Lambda is a GPU cloud built for training and serving AI models. It rents NVIDIA hardware, including B200, H100, A100 and GH200 GPUs, through three routes: self-serve instances launched in minutes, 1-Click Clusters ranging from 16 to more than 2,000 GPUs, and Superclusters for the largest workloads. Instance pricing is per GPU per hour, starting at $0.69 for a Quadro RTX 6000 and reaching $6.99 for a B200 SXM6, with H100 SXM instances from $3.99.
The audience runs from individual machine learning engineers who need a single GPU for a weekend experiment to research labs and enterprises training foundation models. Companies with strict security needs can choose single-tenant, dedicated infrastructure with SOC 2 Type II compliance, and teams without much operations staff can hand cluster management to Lambda, whose engineers also co-engineer workloads. Security-focused buyers get single-tenant environments, and launching an instance takes only an account and a few clicks.
Typical scenarios include fine-tuning a model on a few A100 or H100 GPUs, launching a 64-GPU B200 cluster for a multi-week pretraining run, and serving billions of tokens on dedicated hardware for production inference. Reserved clusters last from two weeks to a year, with longer commitments negotiated directly with the sales team at lower prices. Cluster pricing on the H100 runs from $5.54 to $6.16 per GPU hour depending on size, and B200 clusters from $8.87 to $9.86. Organizations building complete AI factories can also get high-density power, liquid cooling and NVIDIA GPUs delivered as one system.
