CRWV earnings call summary

Prepared Highlights

• Q2 2025 revenue was $1.2 billion, up 107% year over year (CEO, CFO).
• Adjusted operating income (AOI) was $200 million; AOI margin 16% (CEO, CFO).
• Adjusted EBITDA was $753 million, up 3x year over year; margin 62% (CFO).
• Net loss was $291 million; adjusted net loss $131 million (CFO).
• Interest expense was $267 million in Q2, up from $67 million in Q2 2024 (CFO).
• Capital expenditures (CapEx) in Q2 totaled $2.9 billion (CFO).
• Ended Q2 with $2.1 billion in cash, cash equivalents and restricted cash (CFO).
• Contracted backlog was $30.1 billion, up $4 billion from Q1 and doubled year to date (CEO).
• Active power at quarter-end: nearly 470 megawatts; total contracted power 2.2 gigawatts (CEO).
• On track to deliver over 900 megawatts of active power before end of 2025 (CEO).
• Proposed acquisition of Core Scientific would add ~1.3 gigawatts gross power capacity, with $500 million in fully ramped annual run-rate cost savings by 2027 (CEO).
• Since beginning of 2024, secured over $25 billion of debt and equity (CFO).
• Q2 2025 stock-based compensation: $145 million (CFO).

Official Guidance (CFO):

• Q3 2025: Revenue $1.26B–$1.30B; AOI $160M–$190M; interest expense $350M–$390M; CapEx $2.9B–$3.4B.
• Full-year 2025: Revenue $5.15B–$5.35B (raised by $250M from prior); AOI $800M–$830M (unchanged); CapEx $20B–$23B (unchanged).

Mgmt Quotes:

• “Revenue grew a better than expected 107% year over year to $1.2 billion for the second quarter with adjusted operating income of $200 million.” (CEO)
• “We are now on track to deliver over 900 megawatts of active power before the end of the year.” (CEO)
• “The acquisition would drive the immediate elimination of more than $10 billion in future lease liability overhead, as well as a more streamlined and efficient operating model.” (CEO)
• “We are still operating in a structurally supply-constrained environment where demand far outstrips supply for our products and services.” (CFO)
• “For the second quarter in a row, we are raising our full-year revenue guidance.” (CFO)

Prepared Metrics

Q&A Batch (1-5 of 5)

Q1 — Nitin Agrawal

• Topic: 技术故障导致问答无效
• Key points: 无有效内容（仅听到“Keith, we can't hear you.”，随后Deborah Crawford请求转接）
• Mgmt stance: 无

Q2 — Kash Rangan

• Topic: 超大规模客户合同续约与资产回报率提升
• Key points:
  • 公司不关注“续约”，而是关注“扩张”，因为客户通常会在新硬件架构发布时购买同级别最新基础设施。
  • 已收购Weights & Biases（向上游堆栈增值），并通过收购向下游垂直整合以持续降低成本，预期2025年全面实现5亿美元（$500 million） 的成本节省。
  • 通过快速部署缩短客户上线时间，并在业务各环节保持成本纪律。
• Mgmt stance: bullish，因为公司通过扩张策略、收购和成本控制持续提升资本回报率。

Q3 — Keith Weiss

• Topic: 工作负载混合（训练vs推理）与供给侧瓶颈
• Key points:
  • 基础设施设计为在训练和推理之间无缝切换（“AI基础设施”而非“训练/推理基础设施”）。
  • 观察到推理工作量大幅增加，通过电力消耗模式（训练为阶跃变化，推理为增量变化）可监控；链式推理（chain of reasoning）进一步推动推理消耗。
  • 当前最严重的供给侧瓶颈是电力外壳（powered shells），其次是电网电力、GPU供应链、中压变压器等多环节约束。
• Mgmt stance: bullish，因推理（AI货币化）使用场景扩大，但承认市场结构性供应受限。

Q4 — Mark Murphy

• Topic: 主权政府需求与两笔超大规模客户扩展合同
• Key points:
  • 多个主权国家正寻求建设AI数据中心，公司对通过最佳技术解决方案扩大主权云足迹有信心；已与加拿大Cohere合作，并在欧洲扩展基础设施。
  • 两笔超大规模客户扩展合同：一笔在Q2签署，已反映在Q2的剩余履约义务（RPO）中；另一笔在Q3签署，将在Q3财报中体现。合同内容均为GPU计算，包含服务组合元素。
• Mgmt stance: neutral–bullish，对主权市场扩张持乐观态度，但承认部分司法管辖区对美国技术接受度较低。

Q5 — Raimo Lenschow

• Topic: 供需不平衡的持续时间与推理延迟对数据中心位置的影响
• Key points:
  • 公司从未动摇对市场结构性供应受限的判断，该判断基于与全球最大消费者的直接关系。
  • 关于延迟：链式推理查询对延迟不敏感，计算能力比距离更重要；其他工作负载下延迟更重要。公司策略是将基础设施靠近人口中心以保留低延迟选项，但同时预见远程区域对延迟不敏感工作负载的需求。
• Mgmt stance: bullish，坚定认为市场供需失衡将持续，且基础设施布局策略具备弹性。

Q&A Batch (6-10 of 10)

Q6 — Brad Zelnick

• Topic: Inference vs. training economics and flexible capacity products
• Key points:
  • Economics of inference and training consumption are identical for CoreWeave’s business model; overwhelming majority of infrastructure is sold in long-term structured contracts.
  • On-demand compute is a very small percentage of overall workloads; short-term pricing spikes for inference occur when new models are released.
  • Older GPU generations (A100s, H100s) are being recontracted on term for inference loads.
  • Flexible capacity (on-demand/spot) products are coming online in the back half; a spot product is in customer preview, but capacity is immediately consumed by existing/new clients due to demand constraints.
• Mgmt stance: Neutral — economics are stable, but building on-demand/spot capacity is challenging in a demand-constrained market.

Q7 — Michael Turrin

• Topic: Backlog variability and demand pipeline
• Key points:
  • Backlog is ~$30 billion, nearly 2x year ago, fairly consistent with last quarter when adding OpenAI expansion.
  • Demand from largest clients is expanding in scale and magnitude; contracts in pipeline are “extremely significant” and will move the needle.
  • Contracts are heavily negotiated and take significant time to move through the cycle.
  • Expect continued step functions in compute as large clients take large blocks over long periods.
• Mgmt stance: Bullish — demand is broad and large-scale, clients like the product and will continue buying as long as infrastructure is built.

Q8 — Gregg Moskowitz

• Topic: Repurposing older GPU clusters off contract
• Key points:
  • Infrastructure coming off contract (H100s, A100s) is being recontracted for additional term (1–3 years), largely for inference.
  • OpenAI contract was contracted for 5 years with two additional 1-year extensions, providing transparency on older-generation compute run-out.
• Mgmt stance: Bullish — excited about recontracting older GPUs for inference, with multi-year extensions.

Q9 — Tyler Radke

• Topic: CapEx timing and cost increases ahead of contract ramp
• Key points:
  • CapEx ramp is backloaded to Q4; building 400+ additional megawatts of power, followed by CapEx spend when power is available, then revenue.
  • Q2 CapEx was lighter than expected; ramp is systematic and comfortable for delivering 900+ megawatts.
  • Costs include data center lease expenses coming online before revenue generation, creating a timing mismatch at unprecedented scale.
  • Full-year revenue guide was raised, but operating income left unchanged due to these upfront costs.
• Mgmt stance: Neutral — comfortable with backloaded ramp, but costs create short-term margin pressure.

Q10 — Brad Sills

• Topic: End-market segments (AI labs, enterprises) and cost of debt
• Key points:
  • Demand is broad-based: massive labs dominate, but green shoots in sectorial growth (e.g., Moon Valley, financial players like big banks, Hippocratic AI).
  • Weights & Biases brings 1,600 new clients (e.g., British Telecom), forming relationships with enterprise clients experimenting with AI.
  • Cost of debt improved significantly: new delayed draw term loan at SOFR + 400; non-investment grade borrowing costs dropped by 900 basis points.
  • Three debt deals (two high-yield, one DDTL) have changed economics embedded in contracts.
• Mgmt stance: Bullish — excited about capital market progress and broadening compute consumption, though large clients still dominate pipeline.