Lambda GPU pricing explained: what Lambda Cloud charges in 2026

18 October 2026

This page explains Lambda's GPU cloud pricing for people choosing where to run a job. Every figure comes from Lambda's own pages, read on 18 October 2026, with the link beside it. Lambda changes its prices, so treat this as a dated snapshot and confirm on the Lambda pricing page before you commit. We do not have a referral link with Lambda.

What Lambda lists for on demand instances

Lambda's pricing page shows one price per GPU per hour for each GPU type, plus applicable sales tax, VAT or GST. These were the listed rates on 18 October 2026:

NVIDIA B200 SXM6, 180 GB of memory per GPU: $6.69 per GPU per hour.

NVIDIA H100 SXM, 80 GB: $3.99 per GPU per hour.

NVIDIA A100 SXM, 80 GB: $2.79 per GPU per hour.

NVIDIA A100 SXM, 40 GB: $1.99 per GPU per hour.

NVIDIA Tesla V100, 16 GB: $0.79 per GPU per hour.

Check the memory size when you compare. The two A100 rows are different products: the lower price is the 40 GB version, not the 80 GB one. The page shows buttons for 1, 2, 4 and 8 GPUs but lists one rate per GPU, not a separate price for each count.

Working out a job cost from the listed rate

The cost of a job is the rate, times the number of GPUs, times the hours the instance is running. As an illustration using the H100 SXM rate above, one GPU for 100 hours is 100 x $3.99, which is $399. Eight GPUs for 10 hours is 8 x 10 x $3.99, which is $319.20. Add tax where it applies, plus storage and any time the machine sits idle. You can also use the ComputeClarity calculator to compare stored rates across providers, each with the date it was collected.

1-Click Clusters

For larger jobs, Lambda lists HGX B200 1-Click Clusters with durations from two weeks to a year. On 18 October 2026 the page showed $9.86 per GPU per hour at 16 GPUs, $9.36 at 64 GPUs and $8.87 at 256 or more. A one year or longer option is listed from 16 GPUs, but no price was shown for it. Lambda asks you to contact it for reserved capacity.

How it bills

Lambda's pricing page lists rates per GPU per hour. It does not state a billing increment, so check how a short job is billed before relying on it.

Storage is billed separately. Lambda's documentation says filesystems are billed per GB used per month, in increments of one hour, and that if you delete an instance but not its filesystem, you keep being billed for the filesystem. Delete or move data you no longer need.

Egress

Lambda's pricing page description says there are no egress fees. We did not see that statement in the visible pricing text or in the on demand documentation we read. If you move a lot of data out, confirm the egress terms with Lambda before you commit.

Regions

Lambda's documentation lists 14 regions: Tokyo and Osaka in Japan, India, Germany, Israel, and US regions in Virginia, Washington DC, Illinois, Texas, California, Arizona and Utah. It notes that each instance is tied to one region and that not every instance type is available in every region. If your data must stay in a particular country or region, check availability for the GPU you need.

Security

Lambda's trust page states a SOC 2 Type II attestation and ISO 27001, 27017, 27701 and 22301 certification. These are Lambda's own statements. Ask Lambda for the reports if a customer asks you for evidence.

What to check before you start

Match the GPU and its memory to your workload, because rows with the same GPU name can differ in memory. Check that the region you need has the instance type available. Work out your storage and data movement, not just the hourly rate. Run a short test job first, and keep a second provider as a fallback if capacity is tight.

For how Lambda compares with another specialist cloud, read Nebius vs Lambda Labs. For other options when Lambda is full, read Lambda Labs alternatives, and for the wider picture see GPU cloud providers compared.