Analysis
The Numbers
CoreWeave posted a comfortable beat against Wall Street expectations, according to CNBC and Bloomberg, a move CNBC characterized as the market rewarding a "cleaner quarter" than investors had braced for:
- Q2 revenue -- $2.6 billion, up 112% from $1.2 billion a year earlier
- Adjusted per-share loss -- $1.03, narrower than the $1.20 loss analysts had modeled
- FY26 revenue guidance -- raised to $12.4 billion-$13.2 billion, backed by a backlog past $104 billion
- Q3 revenue guidance -- $3.4 billion-$3.6 billion
- Stock move -- up 14% after-hours Tuesday, indicated as much as 18% higher premarket Wednesday
Who Signed What
CoreWeave disclosed three specific customer commitments alongside earnings rather than folding them into a single backlog figure:
- Meta -- committed an additional $21 billion in spending on top of its existing CoreWeave relationship
- Anthropic -- signed a new multi-year compute agreement, its latest in a string of infrastructure deals that also includes Riot Platforms' $9.1 billion, 20-year deal this same week
- Jane Street -- the quantitative trading firm committed $6 billion, a sign CoreWeave's customer base now extends beyond AI labs and hyperscalers into finance
Company Background
CoreWeave started as a cryptocurrency-mining operation before pivoting to GPU cloud infrastructure, went public on Nasdaq in March 2025, and has built its business almost entirely around renting out Nvidia GPU capacity to AI labs and enterprises that don't want to build their own data centers. Pulse previously covered CoreWeave's push into classified federal AI cloud work through a Leidos partnership, part of a broader effort to diversify beyond a small number of AI-lab customers that has historically concentrated the bulk of its revenue.
The Competitive Field
CoreWeave competes against the cloud divisions of Amazon, Microsoft and Google, all of which can subsidize GPU capacity with far larger balance sheets, and against smaller "neocloud" rivals like Lambda and Crusoe that are chasing the same GPU-rental business model without CoreWeave's public-market scale or its now-enormous backlog. What differentiates CoreWeave from the hyperscalers isn't infrastructure -- it's speed and specialization: CoreWeave built its entire stack around AI workloads from the start, while AWS, Azure and Google Cloud are retrofitting decades-old general-purpose infrastructure for GPU-heavy demand.
Numbers in Context
A 112% revenue jump against a backlog above $104 billion means CoreWeave now has more contracted future revenue than nearly ten years of its current run rate -- a backlog-to-revenue ratio few public infrastructure companies carry. That backlog is also the number investors are leaning on to justify CoreWeave's valuation despite the company still posting a loss: the bet is that once fully delivered, the contracted capacity converts to revenue at a predictable pace, the same logic bond investors are applying to Nvidia's $500 billion financing plan this week.
The Counterweight
The headline growth obscures a less comfortable detail: operating expenses more than doubled year-over-year and marginally exceeded revenue, meaning CoreWeave is still not profitable even as its topline more than doubles. A $104 billion backlog is also announced, contracted capacity, not delivered and collected revenue -- it depends on customers including Meta, Anthropic and Microsoft continuing to draw down committed capacity at the pace CoreWeave has guided to, and on CoreWeave successfully building and powering the data centers to fulfill those contracts on schedule. Seeking Alpha's own earnings preview flagged rising debt and margin pressure as the real risk sitting underneath the growth headline, a tension that a single strong quarter doesn't resolve.
What This Means for AI-Infrastructure Investors
CoreWeave's results are one of the cleanest public-market proxies available for how much AI compute demand is actually converting into paid contracts rather than remaining speculative capex talk. The Jane Street commitment in particular is worth watching -- a quant trading firm buying GPU capacity at this scale suggests AI compute demand is starting to broaden beyond model training and consumer chatbots into finance workloads that don't carry the same public narrative as a ChatGPT or Gemini launch.
