CoreWeave Inc (CRWV): UBS starts CoreWeave at Buy, arguing growth and GPU monetization outweigh credit concerns

Overview

UBS initiates coverage of CoreWeave with a Buy rating and US$120 target. Its central view is that the market is heavily discounting leverage, financing and long-term profitability risks, while underestimating capacity execution, AI-compute demand, rising GPU pricing and the resulting increase in revenue per GW.

Core views

UBS launches coverage with a Buy rating and a US$120 price target, based on 3.8x CY28 estimated sales and roughly 40% upside from the US$85.43 price on 21 September 2026. The firm characterizes the call as non-consensus because sentiment toward highly leveraged AI-infrastructure providers is cautious. CoreWeave traded at about 3x 2027 revenue and 5.1x adjusted EBITDA despite 112% year-on-year revenue growth; UBS argues these multiples more than reflect concerns around debt, capacity delivery, margins, customer concentration and competition. Its US$120 target uses a multiple below the 4.5x peer-group median because of CoreWeave's higher leverage, interest burden, concentration and early-stage move into higher-margin services. Capacity is the first analytical pillar. Revenue depends on CoreWeave reaching more than 8 GW of active critical-IT-load power by 2030, and UBS believes it can. Active power reached 1.5 GW in 2Q26, nearly 500 MW higher sequentially, and the company guided to more than 1.85 GW by year-end 2026. It had 4.2 GW of incremental contracted power pending build-out as of the mid-August 2Q26 call. UBS finds little evidence of systemic go-live bottlenecks from industry capacity additions, contractor discussions and partner projects, although it recognizes risks from GPUs, electrical equipment, labor, permitting and community opposition. The report distinguishes critical IT load from gross facility load, which can be 10-25% larger, and notes that active power may take about six weeks to become billable while GPUs are installed and tested. The capacity plan is capital-intensive. UBS estimates US$35bn-40bn of capex per GW, compared with a US$57.2bn total cost for a GB200/GB300-based 1 GW data center before considering the use of partner-built shells. CoreWeave spent US$9.4bn in 2Q26 and guided to as much as US$13.5bn for 3Q26; UBS estimates US$62bn of 2024-26 spending to achieve 1.85 GW by year-end 2026. For 2027-30, UBS models US$260bn of capex, above roughly US$230bn Street consensus, as the company increasingly self-builds facilities. Lease obligations were US$16.5bn at 2Q26, with another US$35.5bn not yet commenced. UBS views these obligations as a directional indicator of future build-out. UBS's positive revenue thesis rests on monetization per GW. CoreWeave generated US$2.575bn in 2Q26 revenue, which annualizes to about US$10.3bn; using 925 MW of prior-quarter average online power implies roughly US$11bn of revenue per GW. UBS expects this to move toward US$15bn or more by the end of 2027, driven by GPU price inflation, mix shifting toward Blackwell and Vera Rubin systems, new contract pricing and a larger share of short-duration contracts. CoreWeave said new contracts carried a 25% uplift across SKUs in 2Q26, while new deals had contribution margins 5-10 points above prior-quarter contracts. UBS notes that high-90% of capacity remains in multi-year contracts, so repricing occurs mostly through new capacity and expirations; short-term contracts remain a small share but have reportedly reached as much as US$40m per MW, or US$40bn per GW. That pricing and capacity framework supports UBS forecasts above consensus. CoreWeave reported 112% revenue growth and 246% backlog growth in 2Q26, with US$104bn of committed backlog, or US$129bn including more than US$25bn of early-3Q26 new commitments. UBS estimates US$28.4bn of 2027 revenue versus US$26.7bn consensus and about US$85bn in 2030 versus roughly US$80bn consensus. The report argues that 40% of 2Q26 backlog, or US$41.6bn, is expected to convert to revenue over the following 24 months. CoreWeave's 2027 revenue growth is modeled at 122%, above consensus at 107%, supported by expected active-power growth of about 70% and approximately 35% higher revenue per GW. Profitability remains a key debate. Non-GAAP gross margin was 17% in 2Q26 versus 30% a year earlier, and non-GAAP operating margin was 5%, improving from 1% in 1Q26 but below the company's 25-30% long-term target. UBS argues that newly ramping capacity temporarily suppresses margins, whereas mature five-year take-or-pay contracts are structured for stabilized mid-20% operating margins. It points to 36% incremental gross margin in 2Q26 and 5-10 points better contribution margins on recently signed contracts. CoreWeave cites a mid-30% levered IRR floor on five-year deals, with potential for 50% or more on short-duration or prepaid contracts; UBS views project-level levered IRR as more informative than point-in-time ROIC during an expansion phase. Upside also depends on extending beyond core GPU compute, which still accounts for 95% of revenue: margin-accretive storage, CPU, networking and software reached more than US$400m ARR in 2Q26, while the managed inference platform had US$100m ARR and is expected to reach US$250m exiting 2026. Credit risk is UBS's number-one concern. At 2Q26, CoreWeave had US$6.4bn of cash, US$35.1bn of debt and US$16.3bn of operating lease liabilities. UBS estimates US$158bn of operating cash flow against US$260bn of 2027-30 capex, implying roughly US$102bn of additional financing needs. The company uses delayed-draw term loans to finance about 90% of capex, backed by GPUs and long-dated take-or-pay customer contracts. Its weighted-average cost of debt declined about 300 basis points during 2025, but 2Q26 interest expense was US$640m and 3Q26 guidance implied US$900m. UBS believes credit markets remain open, although at wider spreads, supported by creditworthy customers, customer prepayments, GPU collateral, Nvidia financing and a large contracted backlog. It nevertheless highlights the B+ speculative-grade rating, widening CDS and bond spreads, and sensitivity to funding costs. Customer concentration is high but improving. UBS estimates that Microsoft, OpenAI and Meta accounted for 36%, 26% and 10%, respectively, or 72% of 2Q26 revenue. Microsoft concentration fell from 72% in 1Q25 and is no longer one of the two largest backlog customers. UBS cites expanding relationships with Meta, Google, Anthropic, Jane Street and others, as well as Nvidia's commitment to purchase residual unused capacity through April 2032. Still, the report considers enterprise adoption essential for the next stage of upside: a UBS survey of about 140 organizations found neo-clouds were cited by fewer than 10% of respondents as preferred AI-infrastructure providers. Competition is substantial: UBS profiles 17 AI-infrastructure participants, including hyperscalers, Nebius, IREN, Nscale, Lambda, Crusoe, SpaceX, Meta, SB Energy and inference providers Fireworks and Baseten. The report accepts that more supply could eventually pressure GPU pricing and returns, especially if hyperscalers add first-party capacity or major customers build their own infrastructure. But customer checks indicate CoreWeave is selected despite premium pricing because of GPU availability, deployment speed, uptime, utilization and performance on large training clusters. UBS believes its scale, roughly US$10bn revenue run-rate, US$104bn backlog and Nvidia-optimized orchestration platform provide differentiation, while acknowledging that CoreWeave trails pure-play inference platforms in inference cost optimization and needs to strengthen cybersecurity and enterprise features. UBS frames the valuation through three scenarios. Its US$120 base case assumes CY28 revenue growth of 61%, 28% gross margin, 19.9% operating margin, CY28 revenue of US$45.575bn and a 3.8x EV/sales multiple. The US$190 upside case assumes 66% growth, 30% gross margin, 21.8% operating margin, US$46.995bn of revenue and 4.8x EV/sales. The US$46 downside case assumes 56% growth, 26% gross margin, 18% operating margin, US$44.155bn of revenue and 2.7x EV/sales.

Analysis framework

UBS evaluates CoreWeave in the sequence of capacity delivery, capital required to build capacity, revenue per GW, margin and project-return economics, financing access, customer concentration, competitive positioning and valuation. The analysis combines company disclosures, peer comparisons, customer and industry checks, third-party data, contractor discussions and UBS's Enterprise AI survey.

Methodology notes

• Industry AnalysisSupply-demand framework

  AI-compute supply-demand and GPU pricing analysis

  UBS links strong demand, scarce GPU capacity, new capacity additions and contract duration to GPU pricing, revenue per GW and returns.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  AI infrastructure value-chain analysis

  The report assesses how GPU suppliers, data-center partners, cloud providers, customers and inference platforms affect CoreWeave's capacity, financing, utilization and service mix.

• Valuation methodsPS valuation

  Enterprise-value-to-sales multiple valuation

  UBS applies a 3.8x CY28 estimated sales multiple to derive its US$120 base-case price target and presents upside and downside valuation cases.

• Corporate Fundamentals and FinanceROIC–WACC spread

  Project-level levered IRR and return-on-invested-capital assessment

  UBS examines whether contract economics, customer prepayments, GPU life and pricing can support returns above the cost of capital despite heavy upfront investment.

Asset mapping & comparison

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

• CoreWeave (CRWV)

  Primary covered company; UBS initiates Buy coverage.

  Strengths

  Rapid capacity additions, US$104bn backlog, rising GPU pricing, differentiated training performance and reliability, and Nvidia support.

  Weaknesses

  Capital-intensive model, low current margins, high interest burden, concentrated revenue base and limited higher-margin service mix.

  Comparison

  Trades at about 3x 2027 revenue and 5.1x adjusted EBITDA, below peers such as DigitalOcean and Nebius according to UBS.

  Risks

  Funding access and cost, capacity delays, customer concentration, greater competition and slower move into software and inference.

• Nebius (NBIS)

  Comparable and competitor in AI cloud infrastructure.

  Strengths

  Rapid revenue growth, growing contracted-power target and higher pricing on newer and short-duration deals.

  Weaknesses

  High customer concentration and substantially smaller scale than CoreWeave.

  Comparison

  UBS describes CoreWeave's annualized revenue base as more than four times Nebius's.

  Risks

  Competitive supply and reliance on large customers.

• NVIDIA (NVDA)

  Supplier, customer, investor and capacity-utilization backstop.

  Strengths

  GPU supply relationship, financing support and residual-capacity purchase commitment support CoreWeave's scaling and financing.

  Weaknesses

  CoreWeave remains Nvidia-centric, creating supplier reliance.

  Comparison

  CoreWeave is optimized for Nvidia-based AI infrastructure rather than developing proprietary chips.

  Risks

  The competitive benefit of Nvidia access could diminish if supply constraints ease.

Key data

• Rating / price targetBuy / US$120.0012-month target; approximately 40% upside from US$85.43 on 21 Sep 2026.
• 2Q26 revenue growth112% year-on-yearRevenue was US$2.575bn, equivalent to about US$10.3bn annualized.
• 2Q26 backlog growth246% year-on-yearCommitted backlog was US$104bn; US$129bn including more than US$25bn of early-3Q26 commitments.
• Active power1.5 GW in 2Q26Nearly 500 MW sequential addition; management guided to more than 1.85 GW by year-end 2026 and targets 8+ GW by 2030.
• Revenue per GW~US$11bn currently; US$15bn+ over timeUBS expects uplift from GPU pricing, newer GPU mix and contract mix.
• 2027 revenue estimateUS$28.4bnUBS estimate versus US$26.7bn Street consensus.
• 2027-30 capex~US$260bnUBS estimate, compared with about US$230bn Street consensus.
• Debt and interest expenseUS$35.1bn debt; US$640m 2Q26 interest expense3Q26 midpoint interest-expense guidance was US$900m.
• Customer concentration72% of 2Q26 revenue from three customersUBS estimates Microsoft at 36%, OpenAI at 26% and Meta at 10%.

Impact & implications

UBS argues that CoreWeave's equity case hinges on proving that capacity can be built on schedule and monetized at higher revenue per GW while financing remains available. If those conditions hold, it believes the current valuation understates growth and margin potential; the report nevertheless treats financing, customer concentration and new supply as continuing constraints on the valuation multiple.

Risks

• Frontier labs could pace training activity or slow compute spending.
• Power, equipment, labor, permitting or community constraints could delay new capacity.
• Tighter credit markets or higher funding costs could impair the ability to finance growth.
• Customer concentration may remain elevated, and major customers could internalize infrastructure.
• Hyperscalers, neo-clouds and new entrants could increase supply and pressure GPU pricing, renewals and returns.
• Progress into higher-margin software and inference services could be slower than expected.

What to watch

• Progress toward more than 1.85 GW of active power by year-end 2026 and the 8+ GW 2030 target.
• GPU pricing, the Vera Rubin mix, revenue per GW and short-duration-contract adoption.
• Capex, customer prepayments, debt spreads, financing availability and interest expense.
• Backlog conversion, enterprise customer traction and further diversification beyond Microsoft, OpenAI and Meta.
• Competitive moves by hyperscalers, SpaceX, Meta, SB Energy and inference platforms.
• CoreWeave's upcoming customer event in San Francisco on 30 September and 1 October.