Ngành điện toán đám mây AI ghi nhận nhu cầu tăng cao và sự thay đổi về giá cả

The Huoxing Finance report states that on August 13, analyst qinbafrank identified five key trends in the AI cloud computing power (CSP) sector based on the latest financial reports from CoreWeave and Nebius. Demand continues to significantly outstrip near-term deliverable supply—CoreWeave had approximately 1.5 GW of power operational by the end of Q2, with contracted power of about 3.7 GW, and a backlog of roughly $104 billion in orders, excluding over $2.5 billion in new commitments added at the beginning of Q3; Nebius plans to deploy over 1 GW annually starting in 2027, with a medium-term target of 5 GW in contracted capacity. The industry’s bottleneck has shifted from “whether there are customers” to “when power can be delivered, GPUs installed, and billing initiated.” Pricing power is indeed strengthening, but not uniformly: CoreWeave raised prices across most SKUs by approximately 25% in July; Nebius’s older-generation GPU prices rose over 30% compared to Q1, with average annualized revenue from new Q2 contracts exceeding $20 million per MW, and short-term emergency capacity in early Q3 reaching $40 million to $50 million per MW. The most pronounced price increases occurred in short-term capacity, next-generation GPUs, large-scale clusters, and production-grade inference—not traditional low-priority, long-term fixed-price bare compute. Project-level ROI is now calculable, but company-level ROIC has yet to be proven: Nebius disclosed its first clear project payback period, while CoreWeave covers GPU investments through five-year contracts and asset-level financing, yet depreciation and interest still consume the majority of operating profits. Both companies are transitioning toward becoming “AI infrastructure operating systems,” shifting their competitive focus from hourly GPU rentals to integrated platforms offering training, inference, storage, networking, model deployment, monitoring, governance, agent execution environments, and cross-cloud operations. Paths for third-party capital inflow are diverging: Nebius’s lightweight model more closely resembles “capital partners fund AI factories while Nebius provides the operating system and operational capabilities”; CoreWeave Omni more closely resembles “deploying a full cloud onto customers’ own data centers and GPUs.” While the direction is the same, Nebius currently emphasizes reducing capital expenditure, whereas CoreWeave emphasizes hybrid cloud and sovereign AI delivery.