CoreWeave and Nebius financial reports reveal AI cloud computing power trends: supply shortages continue, CSP is moving towards an “AI infrastructure operating system”
Comparing news, analyst Qinbafrank said in an article on the X platform that CoreWeave (CRWV) and Nebius's latest financial reports show that the AI cloud computing power (CSP) industry is entering a stage of rapid expansion, and the focus of industry competition is shifting from simply providing GPU leasing to building an AI infrastructure platform covering computing power, software, data, and operational capabilities.
Currently, demand for AI computing power still clearly exceeds short-term deliverable supply. At the same time, AI computing power pricing power is being strengthened, but price increases are mainly focused on high-value resources. CoreWeave said that the price of various GPU computing SKUs generally increased by about 25% in July; Nebius revealed that the price of previous-generation GPUs increased by more than 30% compared to the first quarter, the average annualized revenue of new contracts in the second quarter exceeded 20 million US dollars per MW, some projects reached 20 million to 25 million US dollars, and short-term emergency capacity prices even reached 40 million to 50 million US dollars per MW.
However, the price increase mainly occurred in short-term capacity, next-generation GPUs, large-scale clusters, and production-level AI inference scenarios. The raw computing power of traditional low priority and long-term price locking did not rise at the same time. Judging from the profit model, the project-level return on AI computing power is becoming more clear, but the company's overall return on capital (ROIC) still needs time to be verified. Nebius revealed a more clear project payback cycle for the first time, and CoreWeave reduced the pressure on GPU investment through long-term contracts and asset-level financing. However, the two companies are currently still in a phase of high capital investment, and depreciation and financing costs continue to reduce profit margins.
However, more and more individual projects can close the economic model, which means that the AI infrastructure business model is gradually maturing. Additionally, both CoreWeave and Nebius are upgrading to an “AI infrastructure operating system.” In the future, CSP competition will not only rent GPU hours, but will cover complete service systems such as AI training, inference, storage, networking, model deployment, monitoring, security governance, and agent operation environments.
In terms of capital models, the two companies are also taking different paths: Nebius favors an asset-light model, investing in the construction of AI data centers through capital partners to provide AI infrastructure operations and software capabilities; CoreWeave promotes a hybrid cloud model through the Omni strategy, deploying a complete AI cloud platform to the customer's own data center and GPU resources, and places more emphasis on enterprise-level and sovereign AI delivery.
Overall, the AI cloud computing power industry is evolving from a “GPU renter” to an “AI infrastructure platform.” Short-term supply constraints will still support computing power prices, while long-term competitive focus will shift to capital efficiency, software capabilities, and whether it can become an infrastructure operating system in the AI era.
