Citi said in its latest report dated Aug. 13 that Nebius's main growth constraint is not demand but the timing of capacity coming online, while maintaining a buy/high risk rating and a $278 target price. The bank said second-quarter revenue was strong, helped by asset SLA revenue, Token Factory, Tavily, higher utilization and on-demand demand. Management said demand remains very strong, with multiple buyers behind each GPU (graphics processing unit, a key AI chip), and that backlog growth is being driven by larger average order sizes and more medium-term contracts in its core AI cloud business. Citi said neoclouds (cloud providers renting GPU capacity for AI workloads) have become a new focal point in the recent U.S. AI trade after Nvidia's $500 billion AI infrastructure financing plan, CoreWeave's earnings and Lumentum's strong guidance, as investors look for which companies can secure GPUs, finance buildouts and convert computing power into revenue. Nebius also stands out because about 50%-60% of its infrastructure capital spending is backed by customer prepayments, allowing project payback on a cash basis in less than a year, at about 10 months. Citi said the bigger test comes in the second half, with most signed capacity expected to go live in the second half of 2026 and Microsoft-related deployments following a similar timetable. Management said the 800MW to 1GW connected power target remains achievable, though converting that into active power and recognized revenue will take network testing, integration and commissioning. It also said the $7 billion to $9 billion ARR (annual recurring revenue) framework does not depend on a single project, but on utilization, pricing and capacity growth together, with most of the 5GW of signed capacity expected to ramp over the next two to three years. Citi still warned that Nebius is a high-risk name because of customer concentration, heavy capital spending, and uncertainty around GPU supply and financing conditions, leaving the stock highly sensitive to delivery timing even if the AI compute story remains intact.