Lambda
6/10Lambda is a GPU cloud that rents NVIDIA GPU virtual machines and multi-node clusters, billed by the hour, to AI teams that train and serve models.
Data last updated November 2026 · Our methodology
Overview
Lambda is an AI infrastructure company that runs its own GPU data centers. Its public cloud has two main products. On-Demand Cloud gives you Linux virtual machines with NVIDIA GPUs, from older V100 and RTX 6000 cards up to H100, GH200 and B200. 1-Click Clusters are multi-node H100 or B200 clusters of 16 GPUs or more, reserved for two weeks to one year. Instances come with Lambda Stack, which has PyTorch, CUDA and drivers preinstalled. You can attach persistent network filesystems, set inbound rules in a console firewall, and use the Cloud API to launch, restart, list and terminate instances. The docs list 14 regions in the US, Europe, Asia and Israel. Pricing is metered per GPU per hour, and cluster rates fall as the GPU count rises. Lambda's pricing page lists new rates from November 16, 2026, for example H100 at $3.79 to $4.89 per GPU-hour and B200 at $7.69 to $7.99. Filesystem storage is billed per GB per month. A December 2025 third-party comparison reports no egress fees. Longer reservations and Superclusters are priced through sales. Lambda suits ML researchers, AI startups and enterprises that need raw GPU capacity. It is not a general web host: there is no WordPress, CDN or managed databases, and reviewers describe its Kubernetes offering as early-stage. The most common complaint is availability. Popular GPU types often sell out on demand, and SemiAnalysis also criticised the console and onboarding for being immature.
Our Rating
Hosting & Infrastructure6.1/10
AI Software5.8/10
Who Lambda is best for
Good fit
ML engineers get SSH access to GPU machines with frameworks preinstalled and an API for automation.
Hourly on-demand GPUs with no minimum commitment suit AI startups training and testing models.
Reserved clusters of hundreds of GPUs and SOC 2 Type II suit large AI programs, though reviewers flag gaps in cluster tooling.
Less suited
It offers only GPU compute and none of the general web hosting most small businesses need.
Pros & Cons
Pros
- A wide GPU range, from $0.69 per hour V100-class cards to B200, suits both small experiments and large training runs.
- A December 2025 comparison reports no egress or data transfer fees, which helps when moving large datasets in and out.
- Lambda Stack images come with PyTorch, CUDA and drivers preinstalled, so ML engineers can start training soon after launch.
- SemiAnalysis testing found full expected interconnect bandwidth and strong health checks on Slurm clusters.
- Lambda's trust page lists SOC 2 Type II, and a trust portal provides the report and pentest results on request.
Cons
- Popular GPUs often sell out on demand, according to several reviews.
- The new list prices from November 2026 are well above the H100 rates of around $2.99 reported in 2025.
- Kubernetes and Slurm need Lambda engineers to set up, and SemiAnalysis rated the console and onboarding as immature.
- There is no CDN, managed databases, serverless functions or general web hosting, and object storage was still in development.
- Lambda has no official Terraform provider or CLI documented, only community Terraform providers.
Key Features
Not included
- Managed WordPress
- Shared Hosting
Lambda Pricing
1-Click Clusters
- Reserved for 2 weeks to 1 year, from 16 GPUs
- Requires approval and payment before provisioning
- H100 at $6.16 (16 GPUs), $5.85 (64) or $5.54 (256) per GPU-hour
- B200 at $9.86 (16 GPUs), $9.36 (64) or $8.87 (256+) per GPU-hour
On-Demand Instances
- B200 at $7.69 to $7.99 per GPU-hour
- H100 at $3.79 to $4.89 per GPU-hour
- Local SSD included with each instance
- GH200 at $2.59, A100 at $1.99 to $3.19, A10 at $1.29
- Billed per GPU-hour, with prices effective November 16, 2026
- A6000 at $1.09, V100 at $0.79, RTX 6000 at $0.69 per GPU-hour
Reserved and Superclusters
- Contracts of 1 year or more through sales
- Lambda advertises its lowest prices for reserved capacity
Frequently Asked Questions
On-demand instances are charged per GPU per hour. One comparison reports per-minute billing with no minimums. Filesystems are billed per GB per month and keep billing after you delete the instance.
A December 2025 third-party comparison says Lambda charges no data transfer fees. Confirm this with Lambda before relying on it.
Lambda's trust page lists SOC 2 Type II. You can request the report through its Customer Trust Portal.
Yes. 1-Click Clusters reserve 16 or more H100 or B200 GPUs for two weeks to one year at listed rates. Contracts longer than a year are priced by sales.
The docs list 14 regions across the USA, Europe, Asia and Israel. Not every instance type is available in every region.
No free tier is listed on the pricing page.
Similar Tools
See all alternatives
Amazon's public cloud platform, with over 200 pay-as-you-go services for compute, storage, databases, networking and AI, used by startups through large enterprises and governments.
Google's public cloud platform providing pay-as-you-go virtual machines, Kubernetes, serverless, databases, storage, networking, data analytics and AI services for businesses of every size.
Northflank is a Kubernetes-based developer platform for deploying services, jobs, databases and GPU workloads on its managed cloud or on customers' own cloud accounts.