# Neoclouds Cloud providers built specifically to rent GPU capacity for AI, as distinct from the general-purpose [[Hyperscalers]] (AWS, Azure, Google Cloud, Oracle). Examples: CoreWeave, Nebius, Lambda, Crusoe, Together AI, and many regional players. ## How they differ from hyperscalers - **Focus.** GPUs, high-speed networking ([[InfiniBand]]) and AI software, rather than hundreds of general services. - **Speed.** Often first to deploy new Nvidia generations at scale. - **Contracts.** Many sell large reserved blocks of capacity to a few big customers, including frontier labs and hyperscalers themselves. - **Balance sheet.** Typically financed with debt secured on GPUs and long-term customer contracts, which makes utilisation and contract quality the key risks. ## Why they matter in the chessboard When [[Open Weights]] make the model itself free, margin moves to whoever runs the tokens ([[Token Economics of Inference]]). Neoclouds and hyperscalers are the natural winners of that shift. Neoclouds and GPU rental in third countries (Malaysia, Thailand, Japan) are also how compute reaches labs that cannot buy chips directly under [[AI Chip Export Controls]]. Related: [[AI Inference Infrastructure]], [[Open Source Hyperscaler MoC]], [[Data Center MoC]]