NVIDIA DGX Cloud

NVIDIA DGX Cloud is NVIDIA's managed GPU platform for building and running AI at scale. NVIDIA describes it as a fully managed AI training platform. It runs on GPU clusters that NVIDIA co-engineers with AWS, Google Cloud, Microsoft Azure and Oracle Cloud Infrastructure, and customers get access to NVIDIA experts. NVIDIA also uses it to build its own models, such as Nemotron, and its physical AI work such as Cosmos and Isaac GR00T. Alongside it sits DGX Cloud Lepton, launched in May 2025 after NVIDIA bought Lepton AI. Lepton is a marketplace that routes developers to GPU capacity from partners such as Lambda, CoreWeave, Crusoe and Nebius, and it has since added AWS and Azure. Its documentation covers dev pods, endpoints, batch jobs, storage, region selection, a CLI and an API. Prebuilt NIM microservices and serverless endpoints are offered through build.nvidia.com. NVIDIA does not publish prices for DGX Cloud. You request a private offer through a cloud marketplace, with flexible term lengths, and the Google Cloud, Azure and Oracle listings advertise free trials. On Lepton, each partner sets and bills its own rate. A July 2026 third-party survey put H100 rates at roughly $3.85 to $12.29 per GPU-hour, depending on the provider. The product suits enterprises and well-funded AI teams that need large multi-node GPU clusters with NVIDIA's software stack and support. It is not a general web hosting service. Its main limitation is that pricing is opaque and negotiated. Press reports say NVIDIA has scaled back DGX Cloud as a direct rival to the hyperscalers in favor of the Lepton marketplace model, which raises questions about the product's long-term direction.