Bernstein maintains its Underperform view on CRWV, with the core disagreement centered on long-term revenue rather than near-term execution

Overview

This report is Bernstein's post-initiation investor feedback update on CoreWeave, Inc. (CRWV). The report says the team spoke with more than 70 investors, company contacts, and industry contacts after launching coverage, and the market feedback fell into several camps: bulls believe neocloud demand will not slow, bears agree with the bearish logic but care about timing, cloud optimists are focused on whether neocloud can replicate traditional cloud margins, balance sheet watchers question the financing structure, and other investors are focused on whether NVDA will provide a backstop and whether CRWV's issues represent the entire neocloud space. Bernstein's conclusion is to maintain its original view: it does not expect CRWV to stall immediately in the near term, but over the long run it is more likely to be pressured by hyperscaler scale, supply coming online, and customer bargaining power.

Core views

The report's core view is that CRWV's Underperform rating is mainly a structural revenue judgment, not a denial of the company's near-term execution ability. Bernstein expects the company to meet guidance, long-term operating margins could reach 25%-30%, the blended cost of capital could fall to around 7%, and it could still add about $35B in new contracts by the end of 2027. But the firm believes that as hyperscaler self-built supply comes online around 2028, the pace of signing large hyperscale contracts for CRWV will slow; at the same time, hyperscalers such as Microsoft, Google, Amazon, and Oracle have stronger demand visibility, network advantages, capital strength, and internal build options, and may ultimately compete directly for enterprise GPUaaS demand.

Analysis framework

The report uses an investor-debate decomposition approach, responding point by point to the main arguments from bulls, bears, cloud optimists, balance sheet watchers, the NVDA backstop camp, and the neocloud comparison camp. On valuation, it uses an EV/Adjusted EBIT multiple method, arriving at a $56 target price using 2027E adjusted EBIT per share of $5.31 and a 28.4x multiple. On industry judgment, the report focuses on AI power demand, hyperscaler self-build capacity, GPUaaS investment, data center supply timing, customer concentration, debt structure, and NVDA's ecosystem investment incentives.

Methodology notes

• Valuation methodsEV/Adjusted EBIT multiple method

  Assess CRWV using enterprise value relative to adjusted EBIT multiple.

  The report uses a 28.4x EV/Adj. EBIT multiple and applies it to 2027E adjusted EBIT per share of $5.31 to derive a $56 target price.

• Scenario and consensus comparisonRevenue forecast divergence analysis

  Compare Bernstein's forecast with market consensus for 2026, 2027, and 2028.

  The report says 2026 expectations are broadly in line with consensus, divergence begins in 2027, and the 2028 revenue forecast is more than 30% below consensus, reflecting its structural bearish view on revenue.

• Industry supply analysishyperscaler self-build and leased capacity tracking

  Judge the durability of neocloud demand through AI power demand, data center supply, and hyperscaler self-build timing.

  The report estimates global AI power demand at about 60GW by 2030 and notes that hyperscalers have announced roughly 30GW of self-built capacity by 2030, of which about 23GW could come online before 2028, plus another roughly 70GW built by third parties and potentially leased to hyperscalers.

• Business model comparisonneocloud versus traditional cloud comparison

  Compare the scale, diversification, and vertical integration differences between GPU-specific cloud providers and traditional clouds such as AWS, Azure, and GCP.

  The report argues that neoclouds provide GPU clusters, scheduling, orchestration, and performance optimization, but lack the ability of traditional clouds to spread costs across services, customers, and workloads, making it difficult to achieve traditional cloud long-term margins.

Asset mapping & comparison

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

• CRWV

  Core coverage name

  Strengths

  Strong near-term demand, existing large contracts, relatively mature GPUaaS platform, potential to add about $35B in new contracts, and long-term operating margins could reach 25%-30%.

  Weaknesses

  Customer concentration, capital intensity, dependence on hyperscaler demand, lack of traditional cloud scale and diversification advantages, and long-term revenue may come in below consensus.

  Comparison

  Compared with AWS, Azure, and GCP, CRWV is more focused on GPU workloads but lacks the ability of traditional clouds to spread costs across services and customers; compared with other neoclouds, CRWV is more mature, but NBIS, IREN, and others are also winning large contracts.

  Risks

  If hyperscaler self-build capacity is delayed, CRWV's software platform builds a stronger moat, and data center power shortages persist, that could support CRWV's pricing power and create upside risk to the Underperform rating.

• NVDA

  Ecosystem investor and potential backstop

  Strengths

  Has strong cash generation, invests in multiple neoclouds such as CRWV, NBIS, Lambda, and Crusoe, and helps expand the GPU ecosystem and distribution channels.

  Weaknesses

  The report believes NVDA has no inherent incentive to support CRWV indefinitely, and losing its equity stake in CRWV would have limited impact on NVDA.

  Comparison

  By investing in multiple neoclouds rather than a single winner, NVDA cultivates the ecosystem and can also increase its bargaining leverage against hyperscalers.

  Risks

  If the market relies too heavily on the assumption of an NVDA backstop, CRWV's investment case could come under pressure if NVDA's support proves weaker than expected.

• NBIS

  Comparable neocloud and demand diversifier

  Strengths

  The report notes that NBIS recently won an approximately $27B contract from Meta, showing that other neoclouds can also absorb hyperscaler demand.

  Weaknesses

  It also faces neocloud business-model, financing, and customer-concentration risks.

  Comparison

  NBIS is seen as an alternative supply source that could divert some of CRWV's existing hyperscaler demand.

  Risks

  If more hyperscalers choose multiple neoclouds to diversify procurement, CRWV's share and bargaining power could come under pressure.

• IREN

  Comparable neocloud

  Strengths

  The report notes that IREN has a $9.7B contract with Microsoft and says colleagues are more positive on IREN.

  Weaknesses

  It is still subject to the broader neocloud business-model debate.

  Comparison

  Compared with CRWV, IREN may offer a different risk-reward profile for investors.

  Risks

  Intensifying industry competition could lower long-term returns across the entire neocloud space.

Key data

• RatingUnderperformBernstein maintains an Underperform rating on CRWV.
• Target price$56Based on 28.4x EV/Adj. EBIT and 2027E adjusted EBIT per share of $5.31.
• Investor feedback sample70+The report says it spoke with more than 70 investors, company contacts, and industry contacts after initiating coverage.
• Expected new contractsAbout $35BBernstein expects CRWV could still sign new contracts from the report date through the end of 2027.
• Long-term operating margin assumption25%-30%The report believes CRWV could reach this operating margin range in the long term.
• Blended cost of capital assumptionAbout 7%The report believes CRWV's overall cost of capital could fall to around 7%.
• 2028 revenue forecast gapMore than 30% below consensusBernstein says divergence from consensus begins in 2027 and widens in 2028.
• AI power demand modelAbout 60GW by 2030The report acknowledges significant AI demand but believes efficiency gains and supply release will affect neocloud long-term pricing.
• Hyperscaler self-build capacityAbout 30GW by 2030, with about 23GW online by 2028This is a key basis for the report's view that CRWV hyperscale contract growth may slow.
• Other supply capacityAbout 70GWThe report says another roughly 70GW of capacity will be built by other parties and may be leased to hyperscalers.
• NVDA cash$63BThe report believes NVDA can support CRWV, but the investment case should not be built on indefinite backstop assumptions.
• NVDA ownership11% CRWV ownershipThe report says NVDA owns about 11% of CRWV.

Impact & implications

For investors, the report suggests that CRWV's trading risk may not come from near-term execution misses, but rather from a future slowdown in large contract signing, the release of hyperscaler supply, and an early reversal in market sentiment. Near-term AI demand, contract backlog, and NVDA ecosystem investment may still support share-price momentum, but Bernstein believes these factors are not enough to offset the long-term business model risk. For the AI infrastructure chain, the report positions neoclouds as an important but possibly transitional form of supply, whose long-term value depends on whether they can maintain differentiation, utilization, and financing-cost advantages amid hyperscaler self-build and bargaining pressure.

Risks

• A delay in hyperscaler self-build capacity coming online could extend CRWV's value proposition and pricing power.
• If CRWV's dedicated software platform continues to strengthen, it could sustain demand for its operating data centers.
• A persistent shortage of data center power supply could support CRWV's pricing and contracting ability.
• If AI demand is significantly stronger than Bernstein's forecast, it could delay or soften the revenue inflection lower.
• If NVDA chooses to support CRWV over the long term, it could ease financing and ecosystem risks.
• Sentiment momentum may continue to support the stock before the fundamental divergence appears.

What to watch

• Whether hyperscaler GPUaaS products begin to be marketed and invested in more aggressively.
• The actual commissioning pace of roughly 23GW of hyperscaler self-build capacity around 2028.
• Whether CRWV adds roughly $35B in new contracts by the end of 2027, as the report expects.
• Changes in renewals, pricing, and contract duration between CRWV and large hyperscaler customers.
• The pace at which other neoclouds such as NBIS and IREN win large contracts.
• Whether CRWV's cost of capital really falls to around 7%, and whether the debt structure shifts risk to equity.
• Whether the divergence between revenue expectations and market consensus widens in 2027 and 2028.
• Changes in NVDA's investment, procurement, or support behavior toward CRWV and other neoclouds.