What Lambda currently offers
Lambda's public cloud portfolio currently separates on-demand instances from larger 1-Click Clusters. The instance product is designed for self-service access to Linux-based GPU virtual machines, with current public options including NVIDIA B200, H100, GH200, A100, A10, A6000 and older accelerators. The provider states that instances can be launched through its interface, API or CLI and that supported configurations range from one to eight GPUs.
This structure matters because a team experimenting with a model, fine-tuning a workload or serving an early production endpoint can have a very different procurement need from an organization planning hundreds of GPUs for sustained training. Lambda's current 1-Click Cluster offer covers 16 to 2,000+ H100 or B200 GPUs, with published pricing tiers for shorter commitments and contact-based pricing for longer-duration capacity.
For AIDataCenterHQ's current GPU Cloud Decision Cluster, Lambda therefore intersects directly with the H100 and B200 sourcing guides. Lambda's current public self-service and cluster pricing pages do not list H200 as a current priced offering, so this page does not infer an H200 price or availability state.
Primary sources: Lambda AI cloud pricing and Lambda GPU instances, checked 25 September 2026.
Lambda self-service GPU instance pricing
Lambda currently publishes per-GPU-hour pricing for one-, two-, four- and eight-GPU instance sizes. The visible rate varies by accelerator and, for H100 and B200, by GPU count. At the eight-GPU level, the current public table lists B200 SXM6 at $6.69 per GPU-hour and H100 SXM at $3.99 per GPU-hour. At one GPU, B200 is currently $6.99 per GPU-hour, H100 SXM is $4.29, and H100 PCIe is $3.29.
| Current public instance | GPU count | VRAM/GPU | Price/GPU/hour | Context |
|---|---|---|---|---|
| NVIDIA B200 SXM6 | 8 | 180 GB | $6.69 | Largest currently listed self-service B200 instance. |
| NVIDIA H100 SXM | 8 | 80 GB | $3.99 | Eight-GPU Hopper instance. |
| NVIDIA B200 SXM6 | 1 | 180 GB | $6.99 | Single-GPU self-service option. |
| NVIDIA H100 SXM | 1 | 80 GB | $4.29 | Single-GPU SXM option. |
| NVIDIA H100 PCIe | 1 | 80 GB | $3.29 | Different hardware form factor and system context. |
The instance page states that billing is by the minute and advertises no egress fees. Those terms can be commercially significant for iterative development or data-intensive workflows, but the buyer should still verify storage, regional availability, quotas and current capacity before assuming a workload can run at the listed configuration and price.
Source: Lambda instances. Prices are public observations, exclusive of applicable taxes as stated by Lambda, and may change.
Lambda 1-Click Cluster pricing changes with scale and duration
For larger deployments, Lambda currently publishes separate H100 and B200 1-Click Cluster tiers. The current table lists B200 at $9.86 per GPU-hour for 16 GPUs, $9.36 for 64 GPUs and $8.87 for 256+ GPUs for durations from two weeks to one year. H100 is listed at $6.16, $5.85 and $5.54 per GPU-hour at 16, 64 and 256 GPUs respectively for the same published duration band.
| Cluster system | GPU count | Published duration | Price/GPU/hour | Commercial interpretation |
|---|---|---|---|---|
| HGX B200 | 16 | 2 weeks–1 year | $9.86 | Entry published cluster tier. |
| HGX B200 | 256+ | 2 weeks–1 year | $8.87 | Lower published unit rate at larger scale. |
| H100 | 16 | 2 weeks–1 year | $6.16 | Entry published H100 cluster tier. |
| H100 | 256 | 2 weeks–1 year | $5.54 | Lower published H100 unit rate at scale. |
It would be incorrect to interpret the cluster table as evidence that Lambda's larger committed product is always cheaper than its on-demand instances. The products are not identical. Cluster pricing reflects a different capacity model, topology, procurement path and commitment structure. The correct comparison is between equivalent workload and availability requirements.
Primary source: Lambda AI cloud pricing, checked 25 September 2026.
How should Lambda pricing be normalized?
The first normalization step is to keep product type intact. A one-GPU H100 instance, an eight-GPU H100 instance and a 256-GPU H100 cluster are three different commercial objects even though all contain the same accelerator family. Record GPU model, GPU count, VRAM, vCPUs, system RAM, local storage, geography, billing unit, commitment duration, taxes, timestamp and availability status before calculating any derived metric.
The second step is to model utilization. A lower nominal hourly rate can become expensive if capacity is reserved but underused. Conversely, a higher flexible rate can be economically sensible for bursty experiments if it avoids paying for idle committed capacity. Teams should compare cost per completed workload, cost per useful GPU-hour, training elapsed time and inference throughput where those metrics can be measured.
Lambda's published pricing also illustrates why per-GPU-hour is not the final answer. The current eight-GPU H100 instance lists more host CPU, memory and local storage than the one-GPU configuration. That surrounding system affects data preparation, checkpointing and multi-GPU operation. Use the GPU Cloud Pricing Comparator for normalized price observations and the AI Infrastructure ROI Calculator to test workload-level economics.
Where Lambda can fit different AI workloads
Self-service instances can suit teams that need fast access for experimentation, fine-tuning, benchmarking, development and production workloads that fit within one to eight GPUs. The current instance catalogue provides both H100 and B200 options, making it possible to test whether Blackwell's larger memory and architecture deliver enough workload improvement to justify its higher visible hourly price.
1-Click Clusters address a different scale. Lambda describes them as production-ready clusters spanning 16 to 2,000+ B200 or H100 GPUs. Large distributed training programs may therefore evaluate Lambda alongside other AI-focused providers with dedicated multi-node infrastructure. At that scale, networking topology, scheduling, storage throughput, failure handling and capacity certainty become as important as the accelerator specification.
H100 can remain attractive where software, model size and budget align with Hopper infrastructure. B200 may offer additional value for memory-intensive or newer-generation workloads, but procurement should be based on measured throughput and total workload cost rather than architecture generation alone. For broader accelerator context, see the Technology hub.
Lambda compared with the broader GPU-cloud market
Lambda's combination of self-service instances and large 1-Click Clusters creates a different comparison profile from a provider that focuses mainly on fixed multi-GPU nodes or a marketplace that emphasizes very granular rental supply. CoreWeave, for example, publishes system-level HGX prices for several accelerators. RunPod exposes a different mix of Pods and Serverless products. Crusoe has its own AI-cloud infrastructure and commercial model.
The purpose of the GPU Cloud Providers hub is to preserve those distinctions. A future direct CoreWeave-versus-Lambda page should compare equivalent deployment scenarios rather than place unrelated public price numbers side by side. No single provider is inherently suitable for every workload.
Constraints and questions to resolve before choosing Lambda
Self-service access is described as first-come, which means a visible instance type should not be treated as guaranteed capacity in every region at every moment. Teams with fixed start dates or large sustained needs should confirm availability and reservation options. Lambda's documentation also notes that instances are tied to geographical regions, which makes regional availability a practical procurement field rather than a cosmetic detail.
For large cluster commitments, buyers should clarify duration, cancellation and expansion terms, network architecture, storage, support, security requirements, data movement and operational responsibilities. Pricing tables do not answer all of these questions. They are input data for procurement, not substitutes for technical diligence.
Finally, compare cloud rental with owned or colocated infrastructure when utilization is high and predictable. The economics can change materially with financing, power, cooling, staffing, depreciation and refresh cycles. AIDataCenterHQ's Data Centers, Power and Cooling sections provide that wider infrastructure context.
Methodology, freshness and limitations
This page uses Lambda's public pricing, instances page and public cloud documentation as the primary evidence for current configurations, billing structure and list prices. Facts were checked on 25 September 2026 and should be treated as fast-changing F2 information. AIDataCenterHQ does not infer private discounts, future capacity, unlisted regions, negotiated terms or unsupported hardware availability.
Each consequential price observation should retain source, product type, GPU count, region where applicable, configuration, billing unit, commitment assumption and timestamp. Current public pricing is not a guarantee of real-time inventory. Before procurement, recheck Lambda's current pages and verify capacity directly.
Related GPU cloud pages
Adjacent technology, IP and commercialization resources
Lambda procurement can intersect with AI infrastructure contracting, accelerator commercialization, patent strategy and cross-border deployment issues. Relevant specialist resources include Patent Business Lawyer, GIP Research, GIPResearch.org, Tech Law Attorney, US Tech Law Attorney, Patent Business Attorney, International Patents, TechCorpLegal and Advocate Rahul Dev.
Frequently asked questions
Does Lambda currently publish H100 and B200 pricing?
Yes. Lambda currently publishes self-service instance prices for H100 and B200 and separate 1-Click Cluster pricing for H100 and B200 deployments.
Does Lambda offer single-GPU H100 and B200 instances?
Yes. Its current public instance table lists one-GPU H100 SXM, H100 PCIe and B200 SXM6 configurations, alongside larger multi-GPU options.
Is Lambda's cluster price directly comparable with its instance price?
No. The cluster and instance products differ in scale, topology, commitment and procurement model. Compare equivalent workload requirements rather than the visible rate alone.
Does Lambda currently publish H200 pricing?
Not on the current public pricing and instance tables reviewed for this page. This page therefore does not state a current Lambda H200 list price.