Execution delivered its best quarter, but high capital intensity and long-term competitive pressure still support the Underperform view

Overview

The report believes CoreWeave delivered its strongest quarter of execution since going public. Revenue, adjusted operating profit, adjusted EBITDA, backlog, and power capacity all made positive progress, and management also broadly raised near-term operating guidance. However, Bernstein believes these improvements mainly strengthen the credibility of the company achieving its 2026 targets and are not yet sufficient to eliminate concerns about the long-term business model. The company still needs to rely on large-scale capital investment and financing expansion, while future increases in data center supply, direct competition from hyperscale cloud providers, contract pricing, and changes in GPU residual values could cause the growth trajectory to weaken rapidly.

Core views

First, the second quarter validated the company's ability to convert contracted capacity into revenue, with near-term demand visibility and deployment efficiency clearly improving. Second, operating leverage is beginning to emerge, but capital expenditure grew even faster over the same period, and interest expense and balance sheet pressure remain core constraints. Third, managed inference could broaden the customer base and absorb expiring GPU capacity, but the business is still in its early stages, and its margins, retention rate, utilization rate, and incremental capital requirements have not been validated. Fourth, the company has value in the current environment of constrained data center supply, but Bernstein expects supply to become more abundant after 2028, and hyperscale cloud providers will intensify competition with scale, cost, and software capabilities. Fifth, the firm's 2026 forecasts are broadly close to consensus expectations, but become more cautious starting in 2027, with its 2028 revenue forecast about 15% below consensus expectations.

Analysis framework

The report combines quarterly actuals with the firm's forecasts, consensus expectations, and company guidance to conduct variance analysis, and evaluates execution quality from backlog conversion, power capacity deployment, margins, capital expenditure, interest expense, and financing conditions; the long-term assessment is based on changes in data center supply and demand, competition from hyperscale cloud providers, customer concentration and potential coopetition, contract economics, GPU residual values, and new product maturity. Valuation uses a forward enterprise value multiple method.

Methodology notes

• Valuation methodsForward enterprise value to adjusted operating profit multiple method

  Estimate enterprise value based on forward adjusted operating profit, then derive the equity price target.

  The $74 price target is based on a 25.5x next-year-plus-one-period adjusted operating profit multiple, using adjusted operating profit of $5.2 billion; the previous multiple was 28.4x.

• Performance analysisVariance analysis of actuals, firm forecasts, and consensus expectations

  Compare quarterly actual performance with forecast deviations to assess the degree of outperformance and operating inflection points.

  The report focuses on comparing revenue, adjusted operating profit, margins, adjusted EBITDA, capital expenditure, and net loss, and analyzes near-term forecast changes after the company raised guidance.

• Industry analysisSupply-demand and competitive scenario analysis

  Assess the impact of changes in data center capacity supply and competitive structure on long-term revenue and pricing.

  The base-case view is that current capacity shortages still support CoreWeave, but after 2028 supply will loosen, and hyperscale cloud providers and new entrants with scale and cost advantages may weaken its competitive position.

• Financial riskCapital intensity and debt-servicing capacity analysis

  Evaluate balance sheet pressure by combining capital expenditure, financing scale, interest expense, and profit generation capacity.

  The report believes lower financing costs are positive, but capital needs and the debt service burden are rising at the same time, and the pace of cost reduction is still insufficient to offset the capital-intensive risks of the business model.

Asset mapping & comparison

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

• CoreWeave, Inc. (CRWV.US)

  Primary research subject

  Strengths

  Strong demand for AI infrastructure, a large backlog, accelerating capacity deployment, emerging operating leverage, improved economics of newly signed contracts, and managed inference creating a potential new growth driver.

  Weaknesses

  The business model is capital-intensive, the company continues to incur losses with a heavy interest burden, growth depends on financing and large customer contracts, and long-term pricing, contract economics, and GPU residual values are uncertain.

  Comparison

  Currently benefits from data center capacity shortages, but compared with hyperscale cloud providers, it may be disadvantaged in scale, cost of capital, software ecosystem, and enterprise customer coverage.

  Risks

  Looser supply, competition from hyperscale cloud providers, construction and delivery delays, financing costs, customer coopetition, margin volatility, and insufficient maturity of the managed inference business.

• META.US

  An important customer and potential competitor of CoreWeave

  Strengths

  Possesses significant capital investment capacity, scale advantages, and the ability to build its own AI infrastructure.

  Weaknesses

  Large-scale self-built capacity requires a long construction cycle and faces power and supply chain constraints.

  Comparison

  Meta's transition from customer to potential competitor reflects the long-term risk that CoreWeave customers may gradually internalize compute infrastructure.

  Risks

  If Meta accelerates self-built capacity, it could reduce demand for external procurement and compress the bargaining power of new cloud service providers.

• S&P 500

  Bernstein's U.S. equity rating benchmark

  Strengths

  Represents overall performance of the U.S. large-cap equity market and can be used to measure relative returns.

  Weaknesses

  Cannot directly reflect the high capital intensity and idiosyncratic risks of AI infrastructure companies.

  Comparison

  Bernstein's Underperform rating means it expects CRWV to lag the S&P 500 by more than 15 percentage points over the next 12 months.

  Risks

  If valuations in the AI infrastructure sector continue to expand, CRWV could significantly outperform the broad market benchmark for a period.

Key data

• Second-quarter revenue$2.575 billionUp 112.3% year over year and approximately 24% quarter over quarter, above the $2.563 billion consensus expectation.
• Adjusted operating profit$128 millionMargin was 5%, above the $67 million consensus expectation and above the previous guidance range of $30 million to $90 million.
• Adjusted EBITDA$1.510 billionUp 100.5% year over year, with a margin of 58.6%.
• Backlog$104.0 billionUp 46% year over year; approximately $129.0 billion after including more than $25.0 billion of new commitments added early in the third quarter.
• Active power capacity1.5 GWNearly 500 MW added in the second quarter, and the year-end target was raised from more than 1.7 GW to more than 1.85 GW.
• Contracted power capacity3.7 GW at quarter-end, approximately 4.2 GW at the earnings callThere is also more than 1.5 GW of potential capacity from powered land, expansion options, and signed letters of intent.
• Second-quarter capital expenditure$9.352 billionAbove the high end of guidance; construction in progress increased quarter over quarter from $9.6 billion to $11.9 billion.
• Full-year revenue guidance$12.4 billion to $13.2 billionPrevious guidance was $12.0 billion to $13.0 billion.
• Full-year capital expenditure guidance$35.0 billion to $39.0 billionPrevious guidance was $31.0 billion to $35.0 billion, and the capital expenditure increase was larger than that of key operating metrics.
• Third-quarter revenue guidance$3.45 billion to $3.60 billionThe midpoint was $3.525 billion, above consensus expectations at the time.
• Second-quarter net loss and net interest expenseNet loss of $626 million; net interest expense of $640 millionThe net loss narrowed from $740 million in the first quarter, but was still more than double the prior-year period.
• Managed inference booked annual recurring revenueMore than $100 millionRapidly increased from about $1 million at launch; the company's target is at least $250 million by the end of 2026.
• Valuation and price target25.5x forward enterprise value to adjusted operating profit multiple; $74 price targetThe price target was raised from $67, but the valuation multiple was lowered from 28.4x; it implies about 18% downside versus the $90.32 closing price.

Impact & implications

In the short term, strong orders, capacity ramp-up, and margin improvement are expected to drive the market to raise 2026 revenue and operating profit expectations, while managed inference provides an additional growth option. In the medium to long term, revenue growth remains highly dependent on continuing to sign large contracts, building on schedule, and obtaining low-cost financing; once data center capacity shortages ease, pricing, contract returns, and asset utilization may come under pressure simultaneously. The report therefore only increases confidence in the company's ability to achieve near-term targets and does not change its cautious view on long-term fundamentals and the stock's relative performance.

Risks

• Data center capacity may be released faster after 2028, potentially weakening CoreWeave's scarcity premium and pricing power.
• Hyperscale cloud providers may directly compete for enterprise customers with scale, capital, and software ecosystems, potentially limiting renewals and additions of large contracts.
• Capital expenditure and debt service needs continue to rise, and the decline in financing costs may be insufficient to offset the interest burden.
• The deployment complexity of power, land, equipment, networking, and cooling systems may cause construction delays or pre-revenue cost drag.
• The pricing, contribution margins, and GPU residual values after contract expiration for long-term contracts may be lower than current valuation assumptions.
• The managed inference business remains in an early stage, and its margins, customer retention, utilization stability, and incremental capital requirements have not yet been validated.
• If hyperscale cloud providers' self-built capacity is delayed, power shortages persist, or CoreWeave's software platform develops stronger differentiation, the company may continue to maintain pricing power, creating upside risk to the Underperform rating and price target.

What to watch

• Whether third-quarter revenue, adjusted operating profit, and capital expenditure can meet the latest guidance.
• Whether active power capacity can exceed 1.85 GW by the end of 2026 and continue progressing toward the 8 GW target for 2030.
• The actual start of construction, commissioning, and revenue conversion speed of the $104.0 billion backlog and new commitments added in the third quarter.
• The actual margin impact of a 5 to 10 percentage point improvement in contribution margins on newly signed contracts and an approximately 25% product price increase in July.
• The coverage relationship between approximately $3.3 billion of expected 2026 interest expense and approximately $7.4 billion of adjusted EBITDA.
• Whether managed inference annual recurring revenue can reach at least $250 million by the end of 2026 and demonstrate stable utilization and profitability.
• Whether the customer mix continues to diversify, and the impact of customers such as Meta becoming competitors on contract renewals and incremental demand.
• Whether the divergence between the company's revenue and consensus expectations from 2027 to 2028 begins to emerge.