- HGX H100 16-GPU: $6.16/GPU-hr (2-week to 1-year)
- HGX H100 64-GPU: $5.85/GPU-hr
- HGX H100 256-GPU: $5.54/GPU-hr
- HGX B200 16-GPU: $9.86/GPU-hr
- HGX B200 64-GPU: $9.36/GPU-hr
- HGX B200 256+ GPU: $8.87/GPU-hr
- 1-year+ reservation (both H100 and B200): Custom — Talk to our team
1-Click Clusters are invoiced on separate terms from On-Demand Cloud — weekly increments under reservation terms, with explicit GPU count and duration negotiated at approval. The 256+ tier delivers 32% off the on-demand 8x H100 rate ($3.99/GPU-hr x 1.55x volume multiplier); the 16-GPU reservation tier delivers 54% off on-demand because Lambda amortizes the deployment cost across a fixed GPU count.
Storage, snapshot, and inter-region transfer are billed separately. There is no free egress fee on Lambda regardless of tier — a meaningful differentiator versus hyperscalers ($0.087-$0.12/GB on AWS/GCP/Azure).
How Lambda bills: per-minute, no setup fees, no hidden charges
Lambda's billing model is one of the cleanest in the GPU cloud market.
On-Demand Cloud instances bill by the minute in 1-minute increments, starting after launch and health checks and ending at termination. Running instances accrue charges whether or not they are actively used — so shutting down idle instances saves real money. There is no setup fee. There are no commitment minimums on On-Demand Cloud (1-Click Clusters require 2-week minimum).
Storage is metered separately and billed per GB-month for attached volumes. Snapshot fees apply for stored images. Inter-region transfer is metered by the GB. Egress to the internet is free. Lambda publishes its full billing structure on docs.lambda.ai and surfaces the line items on the dashboard — no surprise charges.
Side-by-side Lambda vs RunPod vs CoreWeave pricing
| GPU / SKU | Lambda | RunPod Community | RunPod Secure | CoreWeave |
|---|
| Cheapest published | Quadro RTX 6000 $0.69 | RTX A6000 $0.33 | L40S $0.99 | L40 $1.25 |
| A100 80GB SXM | $2.79 (8x) | $1.39 | $1.59 | $2.70 (8x) |
| H100 PCIe 1x | $3.29 | $1.99 | $2.89 | Not bookable single |
| H100 SXM 1x | $4.29 | $2.69 | $3.29 | Not bookable single |
| H100 SXM 8x | $3.99 (8x) | $2.69 (8x pod) | $3.29 (8x pod) | $6.16 on-demand / $2.46 spot |
| B200 SXM 8x | $6.69 | $5.98 | $6.79 | $8.60 on-demand / $4.26 spot |
| Egress | Free | Free outbound | Free outbound | Free outbound |
| Billing | Per-minute | Per-second | Per-second | Per-hour |
How to choose between Lambda's pricing tiers in 2026
The decision hinges on workload duration, scale, and capacity urgency.
- Short experiments (hours to a few days): Lambda On-Demand Cloud. Per-minute billing, transparent card, instant setup. H100 PCIe at $3.29/GPU-hr is the right starting point.
- Multi-week fine-tuning runs: Lambda On-Demand Cloud. Pay only for what you use; shut down idle instances. For workloads that benefit from NVLink, use H100 SXM 8x at $3.99/GPU-hr.
- Multi-month training (4+ GPUs, 4+ weeks): Lambda 1-Click Cluster reservation. $6.16/GPU-hr at 16-GPU down to $5.54 at 256-GPU scale delivers 25-50% off on-demand equivalent. Requires sales engagement and 2-week minimum.
- Frontier Blackwell capacity: Lambda 1-Click Cluster HGX B200 reservation. $9.86/GPU-hr at 16-GPU down to $8.87 at 256-GPU scale. On-demand B200 is available at $6.69/GPU-hr 8x for short jobs.
- Edge AI / inference / production serving: Lambda On-Demand Cloud. Run inference workloads on H100 PCIe or A100 PCIe. Per-minute billing keeps the bill tied to active serving.
What enterprise buyers should do next
Three actions for organizations evaluating Lambda in 2026.
- Get a quote on 1-Click Cluster reservation before committing to On-Demand Cloud for a long job. The published 16-GPU HGX H100 reserved rate of $6.16/GPU-hr is a floor — sales-quoted contracts for 1-year terms typically deliver additional 10-20% off for committed volume. The 256+ GPU tier at $5.54/GPU-hr is the headline floor; sales can sometimes do better.
- Model capacity waitlists into your timeline. Lambda's H100 SXM capacity is intermittent in 2026. For workloads with hard deadlines, plan to wait 1-3 weeks for capacity, or reserve capacity in advance via 1-Click Cluster.
- Take advantage of free egress. For data-intensive workloads (training on 10+ TB datasets), Lambda's no-egress policy saves $870-$1,200/month versus AWS on the same workload. Lambda + S3-compatible storage is a meaningful TCO advantage.
What to watch next
Three near-term datapoints. First, Lambda's B300-class pricing — the homepage highlights GB300, B300, B200, H200 infrastructure but the On-Demand Cloud card has not yet listed B300 rates. Expect B300 on-demand at $7.89/GPU-hr or higher by Q4 2026. Second, NVIDIA Vera Rubin production ramps at Lambda's partner hyperscalers (CoreWeave, Google, Azure, OCI, Nebius) will relieve Hopper demand pressure; Lambda's H100 rates likely hold through Q1 2027 before easing. Third, Lambda's pricing for H200 on On-Demand Cloud — currently 'not listed' for H200 on the Lambda card; expect $5.50-$6.50/GPU-hr range when listed (lambda.ai/pricing, July 2026).