CoreWeave Q2 FY '26

TLDR

CoreWeave is still growing at an extreme pace: revenue increased 112% YoY, adjusted EBITDA margins reached 59%, and backlog climbed 246% to $104.2 billion. Guidance suggests that growth is accelerating again, with Q3 revenue of roughly $3.5 billion and FY26 guidance implying about $4.6 billion of revenue and $676 million of adjusted operating income in Q4.

 

More importantly, the quarter reinforces the strategic case for neoclouds: AI infrastructure increasingly requires purpose-built power, liquid cooling, networking, orchestration, and cluster optimization rather than simply adding GPUs to a traditional cloud. CoreWeave’s ability to be first to validate NVIDIA’s Vera Rubin NVL72 is a good example of that advantage. The key question for CRWV is increasingly not demand, but how quickly it can deploy capacity against a $104 billion backlog while preserving attractive economics.

 

Overview

• They surpassed 1 billion model training runs tracked on our platform
• They recently signed an A100 contract that extends into 2029 at an attractive price. As a reminder, this SKU was introduced in 2020. Clusters of prior generations of architecture offer installed, energized, production-grade compute already running at scale. They come with a proven ROI for customers.
• In Q2, they became the first cloud provider to bring up and validate NVIDIA’s Vera Rubin NVL72, leveraging innovations in software-defined liquid cooling and rack management to extend their track record of being first to market.
• Neoclouds have a strategic advantage of traditional cloud providers
  • "The continuous AI life cycle cannot be supported by simply adding GPUs to a general purpose cloud. It requires a new approach from power, cooling, and rack design through networking, orchestration, observability, developer tools, and managed services. That is why CoreWeave is purpose-built for AI. Our platform is singular in its depth, breadth, and technical capability. In Q2, we became the first cloud provider to bring up and validate NVIDIA’s Vera Rubin NVL72, leveraging our innovations in software-defined liquid cooling and rack management to extend our track record of being first to market."
    • It would be great to more color on this from management but I think they're primarily talking about
      • The GPU virtualization tax is the hidden performance penalty and economic loss that occurs when running AI workloads inside virtual machines rather than directly on bare-metal silicon
      • East-West vs. North-South Traditional cloud networking is optimized for "North-South" traffic (data moving in and out of the internet to a server). Hyperscalers often rely on heavily virtualized Ethernet networks or custom interconnects (like AWS's EFA).
      • The NVIDIA NVL72 rack that CoreWeave mentions is a perfect example of why legacy data centers struggle. The newest AI racks, like the Blackwell-based NVL72, can draw over 120kW per rack. You cannot air-cool this much heat; it requires complex liquid cooling piped directly to the chips. Hyperscalers have to expensively retrofit older facilities to support this, while specialized providers build their data centers from the concrete up specifically to handle massive power density and fluid dynamics.

 

Financials

They have several good slides here covering the quarter.

 

The good: revenue, growth, demand numbers are great.

 

The bad: leverage. I don't get how people are so comfortable with it. The interest coverage ratio is negative? In Q2, they had an operating loss of $49 million and interest expense of $640 million. EBIT is negative $689 million.

 

Metrics

• Revenue grew 112% YoY
• Adjusted EBITDA is at a 59% margin
• Revenue backlog is $104.2 billion, up 246% YoY, driven by continued diversification from enterprise and AI-natives customers

​

Guidance

Their guidance is through the roof. Q3 revenue is projected to grow at 40% compared to Q2. Operating income is projected to grow to $230 and $676 mln in Q3 and Q4 '26, respectively.