CoreWeave's Quarter and Guidance
CoreWeave (CRWV) delivered $2.58 billion in revenue, up more than 100% year over year. The company sits at what its leadership calls an important inflection point. The gain comes from accelerating operating leverage: the business is reaching large scale, and that scale compounds into profitability, margin expansion, and operating leverage for clients too. CoreWeave's clients are some of the leading AI companies, and the value they get from the platform shows up in their own growth rates across the sector.
Full-year guidance was raised to between $12.4 billion and $13.2 billion. Q3 guidance implies roughly 150%-plus year-over-year growth at the midpoint. The confidence to sustain that pace without hitting physical capacity limits rests on a track record: more than 51 data centers in operation, eight years in the business, and over 500 megawatts of active power brought online in the last quarter alone.
Backlog and Demand Durability
The backlog stands at $104 billion, plus more than $25 billion of new commitments signed in Q3 that are not counted in that figure. Inference is the economic driver of AI. After years spent deep in training - helping clients get their models to a point where they could monetize their business - inference demand is now a straight line up, inflecting higher each quarter.
Contracts being signed today support an "AI loop": clients run training, fine-tuning, and inference, then repeat the cycle as they gather new information to keep improving their models. The platform is built to enable that process and keep clients on the frontier.
Enterprise vs. AI Lab Contracts
Demand is spreading past frontier labs to names like Caterpillar (CAT), Leidos (LDOS), and Flow Traders. The main difference between hosting real-time enterprise workloads and training massive foundation models is contract structure. AI lab clients want longer contracts, five to six year committed take-or-pay agreements. Enterprises want shorter terms, roughly two to three years.
Geography and network latency differ very little between the two. Customers want to deploy in North America - the US and Canada. There is strong demand in Europe, and CoreWeave recently announced its first data center expansion into the APAC region. Enterprises want much the same infrastructure that AI labs run; the difference is contract duration.
Power and Building Constraints
CoreWeave has 1.5 gigawatts of active power and added nearly 500 megawatts in the quarter. The hard part is less about power itself and more about physically delivering a data center or powered shell. The company does two things: it leases data center capacity from trusted partners, who build and deliver the powered shell (CoreWeave then installs its GPU and cloud infrastructure inside), and it also builds some sites directly, which adds operating leverage across the portfolio.
Accessing electrons is not the main issue. The real limits are whether there are enough electricians and enough components to build out the data centers. These supply chain challenges are not new; navigating them for years, plus 51 data centers in operation, gives the partnerships and experience to deliver on time.
CoreWeave does not market capacity to clients until the data center and its power are secured in the portfolio. It does not sign GPU contracts and then hope to find capacity in a constrained, hard-to-navigate market - the process is flipped to avoid disappointing clients. On power type, the focus is grid power. There is little behind-the-meter or self-build generation right now, no push into co-located nuclear, geothermal, or micro grids. Grid power is seen as the right resource, and there is plenty of availability across the US. For context, the IEA forecasts global data center electricity use could double by 2030.
Local Opposition and Site Selection
Local opposition to data centers is rising, including a NIMBY movement and the governor of New York signing an executive order for a moratorium on large-scale data center construction. Regulatory pushback makes some conversations harder. But that opposition does not change AI demand. It may change where data centers get built, not whether they get built. Demand for compute is insatiable with no near-term way to satisfy it, so the location may shift while the need to build in the US stays.
Competitive Position
CoreWeave competes with Microsoft (MSFT), Amazon (AMZN), and Google (GOOGL), all with far larger balance sheets, while Meta (META) and others expand their own infrastructure. The claimed edge: the best technology solution in the market, recognized by clients, third-party consultants, and suppliers. The reason much larger companies have not delivered a better product comes down to focus - CoreWeave delivers this one product in its best available format, and has done so for years.
