Deutsche Bank Deep Dive on CoreWeave: Validates 25% Contract Contribution Margin, Maintains Buy

Overview

This report provides a deep quantitative validation addressing market skepticism about CoreWeave’s profitability and returns on AI infrastructure investments. Deutsche Bank dissects the company’s cost structure and adjusts for 'pre-revenue scaling costs,' confirming management’s claim of an approximately 25% contribution margin for steady-state contracts. It further estimates that a typical 5-year contract delivers a levered pre-tax IRR of 35%-40%. The report argues that while CY26 margins are temporarily pressured by rapid capacity ramp-up, the underlying unit economics remain robust and sustainable, and the current valuation already reflects risks—thus maintaining its Buy rating and $135 price target.

Core views

The core thesis centers on the idea that 'true profitability is obscured by accounting treatment.' The analysis notes that fluctuations in CoreWeave’s reported non-GAAP gross margins primarily stem from 'pre-revenue scaling costs'—including rent, power, and equipment depreciation—incurred 1-2 months before new capacity goes live, rather than any fundamental erosion in business profitability. By adding back these costs and excluding corporate-level expenses, the report validates a CY25 steady-state contract contribution margin of approximately 25%, closely aligning with the company’s disclosed example contracts. Even after allocating R&D expenses, this margin level remains comparable to leading cloud providers, demonstrating the commercial viability of its AI infrastructure business. On investment returns, the report conducts a detailed cash flow waterfall analysis based on the company’s 5-year contract model. Results show an unlevered pre-tax IRR of approximately 15% per contract, assuming no residual GPU value at the end of year 5. With CoreWeave’s typical SPV financing structure (80% loan-to-cost ratio at 6.5% interest), the levered pre-tax IRR at the parent company level rises to 35%-40%. This return significantly exceeds the cost of capital, and thanks to long-term 'take-or-pay' contracts and investment-grade customer backing, the associated risk remains relatively contained. The report emphasizes that such high returns are scarce amid the current AI infrastructure demand surge. Looking ahead, the report views CY26 as a cyclical low point for margins. With ARR growth far outpacing revenue growth, pre-revenue scaling costs are expected to drag down operating margins by approximately 700 basis points in CY26, pushing non-GAAP operating margins to around 8%. However, as capacity expansion stabilizes and operating leverage kicks in during CY27, operating margins are projected to rebound into the 15% range. Furthermore, if pre-revenue costs are capitalized and amortized over the GPUs’ 6-year useful life to better reflect economic reality, CY27’s adjusted EV/EBIT multiple would be approximately 18x—a reasonable valuation given CoreWeave’s significantly higher compound growth rate compared to the S&P Tech Index.

Analysis framework

The report combines 'unit economics reconstruction' with 'full lifecycle return modeling.' First, instead of relying directly on reported gross margins, analysts identified the unique accounting mismatch in AI cloud providers—where costs are incurred before revenue recognition—and constructed adjustments to strip out pre-revenue scaling costs from current-period P&L, thereby revealing comparable steady-state contribution margins. Second, in assessing investment value, the report moved beyond traditional annual P/E or EV/Sales metrics and adopted a project-level discounted cash flow mindset. By incorporating SPV financing structures to calculate levered IRR, it translated abstract 'AI infrastructure investments' into concrete project-level return metrics. This approach is especially effective for evaluating capital-intensive, long-cycle infrastructure companies in hyper-growth phases, as it filters out short-term accounting noise and reveals long-term value creation potential.

Methodology notes

• Company Fundamentals & Financial FrameworkProfitability Quality Analysis

  Pre-revenue Scaling Costs Adjustment

  AI cloud providers incur rent, power, and depreciation costs before new capacity goes live, distorting reported margins during expansion phases. The report quantifies and adds back these 'timing mismatch' costs to reveal the true steady-state contribution margin, helping investors distinguish between 'growth-related losses' and 'structural low profitability.'

• Valuation Methodology

  Project-level Levered IRR

  For capital-intensive infrastructure businesses reliant on debt financing, equity-only returns can understate value. The report incorporates SPV financing structures (e.g., 80% LTC) to calculate parent-company levered IRR after interest and principal repayments, offering a more accurate reflection of shareholder capital efficiency when leveraging low-cost debt to amplify high-return projects.

• Valuation MethodologyEV/EBITDA valuation

  Normalized EV/EBIT

  When capital-intensive firms experience depressed EBIT due to lumpy capex, simulating amortization of expansion costs over asset life (e.g., 6 years) yields a smoother, more comparable 'normalized' earnings base, preventing valuation distortions caused by short-term investment timing.

Asset mapping & comparison

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

• CoreWeave (CRWV.US)

  Primary beneficiary: Surging AI infrastructure demand has driven backlog close to $100B, with validated healthy and sustainable unit economics

  Strengths

  Steady-state contract contribution margin ~25%; levered IRR of 35%-40%; high proportion of long-term take-or-pay contracts with investment-grade or top-tier AI lab clients; OpEx stable at ~10% of revenue, strong operating leverage

  Weaknesses

  Temporary CY26 margin pressure due to expansion; high customer concentration; less IP depth and ecosystem breadth compared to hyperscalers

  Comparison

  Trades at a premium to S&P Tech Index but with faster growth; unit economics comparable to mature cloud providers, though uses more aggressive financing to boost ROE

  Risks

  High customer concentration; well-funded competitors with strong incentives; high financial leverage amplifies downside risk; construction delays could postpone margin recovery

Key data

• Steady-State Contract Contribution Margin~25%CY25 estimate; adjusted for pre-revenue scaling costs and corporate expenses, validating management guidance
• Levered IRR for 5-Year Contract35%-40%Pre-tax; based on SPV financing with 80% loan-to-cost ratio and 6.5% interest rate
• CY26E Non-GAAP Operating Margin~8%Down ~500bps from CY25 due to pre-revenue scaling costs; expected cyclical trough
• CY27E Adjusted EV/EBIT~18xNormalized by amortizing pre-revenue costs over 6 years; implies growth premium vs. S&P Tech Index
• Pre-Revenue Scaling Cost per MW$0.7-0.8mIncurred 1-2 months before revenue recognition; primarily for equipment installation, testing, and site preparation

Impact & implications

The report’s findings offer critical insights into valuing AI infrastructure companies. It demonstrates that for hyper-growth Neoclouds like CoreWeave, investors should not apply mature-cloud profitability benchmarks but instead focus on unit economics health and long-term contract IRRs. If new contracts continue meeting similar underwriting standards, CoreWeave could achieve structural margin recovery post-CY26. Additionally, monetization potential from software stacks (e.g., Mission Control) within NVIDIA’s reference architecture provides an optionality to further enhance gross margins.

Risks

• High customer concentration—changes from one or a few large clients could significantly impact revenue
• Intense competition from well-capitalized large tech firms and peer Neoclouds
• Aggressive financial leverage (via SPV financing) magnifies risk if demand falls short or financing conditions deteriorate
• Delays in capacity build-out or delivery could postpone the timeline for margin recovery

What to watch

• Whether new contract underwriting maintains ~25% contribution margin standards
• If actual CY26 pre-revenue scaling costs align with the $0.7-0.8m/MW estimate
• Commercialization progress of software products like Mission Control within NVIDIA’s reference architecture
• CY27 capacity ramp pace and whether non-GAAP operating margin rebounds as expected to the 15% range