Nebius Demand and Order Economics Are Strong, but Expansion Financing and Execution Risks Limit a Rating Upgrade

Overview

This report comments on Nebius's second-quarter results and analyzes neocloud providers in conjunction with the update on CoreWeave. Against a backdrop of strong demand for AI compute, Nebius achieved rapid growth in revenue, EBITDA, and ARR, with improvements in pricing, order size, and payback period, while also raising its contracted capacity target. Morgan Stanley recognizes its vertically integrated AI Cloud, lower total cost of ownership, and customer expansion potential, but believes the current valuation already reflects many positive expectations and that future large-scale expansion still depends on continued financing and stable execution.

Core views

Nebius's key positive signals include growth in contracted capacity, record-high pricing for Blackwell compute, a sharp increase in total contract value of orders, and a shorter payback period. New transactions are mainly scheduled to come online from late 2026, with limited impact on 2026 revenue but significant impact on 2027 and later performance. If the company can continue to bring capacity online, expand its customer base, and improve software monetization, long-term margins and free cash flow have upside potential; however, near-term targets are relatively aggressive, and new booking demand, capital expenditure, project approvals, and financing costs remain the main constraints on valuation realization.

Analysis framework

The report evaluates the neocloud businesses of Nebius and CoreWeave through comparisons of results with market expectations, a breakdown of revenue and capacity drivers, order economics analysis, scenario valuation, and a risk-reward framework, focusing on revenue, EBITDA, capital expenditure, free cash flow, contracted capacity, contract value per unit of capacity, and customer structure.

Methodology notes

• Relative valuationEnterprise value multiple approach

  Uses forward EBITDA or free cash flow as the basis, combined with capital intensity and business risk to determine valuation multiples.

  Nebius's target price mainly references CY28 EBITDA valuation; the $144 target price corresponds to approximately 8x CY28 EBITDA of $11.5 billion. This multiple is at a slight discount to the software industry average to reflect higher capital intensity.

• Scenario analysisBull-base-bear scenarios

  Constructs multiple valuation outcomes based on capacity delivery, customer expansion, margins, financing capability, and competitive environment.

  Nebius's bull case is $400, the base range is $130-$300, and the bear case is $70; the different scenarios mainly reflect differences in capacity coming online before 2030, customer diversification, software monetization, and long-term margins.

• Fundamental analysisOrder and unit capacity economics

  Assesses the profitability quality of compute assets through ACV/MW, total contract value, contract duration, pricing, and investment payback period.

  The report focuses on second-quarter ACV/MW above $20 million, short-term transactions above $40 million, Blackwell auction pricing 15% above the previous historical high, and the transaction payback period shortening from 2 to 3 years previously to about 1 year and 10 months.

Asset mapping & comparison

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

• NEBIUS GROUP NV (NBIS)

  Primary covered stock

  Strengths

  The vertically integrated AI Cloud platform has gained validation from external customers, with strong demand, pricing, and order economics; lower total cost of ownership, software monetization potential, balance sheet, and strategic equity assets can support long-term margin improvement.

  Weaknesses

  The business and profitability track record is short, near-term capacity and booking targets are aggressive, and free cash flow is pressured by high capital expenditure.

  Comparison

  The valuation multiple is at a slight discount to the software industry average to reflect its higher capital intensity; compared with CoreWeave, the report places more emphasis on Nebius's vertical integration, customer selection, and ability to reduce total cost of ownership.

  Risks

  Capacity delivery delays, Vineland approvals, rising financing costs, equity dilution, normalization of AI demand, competition from hyperscale cloud providers, and shorter GPU useful life.

• CoreWeave (CRWV)

  Comparable company in the neocloud industry

  Strengths

  Has specialized capabilities in building and operating GPU clusters, has developed more than $25 billion in business orders in less than five years, and addresses an approximately $361 billion generative AI data center opportunity.

  Weaknesses

  Lacks a long operating and financial track record, has a high debt burden, with debt expected to be approximately $38 billion by the end of 2026, and has significant customer concentration.

  Comparison

  CoreWeave stands out in scale expansion and order accumulation, but its leverage and customer concentration risks are higher than Nebius's; its valuation range already reflects a wide distribution of operating outcomes.

  Risks

  Normalization of AI demand, customers shifting to self-built compute, downward pressure on GPU instance prices, shorter server useful life, difficulty accessing capital, and customer concentration.

Key data

• Second-quarter total revenue$582 millionUp 454% year over year and slightly above market expectations.
• Second-quarter adjusted EBITDA$236 millionMargin was 40.6%, above market expectations of about 30%, and improved by more than 800 basis points quarter over quarter.
• AI Cloud revenue$575 millionUp 514% year over year, accounting for 99% of total revenue.
• Quarter-end ARR$3.0 billionHigher than $1.92 billion in the first quarter.
• 2026 contracted capacity target5 GWThe previous target was more than 4 GW.
• Annual deployment target>1 GWPlanned to be deployed annually from 2027 onward, with a goal of exceeding 5 GW of active power capacity by 2030.
• Flagship agreements4Average deal size exceeded $1 billion.
• Contract value per unit of capacityQ2 above $20 million ACV/MWShort-term capacity transactions in the third quarter exceeded $40 million ACV/MW.
• Blackwell capacity auction pricing15% higher than the previous highest priceReflects that supply and demand for current high-end GPU compute remain tight.
• Transaction investment payback periodApproximately 1 year and 10 monthsA notable shortening from the previous 2 to 3 years.
• Revenue forecasts2026E $3.389 billion; 2027E $10.926 billion; 2028E $19.059 billionNew orders mainly affect 2027 and later performance.
• Vineland data center300 MWThe customer served is Microsoft, and project approval progress still needs to be monitored.

Impact & implications

Strong pricing, expanded order scale, and a shortened payback period indicate that the AI compute supply-demand environment remains favorable for neocloud providers and improves visibility into Nebius's future revenue and margin realization. However, deploying more than 1 GW per year from 2027 onward will significantly increase capital expenditure and funding needs. The company may need to use a combination of customer prepayments, asset-backed financing, corporate debt, and equity-linked financing. As a result, the positive impact of order growth on equity value will depend on financing costs, the degree of dilution, on-time data center delivery, and long-term utilization of GPU assets.

Risks

• The speed at which newly contracted capacity is converted into connected and active capacity may be slower than expected.
• Large-scale data center construction requires additional financing, and rising debt costs or equity financing may cause dilution.
• The 300 MW project in Vineland, NJ still faces uncertainty around hearings, approvals, and delivery.
• AI demand may normalize, and large customers may choose to build their own compute in the future.
• Intensifying competition from hyperscale cloud providers and other neocloud providers may depress pricing and investment returns.
• Actual GPU server useful life being shorter than assumed would weaken asset economics.
• Customer concentration, order cancellations, or insufficient new bookings may affect aggressive growth targets.
• Software monetization, long-term margins, and free cash flow capabilities have not yet been fully validated.
• Morgan Stanley has shareholding, underwriting, and investment banking relationships with some covered companies, which may constitute potential conflicts of interest.

What to watch

• Progress toward completing the 2026 target of 5 GW of contracted capacity.
• The pace of construction, energization, and delivery for deploying more than 1 GW per year from 2027 onward.
• New customer contracts, bookings, and changes in customer concentration.
• Trends in ACV/MW, GPU instance prices, contract duration, and investment payback period.
• Auction and contract pricing for Blackwell and subsequent chip generations.
• Progress on hearings, approvals, and Microsoft-related delivery for the Vineland, NJ data center.
• The mix and cost of customer prepayments, asset-backed financing, corporate debt, and equity-linked financing.
• Realization of capital expenditure, free cash flow, and adjusted EBITDA margin.
• Launch of software products and improvement in software stack attach rate.
• Normalization of AI demand, customers building their own compute, and changes in GPU useful life.