Lambda Labs review: Rent NVIDIA GPUs by the hour: instances, clusters and Superclusters for AI teams

What Lambda Labs does

Lambda is a cloud provider that rents NVIDIA GPU compute to teams that train and serve AI models. Capacity comes in three forms. Instances are self-serve machines with B200, H100, A100, GH200 and other GPUs, and the vendor says they can be deployed in minutes on a first-come basis. Superclusters run on single-tenant NVIDIA GB300 NVL72 systems with Quantum-2 InfiniBand networking.

Billing is per GPU per hour rather than a monthly subscription, and listed prices exclude sales tax/VAT/GST.

Larger counts cost less per GPU in the table. The tables stop at 256+ GPUs, and terms of 1 year or more require talking to the sales team. The pricing page also states there are no egress fees.

For security, Lambda cites a single-tenant, shared-nothing architecture and SOC 2 Type II certification. It also offers managed clusters and co-engineering help.

• Self-serve GPU instances launched in minutes
• 1-Click Clusters™ of B200 or H100 GPUs
• Single-tenant Superclusters on NVIDIA GB300 NVL72
• Quantum-2 InfiniBand networking for Superclusters
• Managed clusters handled by the vendor
• SOC 2 Type II certification
• No egress fees
• Reserved capacity via the sales team

Our take

Lambda fits teams that want to rent NVIDIA GPUs by the hour, from a single self-serve instance to larger clusters, and that care about single-tenant isolation and SOC 2 Type II. Cluster prices depend on GPU count and term, and terms of 1 year or more or Superclusters require talking to sales. Listed prices also exclude sales tax/VAT/GST. This summary comes from reading the official pages, not from using the service.