Nebius is still in the early-benefit stage of agentic workloads

Overview

This report is Citi's meeting note following attendance at Nebius's Inflection event held in San Francisco. The event was product- and customer-focused, with no major financial or capacity plan updates, but Citi believes the meeting reinforced the investment thesis that Nebius is evolving from a pure AI compute supplier into a full-stack AI platform. The company's investments in hosted inference, system-level optimization, and agent infrastructure may enable it to capture more value as multi-step agentic workloads expand.

Core views

Core views include: first, AI compute supply and demand remain tight, and Nebius's GPU demand is reportedly oversubscribed by 2-4x, supporting strong pricing; second, enterprise customers are moving from experimentation to scaled adoption, increasing the importance of cost visibility, model routing, and internal tools; third, Nebius's software stack and capabilities such as Agent Blueprint help the company shift from raw compute toward pricing based on tokens, workflows, and outcomes; fourth, feedback from customers or partners such as Shopify, Revolut, Mastercard, Databricks, DataRobot, and Cognition shows that demand sources are becoming more diversified.

Analysis framework

The report combines event observations, management comments, statements from customers and ecosystem partners, and Citi's existing valuation framework to assess Nebius's positioning in AI training, inference, and agentic workloads. The valuation uses regression analysis of software sector trading multiples against growth and profitability, along with qualitative adjustments based on the company's competitive position, growth sustainability, and financial leverage during its future build-out phase.

Methodology notes

• Valuation methodsGrowth and profitability regression multiple method

  Target EV/Revenue multiple

  Citi estimates Nebius's valuation by forecasting software company growth and profitability and referencing the regression relationship of software sector trading multiples; the $287 target price implies about 6.3x FY27E EV/Revenue.

• Risk ratingCiti Research High Risk

  High-risk stock risk rating

  Citi classifies Nebius as High Risk due to factors including its short trading history, the high capital intensity of multi-GW AI data center expansion, volatility in the AI theme, and significant exposure to a small number of large hyperscaler customers.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Nebius Group NV (NBIS.O)

  Core covered name

  Strengths

  GPU demand is oversubscribed, demand for AI training and inference is strong, the software stack is expanding, the customer base is extending from AI-native companies to enterprise customers, and preparation for the Vera Rubin architecture transition is more complete than in the Blackwell phase.

  Weaknesses

  Limited trading history, high business capex intensity, still in the large-scale data center construction phase, and visibility on part of its revenue depends on a small number of large customers.

  Comparison

  Compared with traditional cloud or software companies, Nebius is more exposed to GPU supply-demand dynamics, data center construction, and shifts in AI workloads; compared with pure compute suppliers, the company is attempting to capture higher value through inference, orchestration, and agent infrastructure.

  Risks

  Customer concentration, GPU supply chain delays, data center construction execution, falling GPU prices, regulation and geopolitics, financing costs, and potential equity dilution.

• Shopify Inc. (SHOP.US)

  Customer or partnership case

  Strengths

  The report mentions that Shopify uses Nebius for large-scale training and emerging production workloads, and views inference as a potential expansion opportunity.

  Weaknesses

  The report does not provide Shopify's own financial or investment rating analysis.

  Comparison

  As an enterprise customer case, Shopify shows that Nebius serves not only AI-native companies, but can also support large enterprise production workloads.

  Risks

  Its revenue contribution to Nebius, contract size, and continuity were not disclosed in the material.

Key data

• Report date2026-06-10Shown on the report cover as 10 Jun 2026 16:30:00 ET.
• Investment ratingBuy/High RiskThe report lists Nebius's rating as Buy/High Risk.
• Target price$287Citi's one-year target price for NBIS.
• Implied target multipleabout 6.3x FY27E EV/RevenueFrom the report's valuation section.
• GPU demandoversubscribed by about 2-4xManagement said demand per GPU remains oversubscribed.
• Higgsfield AI demand growthabout 2x over the past two monthsCustomer feedback indicates rapid growth in AI adoption and agentic usage.
• Higgsfield agentic traffic shareabout 10%After the Supercomputer launch, about 10% of total traffic was related to agentic usage.
• 2030 active power targetabout 5GWMentioned in the upside risks as NBIS progressing toward an approximately 5GW 2030 active power target.
• FY26 ARR customer concentrationMSFT and META account for about 40%Mentioned in the downside risks that the two customers account for about 40% of FY26 ARR guidance.
• ClickHouse stakeabout $3.75BMentioned in the upside risks that monetizing the ClickHouse stake could improve capital efficiency.

Impact & implications

If Citi's view plays out, Nebius's investment story will rely not only on GPU capacity expansion, but will shift toward a higher-value, software-driven AI infrastructure platform, benefiting from enterprise demand for inference, agentic workflows, model routing, observability, and cost control. However, this thesis also raises requirements for execution, financing, supply chain, and customer retention, and the share price may fluctuate significantly with the AI capex cycle, GPU supply-demand dynamics, and the timing of large customer contracts.

Risks

• MSFT and META account for about 40% of FY26 ARR guidance, and customer concentration may affect growth visibility.
• If supply or delivery of next-generation GPU architectures such as Blackwell, Rubin, and Vera is delayed, revenue conversion could be constrained.
• Multi-GW data center construction faces execution risks related to land, power, permitting, and construction.
• If GPU pricing declines faster with the hardware cycle, revenue and profit expectations will be compressed.
• EU data sovereignty, AI governance, and energy usage regulation may increase compliance pressure.
• Although the historical Yandex relationship has been separated, it may still create investor concerns.
• If hyperscaler prepayments decline or the credit environment tightens, the company may face higher debt costs or equity dilution.

What to watch

• Customer mapping for the Vera Rubin architecture and actual capacity delivery progress.
• The degree of GPU supply-demand tightness and changes in pricing per GPU.
• Changes in the revenue mix across training, inference, and agentic workloads.
• Adoption rates of high-margin services such as Token Factory, SLURM/Kubernetes orchestration, agentic workflows/search, and Tendem integration.
• Contract renewals, expansion, and prepayments from MSFT, META, and other large enterprise customers.
• The conversion cycle from connected power to active power in data centers and progress toward the approximately 5GW 2030 target.
• Financing structure, ABS/convertible bond costs, and potential monetization of the ClickHouse stake.