expected to broaden further toward inference workloads over time. The company’s positioning benefits from its integrated software stack and platform usability; management believes this factor is becoming increasingly important relative to underlying chip types. Business mix optimization: Demand from enterprises—beyond hyperscale bare-metal transactions—is gaining strong momentum, with new clients added in Q1 including Revolut and Monday.com. Management notes that, excluding hyperscalers, GPU demand per unit deployed is oversubscribed by 3–5x. Goldman estimates these enterprise clients deliver significantly higher margins than hyperscaler contracts, supporting Nebius’s continued investment in its cloud service offerings—which are viewed as a strategically more durable long-term revenue stream. Additionally, client prepayments are seen as a positive signal for demand visibility. Infrastructure and energy: Nebius guides to >4 GW of contracted power capacity by end-2026, of which >3 GW has already been secured; its 2026 grid-connected power target stands at 800 MW–1 GW. Site targets are primarily supported by contracted grid power, while on-site generation (requiring permitting) may increase per-site capacity ceilings over time. Construction plans incorporate buffers around grid interconnection timelines to mitigate risks from energy delivery delays. Technology partnerships: Bloom Energy—offering low-emission, low-noise, combustion-free power—delivers multiple operational advantages for Nebius. First, its lighter hardware footprint enables easier and potentially faster deployment; second, in certain regions it may reduce permitting requirements and improve rollout timelines; third, management believes the technology holds potential to support future 800V configurations. Competitive moat: In Goldman’s view, Nebius’s full-stack software capabilities and multi-tenancy functionality constitute key differentiators. These enable the company to operate a multi-tenant cloud serving a broader customer base than pure bare-metal providers. As the share of non-hyperscaler clients grows—and given stronger pricing dynamics versus large hyperscaler contracts—profitability is expected to improve meaningfully. Its recent agreement with Meta is viewed favorably: it provides guaranteed scale and downside protection, while retaining flexibility to allocate capacity to higher-value enterprise and startup workloads.
Analysis framework
Goldman employs a combined top-down and bottom-up analytical framework. It first confirms the industry-level, multi-year AI compute cycle and its broadening toward inference workloads; second, it examines the company’s business mix—comparing margin profiles and oversubscription multiples between hyperscaler and enterprise clients—to validate the strategic rationale for shifting toward higher-value cloud services; third, it assesses Nebius’s ability to overcome infrastructure bottlenecks (especially power) using technical solutions such as Bloom Energy; finally, it builds valuation models anchored on an expected 2027 enterprise value-to-sales (EV/Sales) multiple.
Methodology notes
EV/Sales Multiple Valuation
For high-growth technology infrastructure companies that have not yet fully realized profitability, analysts often use the enterprise value-to-sales ratio (EV/Sales) as a valuation anchor. This report applies an 8x EV/Sales multiple to projected 2027 sales to derive the target price, reflecting market pricing of the company’s high-growth potential.
Compute Supply-Demand and Oversubscription
By analyzing GPU deployment demand oversubscription multiples (3–5x), analysts gauge supply scarcity and pricing power. Such supply-demand imbalance typically signals stronger pricing authority and higher margins for suppliers—making it a critical indicator of sector health.
Impact of Customer Mix on Margins
Distinguishing margin characteristics across customer types (hyperscalers vs. enterprises). The report highlights that although enterprise clients may be smaller in scale, their margins are significantly higher—and their prepayment model improves cash flow visibility—representing a key dimension for assessing improvements in profit quality.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- Nebius Group (NBIS.US)Direct beneficiary, AI compute infrastructure provider
- Strengths
- Distinctive software stack and multi-tenancy capabilities; high-margin enterprise clients; Bloom Energy technology accelerating deployment; ample power reserves
- Comparison
- Compared to pure bare-metal providers, offers broader customer reach and stronger pricing power
- Risks
- Competitive pressure from hyperscalers; slower-than-expected AI adoption; reduced visibility due to shorter-term contracts
- Bloom Energy Corp (BE.US)Indirect beneficiary / technology partner
- Strengths
- Low-emission, low-noise, combustion-free technology; reduced hardware footprint; potential for simplified permitting in certain regions
Key data
- 12-Month Target Price$234Based on 8x 2027 expected EV/Sales multiple
- Current Share Price$208.37As of close on May 27, 2026
- Implied Upside12.3%Calculated from target price minus current price
- Contracted Power Capacity Guidance (End-2026)>4 GWOver 3 GW already secured
- 2026 Grid-Connected Power Target800 MW–1 GWAnnual construction target
- GPU Demand Oversubscription (Non-Hyperscaler Clients)3–5xReflecting strong enterprise demand
- 2027 Expected EV/Sales Multiple8xCore valuation assumption
Impact & implications
The report concludes that Nebius is building a more sustainable competitive advantage by optimizing its customer mix (increasing high-margin enterprise clients) and resolving infrastructure bottlenecks (power deployment). Collaborations with major clients like Meta provide a floor on scale, while its software stack and multi-tenancy capabilities open upside potential. This implies Nebius can capture both the beta of AI industry growth and alpha from improved operational efficiency.
Risks
- Competitive pressure from hyperscalers
- Slower-than-expected AI adoption
- Reduced demand visibility due to increased share of short-term contracts
What to watch
- Whether contracted power capacity reaches >4 GW guidance by end-2026
- Trends in enterprise client share and margin evolution
- Actual permitting acceleration effect of Bloom Energy technology at deployment sites
- Sustained oversubscription multiples for GPU demand from non-hyperscaler clients