Lambda Cloud review

Lambda Cloud provides GPU infrastructure for training AI models and serving inference. Its homepage lists self-serve GPU instances, production-ready clusters, managed superclusters, and co-engineering support. That puts it in a different category from a general-purpose cloud where GPUs are one option among many: the product is organized around AI compute and larger GPU deployments.

The choices range from individual instances for testing and prototyping to clusters built for distributed workloads. Lambda also offers managed clusters and single-tenant superclusters, including NVIDIA GB300 NVL72 systems with Quantum-2 InfiniBand. Its listed cluster options include HGX B200 and H100 configurations. Lambda describes its infrastructure as single-tenant and shared-nothing, and lists SOC 2 Type II certification and hardware-level isolation. In practical terms, the decision is less about whether Lambda offers GPUs and more about which deployment scale, GPU generation, and level of management your workload needs.

Lambda’s range gives teams a path from smaller GPU jobs to production-scale training without treating every workload as a one-off cloud build. The product lineup includes self-serve instances, 1-Click Clusters, managed clusters, and superclusters. The homepage describes 1-Click Clusters as production-ready HGX B200 and H100 systems for distributed AI workloads; managed options may suit teams that want Lambda to manage the cluster rather than focus their own staff on infrastructure.

There is also useful hardware detail in the pricing list, rather than just a starting price. For example, its self-serve HGX B200 listing is $7.69 per GPU-hour and specifies 180 GB VRAM per GPU, 208 vCPUs, 2,900 GiB RAM, and 22 TiB SSD. The HGX H100 listing is $4.59 per GPU-hour, with 80 GB VRAM per GPU. Other listed options include A100, A6000, A10, GH200, and V100 GPUs, so buyers can compare older and newer hardware against workload needs.

For organizations with security requirements, Lambda lists single-tenant infrastructure, shared-nothing architecture, hardware-level isolation, and SOC 2 Type II certification. Those are concrete differentiators to assess against your own compliance requirements.

The main trade-off is cost visibility. Lambda prices GPU capacity by the hour, not with a published monthly or annual rate. The pricing data lists no free plan, and trial details are not published. Usage totals can add up quickly, so compare the hourly rate against expected runtime and include applicable sales tax, VAT, or GST.

Cluster pricing also comes with scale and term commitments. The listed production cluster options start at 16 GPUs and cover terms from two weeks to one year; larger price tiers are listed at 64 and 256 or more GPUs. The 16-GPU B200 cluster is $9.86 per GPU-hour, while the comparable H100 option is $6.16. Reserved capacity for 16 or more GPUs on terms of one year or longer has no published price. These cluster terms are distinct from self-serve instance listings, which include smaller configurations.

There is a pricing-date caveat, too: the published pricing information says displayed instance prices take effect November 16, 2026, and current prices are available after login until then. Confirm the rate before budgeting or switching workloads.

Lambda lists GPU-hour pricing, not monthly or annual plans. Self-serve rates range from $0.69/GPU-hour for a Quadro RTX 6000 to $7.99/GPU-hour for a two-GPU B200 SXM6 listing. HGX instances are $1.99/GPU-hour for A100 (40 GB), $4.59 for H100 and $7.69 for B200. Cluster rates are higher: a 16-GPU H100 cluster is $6.16/GPU-hour, while a 16-GPU B200 cluster is $9.86. Larger cluster listings lower the per-GPU rate: 256 H100 GPUs cost $5.54 each per hour, and 256+ B200 GPUs cost $8.87. Reserved capacity for 16+ GPUs over a year has no published price.

There’s no free plan or trial listed. Applicable sales tax, VAT or GST may apply. The pricing page says listed instance rates take effect November 16, 2026; until then, current rates are available after login. Budget by GPU-hours and cluster size, not a monthly sticker price.

Switching is likely more involved than moving a general-purpose app between hosts: Lambda sells GPU instances and managed or production-ready clusters for AI workloads. Before committing, check that your deployment scripts, container images, storage and networking assumptions fit the specific instance or cluster you’ll use. The published material lists co-engineering support, but doesn’t specify migration services, data-export tools, API compatibility or transfer fees. Treat those as questions for Lambda, not established features.

Leaving means planning where model files, datasets and workload configuration will go, and checking any transfer costs and access requirements before ending capacity. The published material doesn’t document export limits or a standard offboarding process. It also includes no Reddit posts or comments, so we can’t substantiate a claim about why users leave. For a switch decision, get written answers on data retrieval, cluster reservations and contract end dates—especially for 1-year-plus capacity, whose price isn’t published.