Menu

Lambda (formerly Lambda Labs) GPU prices: A100, H100, GH200, B200 per hour

We may earn commissions when you shop through the links below. Learn more

GPU cloud with per-minute 1x to 8x NVIDIA B200, H100, A100 and GH200 instances and reservable InfiniBand clusters.

GPU cloudInference hosting

GPU-focused cloud · Headquarters: San Jose, United States

From published sources, not hands-on tested

Lambda, formerly Lambda Labs, is a GPU cloud. It is not AWS Lambda, Amazon's serverless compute service.

From
$0.79≈ €0.706 (estimated conversion from $0.79)/GPU-hrV100 16GB · On-demand · rented as 8 GPUsChecked lambda.ai
GPU families
7
B200 · GH200 Grace Hopper Superchip · A100 · +4 more · 9 models
Regions
16
1 in the EU
Billing
Per minute
Reserved, usage-based

Prices checked 37 sourcesSources (37)How we check

In short

Lambda (Lambda, Inc.; founded in San Francisco in 2012, with its legal address in San Jose, California) runs a GPU cloud. Its self-serve On-Demand Cloud instances are Ubuntu virtual machines with 1, 2, 4 or 8 NVIDIA GPUs (B200, H100 SXM and PCIe, A100, GH200, RTX A6000, A10, V100 and Quadro RTX 6000), priced per GPU-hour and billed in one-minute increments; H100 SXM costs $4.29 per GPU-hour as a 1x instance and $3.99 in the 8x configuration.

Read full summary

1-Click Clusters, reserved from the console subject to approval, provide 16 or more InfiniBand-connected H100 or B200 GPUs for 2 weeks to 1 year at $5.54 to $9.86 per GPU-hour, billed weekly. Lambda charges no egress fees and has no spot tier, and its serverless Inference API is winding down. Lambda's docs name 16 regions: 14 On-Demand instance regions, one of them in Germany, and 2 US regions (us-east-3, us-midwest-2) that appear only as filesystem endpoints. Private Cloud clusters of 1,000+ GPUs are sold through sales.

From

$0.79≈ €0.706 (estimated conversion from $0.79)/GPU-hrV100 16GB · On-demand · rented as 8 GPUsChecked lambda.ai
Visit Lambda Opens lambda.ai in a new tab · Check the final plan, price and terms there before you order.Compare with alternatives to Lambda
Setup guides for Lambda
Billing increment
Per minute
Egress
No egress feesSource
Persistent storage
Yes
Managed service
Partial

Pricing

Prices checked 27 pricesHow we check

Billing: per minute, reserved, usage-based. Lowest pay-as-you-go GPU price: $0.79≈ €0.706 (estimated conversion from $0.79) per GPU-hour (V100 16GB, on-demand, rented as 8 GPUs).

Prices in USD, as listed, excluding VAT.≈ Estimated conversion from the listed USD price at the ECB reference rate of 23 Oct 2026 (1 EUR = 1.1193 USD). Billing currency not verified: providers may bill in USD, and your bank's rate and fees differ. Prices exclude VAT.

GPU offers27

Lambda GPU offers: 27 offers grouped by GPU; prices per GPU-hour, excl. VAT. Region: varies by offer. Availability: not verified for every offer.

NVIDIA B200 SXM 180GB180 GB VRAM · 7 offers · from $6.69≈ €5.98 (estimated conversion from $6.69) per GPU-hour

Checked lambda.ai

  • On-demand8-GPU node · 208 vCPU · 2900 GB RAM · 22528 GB disk$6.69≈ €5.98 (estimated conversion from $6.69)/GPU-hrPer minute
  • On-demand4-GPU node · 104 vCPU · 1440 GB RAM · 11264 GB disk$6.79≈ €6.07 (estimated conversion from $6.79)/GPU-hrPer minute
  • On-demand2-GPU node · 52 vCPU · 720 GB RAM · 5632 GB disk$6.89≈ €6.16 (estimated conversion from $6.89)/GPU-hrPer minute
  • On-demand26 vCPU · 360 GB RAM · 2816 GB disk$6.99≈ €6.24 (estimated conversion from $6.99)/GPU-hrPer minute
  • Reserved256-GPU node$8.87≈ €7.92 (estimated conversion from $8.87)/GPU-hrHourly
  • Reserved64-GPU node$9.36≈ €8.36 (estimated conversion from $9.36)/GPU-hrHourly
  • Reserved16-GPU node$9.86≈ €8.81 (estimated conversion from $9.86)/GPU-hrHourly

NVIDIA GH200 Grace Hopper Superchip (96GB HBM3)96 GB VRAM · 1 offer · from $2.29≈ €2.05 (estimated conversion from $2.29) per GPU-hour

Checked lambda.ai

  • On-demand64 vCPU · 432 GB RAM · 4096 GB disk$2.29≈ €2.05 (estimated conversion from $2.29)/GPU-hrPer minute
Show all 27 offers (from $0.79≈ €0.706 (estimated conversion from $0.79) per GPU-hour)

NVIDIA A100 SXM4 80GB80 GB VRAM · 1 offer · from $2.79≈ €2.49 (estimated conversion from $2.79) per GPU-hour

Checked lambda.ai

  • On-demand8-GPU node · 240 vCPU · 1800 GB RAM · 19968 GB disk$2.79≈ €2.49 (estimated conversion from $2.79)/GPU-hrPer minute

NVIDIA H100 PCIe 80GB80 GB VRAM · 1 offer · from $3.29≈ €2.94 (estimated conversion from $3.29) per GPU-hour

Checked lambda.ai

  • On-demand26 vCPU · 225 GB RAM · 1024 GB disk$3.29≈ €2.94 (estimated conversion from $3.29)/GPU-hrPer minute

NVIDIA H100 SXM 80GB80 GB VRAM · 7 offers · from $3.99≈ €3.56 (estimated conversion from $3.99) per GPU-hour

Checked lambda.ai

  • On-demand8-GPU node · 208 vCPU · 1800 GB RAM · 22528 GB disk$3.99≈ €3.56 (estimated conversion from $3.99)/GPU-hrPer minute
  • On-demand4-GPU node · 104 vCPU · 900 GB RAM · 11264 GB disk$4.09≈ €3.65 (estimated conversion from $4.09)/GPU-hrPer minute
  • On-demand2-GPU node · 52 vCPU · 450 GB RAM · 5632 GB disk$4.19≈ €3.74 (estimated conversion from $4.19)/GPU-hrPer minute
  • On-demand26 vCPU · 225 GB RAM · 2816 GB disk$4.29≈ €3.83 (estimated conversion from $4.29)/GPU-hrPer minute
  • Reserved256-GPU node$5.54≈ €4.95 (estimated conversion from $5.54)/GPU-hrHourly
  • Reserved64-GPU node$5.85≈ €5.23 (estimated conversion from $5.85)/GPU-hrHourly
  • Reserved16-GPU node$6.16≈ €5.50 (estimated conversion from $6.16)/GPU-hrHourly

NVIDIA RTX A6000 48GB48 GB VRAM · 3 offers · from $1.09≈ €0.974 (estimated conversion from $1.09) per GPU-hour

Checked lambda.ai

  • On-demand14 vCPU · 100 GB RAM · 512 GB disk$1.09≈ €0.974 (estimated conversion from $1.09)/GPU-hrPer minute
  • On-demand2-GPU node · 28 vCPU · 200 GB RAM · 1024 GB disk$1.09≈ €0.974 (estimated conversion from $1.09)/GPU-hrPer minute
  • On-demand4-GPU node · 56 vCPU · 400 GB RAM · 1024 GB disk$1.09≈ €0.974 (estimated conversion from $1.09)/GPU-hrPer minute

NVIDIA A100 40GB40 GB VRAM · 5 offers · from $1.99≈ €1.78 (estimated conversion from $1.99) per GPU-hour

Checked lambda.ai

  • On-demand30 vCPU · 225 GB RAM · 512 GB disk$1.99≈ €1.78 (estimated conversion from $1.99)/GPU-hrPer minute
  • On-demand30 vCPU · 220 GB RAM · 512 GB disk$1.99≈ €1.78 (estimated conversion from $1.99)/GPU-hrPer minute
  • On-demand2-GPU node · 60 vCPU · 450 GB RAM · 1024 GB disk$1.99≈ €1.78 (estimated conversion from $1.99)/GPU-hrPer minute
  • On-demand4-GPU node · 120 vCPU · 900 GB RAM · 1024 GB disk$1.99≈ €1.78 (estimated conversion from $1.99)/GPU-hrPer minute
  • On-demand8-GPU node · 124 vCPU · 1800 GB RAM · 5939.2 GB disk$1.99≈ €1.78 (estimated conversion from $1.99)/GPU-hrPer minute

NVIDIA A10 24GB24 GB VRAM · 1 offer · from $1.29≈ €1.15 (estimated conversion from $1.29) per GPU-hour

Checked lambda.ai

  • On-demand30 vCPU · 226 GB RAM · 1331.2 GB disk$1.29≈ €1.15 (estimated conversion from $1.29)/GPU-hrPer minute

NVIDIA V100 16GB16 GB VRAM · 1 offer · from $0.79≈ €0.706 (estimated conversion from $0.79) per GPU-hour

Checked lambda.ai

  • On-demand8-GPU node · 88 vCPU · 448 GB RAM · 5939.2 GB disk$0.79≈ €0.706 (estimated conversion from $0.79)/GPU-hrPer minute

Prices are per single GPU-hour, excluding VAT. Multi-GPU nodes are divided by their GPU count; the node size is shown with the price type. Each row shows when the price was observed and links to the page it was read from.

Research notes for these offers27
NVIDIA B200 SXM 180GB · On-demand · 8-GPU node
NVIDIA B200 SXM6 180 GB, 8x configuration (pricing-page tab '8x'). Published as $6.69 per GPU-hour plus applicable sales tax/VAT/GST; $53.52/hr for the 8-GPU instance (derived). Instance: 208 vCPUs, 2,900 GiB RAM and a 22 TiB SSD root volume (22,528 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Lambda labels the GPU 'NVIDIA B200 SXM6'.
NVIDIA B200 SXM 180GB · On-demand · 4-GPU node
NVIDIA B200 SXM6 180 GB, 4x configuration (pricing-page tab '4x'). Published as $6.79 per GPU-hour plus applicable sales tax/VAT/GST; $27.16/hr for the 4-GPU instance (derived). Instance: 104 vCPUs, 1,440 GiB RAM and a 11 TiB SSD root volume (11,264 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Lambda labels the GPU 'NVIDIA B200 SXM6'.
NVIDIA B200 SXM 180GB · On-demand · 2-GPU node
NVIDIA B200 SXM6 180 GB, 2x configuration (pricing-page tab '2x'). Published as $6.89 per GPU-hour plus applicable sales tax/VAT/GST; $13.78/hr for the 2-GPU instance (derived). Instance: 52 vCPUs, 720 GiB RAM and a 5.5 TiB SSD root volume (5,632 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Lambda labels the GPU 'NVIDIA B200 SXM6'.
NVIDIA B200 SXM 180GB · On-demand
NVIDIA B200 SXM6 180 GB, 1x configuration (pricing-page tab '1x'). Published as $6.99 per GPU-hour plus applicable sales tax/VAT/GST. Instance: 26 vCPUs, 360 GiB RAM and a 2.75 TiB SSD root volume (2,816 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Lambda labels the GPU 'NVIDIA B200 SXM6'.
NVIDIA B200 SXM 180GB · Reserved · 256-GPU node
1-Click Cluster, pricing-page row 'NVIDIA HGX B200': duration '2 weeks – 1 year', GPU count '256+', $8.87 per GPU-hour plus applicable sales tax/VAT/GST ($2,270.72/hr for 256 GPUs, derived). Shown as Reserved because a reservation of at least 2 weeks is required. The billing increment is shown as hourly because the price is quoted per GPU-hour, but Lambda bills 1CCs 'in weekly increments according to the terms of your reservation'; the invoice must be paid within 10 days of approval, and reservations cannot be extended or renewed. Each 1CC includes 3 CPU management nodes; each compute node has 24 TB usable local ephemeral NVMe; GPUs are connected by a Quantum-2 400 Gb/s InfiniBand fabric. Filesystems are billed separately by usage. Terms of '1 year+' are 'Talk to our team' (no public price). These per-GPU rates are higher than the 8x on-demand instance rate for the same GPU ($6.69). Regions are chosen in the reservation wizard and are not listed publicly; capacity is not verified. The docs say 1CCs hold 16 to 512 GPUs, while the product page says 16 to 2,000+. On the pricing page this row's action button is 'Talk to our team'; the 1CC docs and product page describe self-service reservation from the Lambda Cloud console.
NVIDIA B200 SXM 180GB · Reserved · 64-GPU node
1-Click Cluster, pricing-page row 'NVIDIA HGX B200': duration '2 weeks – 1 year', GPU count '64', $9.36 per GPU-hour plus applicable sales tax/VAT/GST ($599.04/hr for 64 GPUs, derived). Shown as Reserved because a reservation of at least 2 weeks is required. The billing increment is shown as hourly because the price is quoted per GPU-hour, but Lambda bills 1CCs 'in weekly increments according to the terms of your reservation'; the invoice must be paid within 10 days of approval, and reservations cannot be extended or renewed. Each 1CC includes 3 CPU management nodes; each compute node has 24 TB usable local ephemeral NVMe; GPUs are connected by a Quantum-2 400 Gb/s InfiniBand fabric. Filesystems are billed separately by usage. Terms of '1 year+' are 'Talk to our team' (no public price). These per-GPU rates are higher than the 8x on-demand instance rate for the same GPU ($6.69). Regions are chosen in the reservation wizard and are not listed publicly; capacity is not verified. The docs say 1CCs hold 16 to 512 GPUs, while the product page says 16 to 2,000+. On the pricing page this row's action button is 'Talk to our team'; the 1CC docs and product page describe self-service reservation from the Lambda Cloud console.
NVIDIA B200 SXM 180GB · Reserved · 16-GPU node
1-Click Cluster, pricing-page row 'NVIDIA HGX B200': duration '2 weeks – 1 year', GPU count '16', $9.86 per GPU-hour plus applicable sales tax/VAT/GST ($157.76/hr for 16 GPUs, derived). Shown as Reserved because a reservation of at least 2 weeks is required. The billing increment is shown as hourly because the price is quoted per GPU-hour, but Lambda bills 1CCs 'in weekly increments according to the terms of your reservation'; the invoice must be paid within 10 days of approval, and reservations cannot be extended or renewed. Each 1CC includes 3 CPU management nodes; each compute node has 24 TB usable local ephemeral NVMe; GPUs are connected by a Quantum-2 400 Gb/s InfiniBand fabric. Filesystems are billed separately by usage. Terms of '1 year+' are 'Talk to our team' (no public price). These per-GPU rates are higher than the 8x on-demand instance rate for the same GPU ($6.69). Regions are chosen in the reservation wizard and are not listed publicly; capacity is not verified. The docs say 1CCs hold 16 to 512 GPUs, while the product page says 16 to 2,000+. On the pricing page this row's action button is 'Talk to our team'; the 1CC docs and product page describe self-service reservation from the Lambda Cloud console.
NVIDIA GH200 Grace Hopper Superchip (96GB HBM3) · On-demand
NVIDIA GH200 96 GB, 1x configuration (pricing-page tab '1x'). Published as $2.29 per GPU-hour plus applicable sales tax/VAT/GST. Instance: 64 vCPUs, 432 GiB RAM and a 4 TiB SSD root volume (4,096 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Offered only as a 1x instance. The GH200 Grace Hopper Superchip has an onboard Arm CPU; Lambda's docs ask users to verify that workloads run on ARM.
NVIDIA A100 SXM4 80GB · On-demand · 8-GPU node
NVIDIA A100 SXM 80 GB, 8x configuration (pricing-page tab '8x'). Published as $2.79 per GPU-hour plus applicable sales tax/VAT/GST; $22.32/hr for the 8-GPU instance (derived). Instance: 240 vCPUs, 1,800 GiB RAM and a 19.5 TiB SSD root volume (19,968 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Offered only as an 8x instance.
NVIDIA H100 PCIe 80GB · On-demand
NVIDIA H100 PCIe 80 GB, 1x configuration (pricing-page tab '1x'). Published as $3.29 per GPU-hour plus applicable sales tax/VAT/GST. Instance: 26 vCPUs, 225 GiB RAM and a 1 TiB SSD root volume (1,024 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Offered only as a 1x instance.
NVIDIA H100 SXM 80GB · On-demand · 8-GPU node
NVIDIA H100 SXM 80 GB, 8x configuration (pricing-page tab '8x'). Published as $3.99 per GPU-hour plus applicable sales tax/VAT/GST; $31.92/hr for the 8-GPU instance (derived). Instance: 208 vCPUs, 1,800 GiB RAM and a 22 TiB SSD root volume (22,528 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login).
NVIDIA H100 SXM 80GB · On-demand · 4-GPU node
NVIDIA H100 SXM 80 GB, 4x configuration (pricing-page tab '4x'). Published as $4.09 per GPU-hour plus applicable sales tax/VAT/GST; $16.36/hr for the 4-GPU instance (derived). Instance: 104 vCPUs, 900 GiB RAM and a 11 TiB SSD root volume (11,264 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login).
NVIDIA H100 SXM 80GB · On-demand · 2-GPU node
NVIDIA H100 SXM 80 GB, 2x configuration (pricing-page tab '2x'). Published as $4.19 per GPU-hour plus applicable sales tax/VAT/GST; $8.38/hr for the 2-GPU instance (derived). Instance: 52 vCPUs, 450 GiB RAM and a 5.5 TiB SSD root volume (5,632 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login).
NVIDIA H100 SXM 80GB · On-demand
NVIDIA H100 SXM 80 GB, 1x configuration (pricing-page tab '1x'). Published as $4.29 per GPU-hour plus applicable sales tax/VAT/GST. Instance: 26 vCPUs, 225 GiB RAM and a 2.75 TiB SSD root volume (2,816 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login).
NVIDIA H100 SXM 80GB · Reserved · 256-GPU node
1-Click Cluster, pricing-page row 'NVIDIA H100' (SXM per the 1CC docs: '16 to 512 NVIDIA H100 or B200 SXM Tensor Core GPUs'): duration '2 weeks – 1 year', GPU count '256', $5.54 per GPU-hour plus applicable sales tax/VAT/GST ($1,418.24/hr for 256 GPUs, derived). Shown as Reserved because a reservation of at least 2 weeks is required. The billing increment is shown as hourly because the price is quoted per GPU-hour, but Lambda bills 1CCs 'in weekly increments according to the terms of your reservation'; the invoice must be paid within 10 days of approval, and reservations cannot be extended or renewed. Each 1CC includes 3 CPU management nodes; each compute node has 24 TB usable local ephemeral NVMe; GPUs are connected by a Quantum-2 400 Gb/s InfiniBand fabric. Filesystems are billed separately by usage. Terms of '1 year+' are 'Talk to our team' (no public price). These per-GPU rates are higher than the 8x on-demand instance rate for the same GPU ($3.99). Regions are chosen in the reservation wizard and are not listed publicly; capacity is not verified. The docs say 1CCs hold 16 to 512 GPUs, while the product page says 16 to 2,000+. On the pricing page this row's action button is 'Talk to our team'; the 1CC docs and product page describe self-service reservation from the Lambda Cloud console.
NVIDIA H100 SXM 80GB · Reserved · 64-GPU node
1-Click Cluster, pricing-page row 'NVIDIA H100' (SXM per the 1CC docs: '16 to 512 NVIDIA H100 or B200 SXM Tensor Core GPUs'): duration '2 weeks – 1 year', GPU count '64', $5.85 per GPU-hour plus applicable sales tax/VAT/GST ($374.40/hr for 64 GPUs, derived). Shown as Reserved because a reservation of at least 2 weeks is required. The billing increment is shown as hourly because the price is quoted per GPU-hour, but Lambda bills 1CCs 'in weekly increments according to the terms of your reservation'; the invoice must be paid within 10 days of approval, and reservations cannot be extended or renewed. Each 1CC includes 3 CPU management nodes; each compute node has 24 TB usable local ephemeral NVMe; GPUs are connected by a Quantum-2 400 Gb/s InfiniBand fabric. Filesystems are billed separately by usage. Terms of '1 year+' are 'Talk to our team' (no public price). These per-GPU rates are higher than the 8x on-demand instance rate for the same GPU ($3.99). Regions are chosen in the reservation wizard and are not listed publicly; capacity is not verified. The docs say 1CCs hold 16 to 512 GPUs, while the product page says 16 to 2,000+. On the pricing page this row's action button is 'Talk to our team'; the 1CC docs and product page describe self-service reservation from the Lambda Cloud console.
NVIDIA H100 SXM 80GB · Reserved · 16-GPU node
1-Click Cluster, pricing-page row 'NVIDIA H100' (SXM per the 1CC docs: '16 to 512 NVIDIA H100 or B200 SXM Tensor Core GPUs'): duration '2 weeks – 1 year', GPU count '16', $6.16 per GPU-hour plus applicable sales tax/VAT/GST ($98.56/hr for 16 GPUs, derived). Shown as Reserved because a reservation of at least 2 weeks is required. The billing increment is shown as hourly because the price is quoted per GPU-hour, but Lambda bills 1CCs 'in weekly increments according to the terms of your reservation'; the invoice must be paid within 10 days of approval, and reservations cannot be extended or renewed. Each 1CC includes 3 CPU management nodes; each compute node has 24 TB usable local ephemeral NVMe; GPUs are connected by a Quantum-2 400 Gb/s InfiniBand fabric. Filesystems are billed separately by usage. Terms of '1 year+' are 'Talk to our team' (no public price). These per-GPU rates are higher than the 8x on-demand instance rate for the same GPU ($3.99). Regions are chosen in the reservation wizard and are not listed publicly; capacity is not verified. The docs say 1CCs hold 16 to 512 GPUs, while the product page says 16 to 2,000+. On the pricing page this row's action button is 'Talk to our team'; the 1CC docs and product page describe self-service reservation from the Lambda Cloud console.
NVIDIA RTX A6000 48GB · On-demand
NVIDIA A6000 48 GB, 1x configuration (pricing-page tab '1x'). Published as $1.09 per GPU-hour plus applicable sales tax/VAT/GST. Instance: 14 vCPUs, 100 GiB RAM and a 512 GiB SSD root volume (512 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Lambda lists 'NVIDIA A6000 48 GB'; shown here as the RTX A6000.
NVIDIA RTX A6000 48GB · On-demand · 2-GPU node
NVIDIA A6000 48 GB, 2x configuration (pricing-page tab '2x'). Published as $1.09 per GPU-hour plus applicable sales tax/VAT/GST; $2.18/hr for the 2-GPU instance (derived). Instance: 28 vCPUs, 200 GiB RAM and a 1 TiB SSD root volume (1,024 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Lambda lists 'NVIDIA A6000 48 GB'; shown here as the RTX A6000.
NVIDIA RTX A6000 48GB · On-demand · 4-GPU node
NVIDIA A6000 48 GB, 4x configuration (pricing-page tab '4x'). Published as $1.09 per GPU-hour plus applicable sales tax/VAT/GST; $4.36/hr for the 4-GPU instance (derived). Instance: 56 vCPUs, 400 GiB RAM and a 1 TiB SSD root volume (1,024 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Lambda lists 'NVIDIA A6000 48 GB'; shown here as the RTX A6000.
NVIDIA A100 40GB · On-demand
NVIDIA A100 PCIe 40 GB, 1x configuration (pricing-page tab '1x'). Published as $1.99 per GPU-hour plus applicable sales tax/VAT/GST. Instance: 30 vCPUs, 225 GiB RAM and a 512 GiB SSD root volume (512 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Lambda lists 'NVIDIA A100 PCIe 40 GB'; shown as the A100 40GB, without a form factor.
NVIDIA A100 40GB · On-demand
NVIDIA A100 SXM 40 GB, 1x configuration (pricing-page tab '1x'). Published as $1.99 per GPU-hour plus applicable sales tax/VAT/GST. Instance: 30 vCPUs, 220 GiB RAM and a 512 GiB SSD root volume (512 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Lambda lists 'NVIDIA A100 SXM 40 GB'; shown as the A100 40GB, which covers both SXM4 and PCIe 40 GB cards.
NVIDIA A100 40GB · On-demand · 2-GPU node
NVIDIA A100 PCIe 40 GB, 2x configuration (pricing-page tab '2x'). Published as $1.99 per GPU-hour plus applicable sales tax/VAT/GST; $3.98/hr for the 2-GPU instance (derived). Instance: 60 vCPUs, 450 GiB RAM and a 1 TiB SSD root volume (1,024 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Lambda lists 'NVIDIA A100 PCIe 40 GB'; shown as the A100 40GB, without a form factor.
NVIDIA A100 40GB · On-demand · 4-GPU node
NVIDIA A100 PCIe 40 GB, 4x configuration (pricing-page tab '4x'). Published as $1.99 per GPU-hour plus applicable sales tax/VAT/GST; $7.96/hr for the 4-GPU instance (derived). Instance: 120 vCPUs, 900 GiB RAM and a 1 TiB SSD root volume (1,024 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Lambda lists 'NVIDIA A100 PCIe 40 GB'; shown as the A100 40GB, without a form factor.
NVIDIA A100 40GB · On-demand · 8-GPU node
NVIDIA A100 SXM 40 GB, 8x configuration (pricing-page tab '8x'). Published as $1.99 per GPU-hour plus applicable sales tax/VAT/GST; $15.92/hr for the 8-GPU instance (derived). Instance: 124 vCPUs, 1,800 GiB RAM and a 5.8 TiB SSD root volume (5,939.2 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Lambda lists 'NVIDIA A100 SXM 40 GB'; shown as the A100 40GB, which covers both SXM4 and PCIe 40 GB cards.
NVIDIA A10 24GB · On-demand
NVIDIA A10 24 GB, 1x configuration (pricing-page tab '1x'). Published as $1.29 per GPU-hour plus applicable sales tax/VAT/GST. Instance: 30 vCPUs, 226 GiB RAM and a 1.3 TiB SSD root volume (1,331.2 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Offered only as a 1x instance.
NVIDIA V100 16GB · On-demand · 8-GPU node
NVIDIA Tesla V100 16 GB, 8x configuration (pricing-page tab '8x'). Published as $0.79 per GPU-hour plus applicable sales tax/VAT/GST; $6.32/hr for the 8-GPU instance (derived). Instance: 88 vCPUs, 448 GiB RAM and a 5.8 TiB SSD root volume (5,939.2 GiB; Lambda uses binary units). Priced per hour and billed in one-minute increments from launch (once health checks pass) until termination; instances cannot be stopped, only terminated. Self-serve, first-come access. Lambda does not publish which regions offer this type, and live capacity is not verified (it is visible only after login). Offered only as an 8x instance. Lambda lists 'NVIDIA Tesla V100 16 GB' without a form factor, so it is shown as the V100 16GB, without a form factor. Lambda's 8x V100 launch post (2020, updated 12 Aug 2022) described 'V100 SXM2 GPUs (with NVLink)', but today's instance differs from that one (88 instead of 92 vCPUs), so SXM2 is not stated as a fact. V100 instances default to the 'Lambda Stack Legacy' image (closed-source driver).

Compare these prices with other providers’: H100 prices compared.

Plans

No fixed-price plans are recorded for Lambda in the dataset.

Storage and data transfer

Lambda Filesystem (regional persistent file storage)
Not verified
File storage · PersistentCurrent price not publicly verifiable: the billing docs bill filesystems 'per GiB used per month in one-hour increments' and give '$0.20 per GiB per month' only as an example that 'might not reflect current pricing. Docs: Billing overview - Lambda Docs (opens in a new tab)
MoreThe actual price will be displayed when you create your filesystem.' A September 2023 Lambda blog post quoted $0.20/GB/month for new regions (historical). No minimum storage period and no ingress or egress charge; billing continues while the filesystem exists, even if unmounted. Mounted over the network at /lambda/nfs/<name>; must be attached when an instance or 1CC is launched; same region and workspace only; cannot be moved between regions; 8 EB capacity per filesystem (10 TB in us-south-1) and up to 24 filesystems. S3-compatible access via the Filesystem S3 Adapter in us-east-2, us-east-3 and us-midwest-2 only. On 1CCs the storage system is encrypted at rest. The Filesystems docs page itself shows no price, and the ODC overview says 'per GB' where the billing page says 'per GiB'. The S3 Adapter works with filesystems created in March 2025 or later.
Instance root volume (SSD, size fixed per instance type)
$0.00≈ €0.00 (estimated conversion from $0.00) per GB-month
Included disk · Not persistentIncluded in the instance price; size is fixed by instance type (512 GiB to 22 TiB, listed as 'SSD' on the pricing page). Docs: On-Demand Cloud overview - Lambda Docs (opens in a new tab)
MoreIt lives only as long as the instance: instances cannot be stopped, only terminated, and Lambda recommends Docker plus regular backups to preserve state.
1-Click Cluster local NVMe (ephemeral)
$0.00≈ €0.00 (estimated conversion from $0.00) per GB-month
Local NVMe · Not persistentIncluded with a 1CC reservation: 24TB usable local ephemeral NVMe per compute node and 208GB per management node. 'All data saved to ephemeral local storage will be deleted and unrecoverable once your reservation ends.' Docs: 1-Click Clusters introduction - Lambda Docs (opens in a new tab)
Egress (data out)
$0.00≈ €0.00 (estimated conversion from $0.00) per GB
Instances page: 'Transparent pricing with no egress fees.' The pricing page's meta description also says 'No egress fees.' Billing docs: filesystems carry no ingress or egress charge. Pricing page: Rent NVIDIA GPUs on demand - Lambda Instances (opens in a new tab)
More1-Click Clusters page: 'no ingress/egress fees'. 1CC Direct Internet Access is 'delivered over shared upstream capacity at the datacenter edge and is not dedicated per tenant'. Per-instance network bandwidth is not published.

Before you compare

From Lambda’s record: its first stated uses and limitation, how its prices are set, and the providers compared below.

Suited to
  • Self-serve single-node training, fine-tuning and experiments on 1x to 8x B200, H100, A100 or GH200 instances billed per minute
  • Multi-node training on InfiniBand-connected H100 or B200 clusters of 16+ GPUs reserved for 2 weeks to 1 year
All 4 uses it suits
Main limitation
No spot or interruptible tier and no serverless or managed inference endpoints; Lambda says its Inference API is winding down
All 10 limitations
Price basis
GPU prices per GPU-hour, billed per minute (lowest price rented as 8 GPUs); egress included; 3 storage options listed
All prices, storage and egress
Alternatives
Hyperstack provider pageModal provider pageOVHcloud provider pageNebius provider page
Compare side by side with LambdaHow they differ

Who it fits

Based on Lambda’s public documentation, checked . We don’t score providers. How we build records

Suited to

  • Self-serve single-node training, fine-tuning and experiments on 1x to 8x B200, H100, A100 or GH200 instances billed per minute
  • Multi-node training on InfiniBand-connected H100 or B200 clusters of 16+ GPUs reserved for 2 weeks to 1 year
  • Teams that want a preinstalled ML stack (Lambda Stack with CUDA, PyTorch, JAX, Docker and JupyterLab) on Ubuntu VMs with sudo
  • Workloads that move a lot of data out of the cloud, since Lambda charges no egress fees

Good fit for

  • TrainingStrong fit8x B200, H100 SXM and A100 SXM instances for single-node jobs, and 1-Click Clusters of 16 to 512 H100 or B200 GPUs (marketed as up to 2,000+) on a Quantum-2 400 Gb/s InfiniBand fabric, with NCCL, Open MPI and PyTorch DDP/FSDP preinstalled and optional Managed Slurm or Managed Kubernetes. Cluster reservations need at least 2 weeks and are billed weekly.
  • Fine-tuningStrong fitSelf-serve 1x to 8x instances billed per minute (1x H100 SXM $4.29/hr, 1x GH200 $2.29/hr, 1x A100 40GB $1.99/hr) with PyTorch preinstalled and persistent filesystems for checkpoints. Instances cannot be stopped, only terminated, and GH200 has an Arm CPU.
  • Batch inferenceSuitablePer-minute billing, no egress fees and official vLLM, SGLang and Ollama guides suit batch jobs on on-demand instances, and dstack can provision an instance per task. There is no spot or preemptible discount tier.

Look elsewhere for

  • Real-time inferencePartial fitModels can be self-hosted with vLLM or SGLang on instances (only port 22 is open until firewall rules are added), and Managed Kubernetes on 1-Click Clusters provides Ingress with TLS. There are no managed or serverless endpoints, no documented autoscaling for on-demand instances, and Lambda's Inference API is winding down.
  • AI agentsPartial fitNo agent templates or agent hosting. OpenClaw or Hermes Agent could be installed by hand (not tested), but On-Demand Cloud lists only GPU instances (cheapest: Quadro RTX 6000 at $0.69/hr), so Lambda fits better as the GPU backend serving an open-weight model for an agent.
  • RAG applicationsPartial fitEmbedding models, LLM servers and a vector database can be self-hosted on GPU instances with persistent filesystems. There is no managed vector database, no object storage product (the Filesystem S3 Adapter is 'not designed to be a replacement for a full-fledged object storage offering') and no RAG template.
  • RenderingPartial fitRTX A6000 (1x to 4x, $1.09 per GPU-hour), A10 and Quadro RTX 6000 instances exist and a VNC desktop setup is documented, but only Ubuntu images are offered and there are no rendering-specific images or guides.
Show 1 more
  • Private LLM hostingPartial fitOpen-weight models can be self-hosted on VMs with sudo and Docker using official vLLM and Ollama guides; 1-Click Cluster compute nodes are single-tenant, Private Cloud offers bare-metal single-tenant clusters, and SOC 2 Type 2 and ISO 27001 are listed. However, the standard Cloud Terms (Platform Guidelines) prohibit submitting sensitive personal data such as government IDs, financial account or payment card data and personal health information unless the documentation or an Order permits it, a DPA applies only if executed separately, no customer-managed keys are documented, and the only EU region (Germany) has no public list of GPU types.

Not sure Lambda fits what you want to build?Help me choose

Strengths and limitations

Summarised from the research record for Lambda. We don’t score providers.

Strengths

  • Public per-GPU-hour rate card for instances and 1-Click Clusters (B200 $6.69-$6.99, H100 SXM $3.99-$4.29, A100 SXM 80GB $2.79, GH200 $2.29), with no sales process needed for on-demand instances
  • On-demand instances are billed in one-minute increments
  • No egress fees on instances, filesystems or 1-Click Clusters; filesystems have no ingress or egress charge and no minimum storage period
  • The default image preinstalls Lambda Stack (NVIDIA driver, CUDA, cuDNN, NCCL, PyTorch, TensorFlow, JAX), Docker with the NVIDIA Container Toolkit, and JupyterLab
Show 3 more
  • 1-Click Clusters use a Quantum-2 400 Gb/s InfiniBand fabric (GPUDirect RDMA up to 3,200 Gb/s), single-tenant compute nodes with 24 TB local NVMe each, and optional Managed Kubernetes or Managed Slurm
  • The Trust Portal lists SOC 2 Type 2, ISO/IEC 27001, 27017 and 27701, and ISO 22301
  • Cloud API (launch, restart and terminate instances, with cloud-init at launch), workspaces, firewall rulesets and six months of audit-event logs; SAML/OIDC single sign-on is available to accounts with a 1-Click Cluster, set up through a Lambda account manager

Limitations

  • No spot or interruptible tier and no serverless or managed inference endpoints; Lambda says its Inference API is winding down
  • Instances can only be launched, restarted or terminated: there is no stop/suspend and no instance snapshot, so billing runs until termination (the Instances marketing page mentions stopping instances through the API, but the docs and the Cloud API reference list only launch, restart and terminate)
  • Filesystems must be attached when an instance or cluster is launched and cannot move between regions; the current storage price is shown only in the console (the docs' $0.20/GiB-month is labelled an example that 'might not reflect current pricing')
  • Lambda does not publish which instance types run in which region; access is first-come, and new accounts start with an instance quota that rises as invoices are paid
Show 6 more
  • Only one EU region (europe-central-1, Germany), with no public list of the GPU types offered there
  • Published 1-Click Cluster rates cover 2-week to 1-year reservations (terms of 1 year+ are quoted by sales); reservations are billed weekly, the invoice is due within 10 days of approval, and they cannot be extended or renewed; their per-GPU rates are higher than the 8x on-demand rates for the same GPU
  • The Cloud Terms of Service contain no uptime SLA and state that Lambda has no obligation to back up customer data; a DPA applies only when executed separately
  • The Platform Guidelines that accompany the Cloud Terms bar customers from submitting sensitive personal data (government ID numbers, financial account or payment card data, personal health information) unless the documentation or an Order expressly permits it
  • On-Demand Cloud lists only GPU instances (cheapest: Quadro RTX 6000 at $0.69/hr), so CPU-only services such as agent runtimes have no low-cost tier; prices exclude sales tax/VAT/GST
  • GH200 instances have an Arm (Grace) CPU, so x86-only software will not run on them

Infrastructure

Regions

16 regions across 5 countries; 1 is recorded as EU region.

Europe (1)
europe-central-1 (Germany) (EU)
North America (11)
us-east-1 (Virginia, USA); us-east-2 (Washington DC, USA); us-east-3 (Washington DC, USA); us-midwest-1 (Illinois, USA); us-midwest-2 (Ohio, USA); us-south-1 (Texas, USA); us-south-2 (North Texas, USA); us-south-3 (Central Texas, USA); us-west-1 (California, USA); us-west-2 (Arizona, USA); us-west-3 (Utah, USA)
Asia (4)
asia-northeast-1 (Tokyo, Japan); asia-northeast-2 (Osaka, Japan); asia-south-1 (India); me-west-1 (Israel)

Sources: Docs: On-Demand Cloud overview - Lambda Docs (opens in a new tab), Docs: Filesystem S3 Adapter - Lambda Docs (opens in a new tab)

Region notes7
asia-northeast-1 (Tokyo, Japan), asia-northeast-2 (Osaka, Japan), us-east-1 (Virginia, USA), us-midwest-1 (Illinois, USA), us-south-2 (North Texas, USA), us-south-3 (Central Texas, USA), us-west-1 (California, USA), us-west-2 (Arizona, USA), us-west-3 (Utah, USA)
Listed in the On-Demand Cloud region table. Lambda does not publish which instance types run here ('Not every instance type will be available in every region'); availability is not verified.
asia-south-1 (India), me-west-1 (Israel)
Listed in the On-Demand Cloud region table. Lambda does not publish which instance types run here ('Not every instance type will be available in every region'); availability is not verified. The docs give no city.
europe-central-1 (Germany)
Listed in the On-Demand Cloud region table. Lambda does not publish which instance types run here ('Not every instance type will be available in every region'); availability is not verified. The only EU region listed; the docs give no city. Secondary evidence only: GPU Finder (gpufinder.dev) lists Lambda's gpu_8x_a100 (8x A100 SXM 40GB, $15.92/hr) with europe-central-1 among its regions, out of stock at its last hourly check on 26 Sep 2026; not confirmed by Lambda.
us-east-2 (Washington DC, USA)
Listed in the On-Demand Cloud region table. Lambda does not publish which instance types run here ('Not every instance type will be available in every region'); availability is not verified. Also a Filesystem S3 Adapter endpoint (files.us-east-2.lambda.ai).
us-south-1 (Texas, USA)
Listed in the On-Demand Cloud region table. Lambda does not publish which instance types run here ('Not every instance type will be available in every region'); availability is not verified. Filesystems here are limited to 10 TB.
us-east-3 (Washington DC, USA)
Not in the On-Demand Cloud region table. Listed only as a Filesystem S3 Adapter endpoint (files.us-east-3.lambda.ai), so filesystems exist here; which compute products run here is not stated.
us-midwest-2 (Ohio, USA)
Not in the On-Demand Cloud region table. Listed as a Filesystem S3 Adapter endpoint (files.us-midwest-2.lambda.ai), and the Managed Kubernetes docs mention Lambda Filesystems in us-midwest-2; which compute products run here is not stated.

Frameworks

How Lambda works with common AI serving, agent and training frameworks. Support levels, strongest first: an official template, documentation, a community guide, or manual setup on a general-purpose machine.

Hosting

Support for running the framework itself on Lambda.

PyTorch
Official template
PyTorch is preinstalled in Lambda Stack, the default image on on-demand instances and 1-Click Cluster nodes (with DDP and FSDP support on 1CCs). Docs: On-Demand Cloud overview - Lambda Docs (opens in a new tab)
Ollama hosting providers
Documented setup
Official guide runs the ollama/ollama Docker image on an on-demand instance (DeepSeek-R1 70B; 1x GH200 or 1x H100 recommended). The education index also lists a Llama 3.2 3B on Kubernetes guide. Docs: Running DeepSeek-R1 70B using Ollama - Lambda Docs (opens in a new tab)
SGLang
Documented setup
The official dstack guide launches an SGLang server (lmsysorg/sglang image) on a Lambda on-demand instance. Docs: Orchestrating AI workloads with dstack - Lambda Docs (opens in a new tab)
vLLM hosting providers
Documented setup
Official guides serve Llama 3.1 8B/70B with the vllm/vllm-openai Docker image on on-demand instances, deploy Nemotron 3 Nano with vLLM, and run a vLLM server for Hermes 4 on Managed Kubernetes with Lambda-hosted manifests. Docs: Serving Llama 3.1 8B and 70B on on-demand instances - Lambda Docs (opens in a new tab)
MoreNo one-click template.
Dify
Manual setup
No template or guide. It would need a self-managed Docker install on a GPU instance, with firewall rules added because only port 22 is open by default (not tested). Docs: Managing your system environment - Lambda Docs (opens in a new tab)
Flowise
Manual setup
No template or guide. It would need a self-managed install on a GPU instance, with firewall rules added because only port 22 is open by default (not tested). Docs: Managing your system environment - Lambda Docs (opens in a new tab)
Hermes Agent
Manual setup
No template or deployment guide; self-install on an instance is possible (not tested). Docs: Creating highly efficient agents: 450M tool-calling tokens (30 Apr 2026) - Lambda (opens in a new tab)
MoreA Lambda blog post (30 Apr 2026) describes generating a 450M-token Hermes Agent dataset on 184 H100 GPUs with vLLM, which is not a customer deployment guide. The docs' Hermes 3 (KubeAI) and Hermes 4 (MK8s) guides serve Nous Research models, not the Hermes Agent framework.
LangGraph
Manual setup
No Lambda guide found. It is a Python library that can be installed in a venv or conda environment on an instance (not tested). Docs: Managing your system environment - Lambda Docs (opens in a new tab)
n8n
Manual setup
No template or guide. It would need a self-managed install on a GPU instance, with firewall rules added because only port 22 is open by default (not tested). Docs: Managing your system environment - Lambda Docs (opens in a new tab)
OpenClaw
Manual setup
No template or deployment guide; it could be installed on an Ubuntu instance (not tested), but On-Demand Cloud has no CPU-only instance type. Docs: Creating highly efficient agents: 450M tool-calling tokens (30 Apr 2026) - Lambda (opens in a new tab)
MoreA Lambda blog post (30 Apr 2026) mentions OpenClaw only as context for agent training data.
Text Generation Inference
Manual setup
No Lambda guide found. It could run as a container using the preinstalled Docker and NVIDIA Container Toolkit (not tested). Docs: Managing your system environment - Lambda Docs (opens in a new tab)

Lambda is not listed on the Hermes Agent hosting and OpenClaw hosting pages. These pages list providers that host the framework with an official template, documentation or a community guide.

Operations

Managed service: partial. Each feature below states what the provider documents, with the source it was checked against. “Not verified” means no reliable source was found.

Features

Compute and access

Custom containers (Docker)
Yes
'Your ODC instance has Docker and the NVIDIA Container Toolkit installed by default.' The GPU Base image also includes Docker; the plain Ubuntu Server image has no extra packages. Docs: Managing your system environment - Lambda Docs (opens in a new tab)
More1CC Slurm nodes include Podman, Apptainer, Pyxis and Enroot.
Root access
Yes
Ubuntu VMs with a default 'ubuntu' user that has sudo (docs use sudo for Docker and user management). On 1-Click Clusters, unmanaged Slurm gives sudo/root; Managed Slurm does not. Docs: Managing your system environment - Lambda Docs (opens in a new tab)
SSH access
Yes
SSH as ubuntu@<instance IP> with a workspace SSH key; 1CC management nodes are reachable by SSH and compute nodes through them. Docs: Connecting to an instance - Lambda Docs (opens in a new tab)
API access
Yes
Lambda Cloud API with API-key authentication (keys have full access to all operations): list instance types and regions, launch instances (with cloud-init user_data), restart, list and terminate them, and manage filesystems, SSH keys, images, firewall rules and rulesets, audit events and support tickets. Docs: Lambda Cloud API Docs (opens in a new tab)
Jupyter notebooks
Yes
JupyterLab is preinstalled on the default image and opened from the console; also available on 1CC nodes. Docs: On-Demand Cloud overview - Lambda Docs (opens in a new tab)
Templates
Partial
No application template marketplace. Docs: On-Demand Cloud overview - Lambda Docs (opens in a new tab)
MoreThe default image is Lambda Stack; through the API you can choose Lambda Stack, Lambda Stack Legacy, GPU Base or Ubuntu Server images (22.04/24.04) and add cloud-init. Framework setups are documented as guides, not one-click templates.
Multi-GPU instances
Yes
2x, 4x and 8x instances (B200 and H100 SXM in all four sizes; A100 PCIe 40GB and A6000 in 1x, 2x and 4x; A100 SXM 40GB in 1x and 8x; A100 SXM 80GB and V100 in 8x only) and 1-Click Clusters of 16+ GPUs over InfiniBand. Pricing page: AI cloud pricing (Instances and 1-Click Clusters) - Lambda (opens in a new tab)
Bare metal
Partial
Private Cloud provides 'bare metal, single-tenant clusters for defined reservation periods' (1,000+ B200 GPUs for one to three years, via sales). Docs: Private Cloud introduction - Lambda Docs (opens in a new tab)
MoreOn-Demand Cloud instances are virtual machines; 1CC compute nodes run on single-tenant hardware while management nodes are multi-tenant VMs.
Kubernetes
Partial
Managed Kubernetes with NVIDIA GPU and Network Operators, InfiniBand RDMA, Traefik ingress with wildcard TLS and shared StorageClasses runs only on 1-Click Clusters (16+ GPUs) or Private Cloud. Docs: Using Lambda's Managed Kubernetes - Lambda Docs (opens in a new tab)
MoreOn on-demand instances Kubernetes must be self-installed (the docs include a SkyPilot guide).
Show all 21 features

Management

Managed service
Partial
Managed Kubernetes (an add-on for 1-Click Clusters, per the 1CC support page) and Managed Slurm on 1CCs; Private Cloud can add Managed Kubernetes. On-Demand instances are self-managed VMs. Docs: Using Lambda's Managed Kubernetes - Lambda Docs (opens in a new tab)
Serverless
No
No serverless GPU or per-token inference product: lambda.ai/inference says 'As the Inference API winds down, you can continue deploying and scaling models seamlessly on NVIDIA GPU instances.' The status page still lists a 'Chat Completions API' component. Docs: Inference - Lambda (opens in a new tab)

Storage and data

Persistent storage
Yes
Regional network filesystems persist independently of instances (attach at launch only; same region and workspace). Docs: Filesystems - Lambda Docs (opens in a new tab)
Snapshots
No
'At the moment, Lambda Cloud instances can only be launched, restarted, or terminated.' Docs recommend Docker and regular backups to preserve state; no instance or filesystem snapshot feature is documented. Docs: Creating and managing instances - Lambda Docs (opens in a new tab)
MoreThe Instances marketing page says the Cloud API can 'create, stop, and restart instances', but the live Cloud API reference exposes only launch, restart and terminate instance operations, and the docs' 'Suspending an instance' section still says instances can only be launched, restarted or terminated.
Managed backups
No
Cloud ToS: unless an Order says otherwise, 'Lambda shall have no obligation to maintain Customer Property, backup Customer Property, or otherwise store Customer Property'. Docs: Terms of Service (Cloud Terms of Service) - Lambda (opens in a new tab)
MoreDocs tell users to back up data themselves (to a filesystem, via rsync or the S3 adapter).

Billing

Per-second billing
No
'ODC prices instances by hourly usage and bills in one-minute increments.' 1CCs are billed in weekly increments; filesystems per GiB-month in one-hour increments. Docs: Billing overview - Lambda Docs (opens in a new tab)
Spot pricing
No
The pricing page lists on-demand instances, 1-Click Cluster reservations and reserved capacity via sales only; no spot, preemptible or interruptible tier appears there or in the billing docs. Pricing page: AI cloud pricing (Instances and 1-Click Clusters) - Lambda (opens in a new tab)
Reserved pricing
Yes
1-Click Clusters publish per-GPU-hour rates for 2-week to 1-year reservations (H100 $5.54-$6.16, B200 $8.87-$9.86); '1 year+' terms and 'reserved capacity at our lowest prices' go through sales. Pricing page: AI cloud pricing (Instances and 1-Click Clusters) - Lambda (opens in a new tab)
MorePrivate Cloud reservations run one to three years (Hyperscale three to ten).

Regions and compliance

EU region
Yes
europe-central-1 (Germany) is in the ODC region table. Docs: On-Demand Cloud overview - Lambda Docs (opens in a new tab)
MoreLambda does not publish which instance types run there: the Cloud API's instance-types endpoint, which reports the regions with capacity, needs an API key, and the status page has no per-region components. A third-party tracker (GPU Finder, secondary) lists Lambda's 8x A100 SXM 40GB type with europe-central-1 among its regions but reported it out of stock at its last hourly check on 26 Sep 2026 (23% availability over 30 days); this is not confirmed by Lambda.
SOC 2 report
Yes
Trust Portal lists SOC 2 Type 2 (report behind an access request) plus ISO/IEC 27001:2022, 27017:2015, 27701:2019, ISO 22301:2019 and CCPA; lambda.ai/trust states 'SOC 2 Type II attestation'. Docs: Lambda Trust Portal (SafeBase) (opens in a new tab)
GDPR data processing agreement
Partial
The Cloud ToS incorporate a DPA only 'Where the Parties execute a Data Processing Agreement'; no public DPA document was found. Docs: Terms of Service (Cloud Terms of Service) - Lambda (opens in a new tab)
MoreThe Privacy Policy (last updated 1 Jan 2026) says Lambda is a data processor for Cloud customers and that the policy does not apply to that processing; its statement that Lambda relies primarily on EU Standard Contractual Clauses for EEA/UK/Swiss transfers covers personal information handled under the policy, not necessarily customer workload data. The Platform Guidelines bar sensitive personal data (government IDs, financial account or payment card data, personal health information) unless the documentation or an Order permits it.
Customer-managed keys (BYOK)
Not verified
No customer-managed or customer-supplied encryption key option is documented. Docs: 1-Click Clusters security posture - Lambda Docs (opens in a new tab)
MoreThe 1CC security posture states that persistent file storage is 'encrypted at rest using industry standard algorithms and parameters' without describing key management.

SLA and support

Uptime SLA
None published
No uptime percentage found. Docs: Terms of Service (Cloud Terms of Service) - Lambda (opens in a new tab)
MoreThe Lambda Cloud Terms of Service (last updated August 2025) contain no service-level or uptime section and state that Lambda has no obligation to back up customer data. For 1-Click Clusters the support page lists initial response times (Severity 1: 4 hours, Severity 2: 8 hours, Severity 3: 10 hours, service requests: 24 hours), and Managed Slurm includes 'Slurm support with SLAs'. Negotiated order forms may differ (not verified).
Support channels
5 listed
  • Support tickets (support.lambda.ai)
  • Community forum (DeepTalk)
  • Documentation (docs.lambda.ai)
  • 1-Click Clusters: 24/7 Support Engineering through the ticketing portal, plus a Customer Success Manager, Technical Account Manager and Machine Learning Expert
  • Sales team ('Talk to our team') for reserved capacity and Private Cloud
1CC initial response targets: Severity 1 4h, Severity 2 8h, Severity 3 10h, service requests 24h; customer code, third-party software installed after handoff and VPN connections are out of scope (1CC support docs). Docs: Lambda Support (opens in a new tab)
MoreResponse targets for on-demand instances were not found. Status page status.lambda.ai lists Website, Cloud API, Chat Completions API, Network, Virtual Machines, Storage and Power, with no per-region breakdown.

Sources and verification

37 sources, accessed between 26 Sep 2026 and 20 Oct 2026. Official provider pages are listed first; the 1 secondary source is labelled as such.

How we verify

Prices, regions, storage, features, framework support, SLA and support channels link to the page they were read from, and every price shows when it was checked. Fit labels, strengths and limitations summarise the same research record. The Lambda record was last checked on 26 Sep 2026. A check date older than 7 days is marked “Stale” with an amber dot, and one older than 30 days “Needs refresh”.

Last verified
26 Sep 2026, 19:05 UTC
Checked · Stale
Method
Source-by-source review by AI agents, independently re-checked
Confidence
High
Checked by
AI research agent; every value re-checked against its source by a separate AI agent

Verification notes

On 26 September 2026 we checked every price, instance size and region in this listing against Lambda's live pricing pages, documentation and Cloud API reference. The Instances page mentions stopping instances, but the docs and API reference offer only launch, restart and terminate; the listing follows the docs. Not verified: the filesystem price (shown only in the console), which GPU types run in each region and live capacity. Later corrections are listed on the methodology page.

Commercial relationship

AIHostCompare has no active affiliate link with Lambda; the Visit provider link goes to its official website. Commissions never change ordering, alternatives, fit labels or conclusions. Read the affiliate disclosure.

Pricing

  1. 1.1-Click Clusters - Lambda (opens in a new tab) · Official pricing page · accessed
  2. 2.AI cloud pricing (Instances and 1-Click Clusters) - Lambda (opens in a new tab) · Official pricing page · accessed
  3. 3.Rent NVIDIA GPUs on demand - Lambda Instances (opens in a new tab) · Official pricing page · accessed

Documentation

  1. 4.1-Click Clusters introduction - Lambda Docs (opens in a new tab) · Official docs · accessed
  2. 5.1-Click Clusters security posture - Lambda Docs (opens in a new tab) · Official docs · accessed
  3. 6.1-Click Clusters support - Lambda Docs (opens in a new tab) · Official docs · accessed
  4. 7.About Lambda (opens in a new tab) · Official docs · accessed
  5. 8.Access and security overview - Lambda Docs (opens in a new tab) · Official docs · accessed
  6. 9.Billing overview - Lambda Docs (opens in a new tab) · Official docs · accessed
  7. 10.Connecting to an instance - Lambda Docs (opens in a new tab) · Official docs · accessed
Show all 37 sources

Documentation

  1. 11.Creating and managing instances - Lambda Docs (opens in a new tab) · Official docs · accessed
  2. 12.Creating highly efficient agents: 450M tool-calling tokens (30 Apr 2026) - Lambda (opens in a new tab) · Official docs · accessed
  3. 13.Deploying NVIDIA Nemotron 3 Nano using vLLM - Lambda Docs (opens in a new tab) · Official docs · accessed
  4. 14.Education index - Lambda Docs (opens in a new tab) · Official docs · accessed
  5. 15.Enabling single sign-on (SSO) - Lambda Docs (opens in a new tab) · Official docs · accessed
  6. 16.Filesystem S3 Adapter - Lambda Docs (opens in a new tab) · Official docs · accessed
  7. 17.Filesystems - Lambda Docs (opens in a new tab) · Official docs · accessed
  8. 18.Inference - Lambda (opens in a new tab) · Official docs · accessed
  9. 19.Lambda Cloud 8-GPU V100 instances (13 May 2020, updated 12 Aug 2022) (opens in a new tab) · Official docs · accessed
  10. 20.Lambda Cloud API Docs (opens in a new tab) · Official docs · accessed
  11. 21.Lambda Support (opens in a new tab) · Official docs · accessed
  12. 22.Lambda Trust Portal (SafeBase) (opens in a new tab) · Official docs · accessed
  13. 23.Managing your system environment - Lambda Docs (opens in a new tab) · Official docs · accessed
  14. 24.On-Demand Cloud overview - Lambda Docs (opens in a new tab) · Official docs · accessed
  15. 25.Orchestrating AI workloads with dstack - Lambda Docs (opens in a new tab) · Official docs · accessed
  16. 26.Partner with Lambda (opens in a new tab) · Official docs · accessed
  17. 27.Persistent storage for Lambda Cloud is expanding! (20 Sep 2023) (opens in a new tab) · Official docs · accessed
  18. 28.Private Cloud introduction - Lambda Docs (opens in a new tab) · Official docs · accessed
  19. 29.Running DeepSeek-R1 70B using Ollama - Lambda Docs (opens in a new tab) · Official docs · accessed
  20. 30.Serving Llama 3.1 8B and 70B on on-demand instances - Lambda Docs (opens in a new tab) · Official docs · accessed
  21. 31.Slurm on Lambda Cloud - Lambda Docs (opens in a new tab) · Official docs · accessed
  22. 32.Trust - Lambda (opens in a new tab) · Official docs · accessed
  23. 33.Using Lambda's Managed Kubernetes - Lambda Docs (opens in a new tab) · Official docs · accessed

Terms & policies

  1. 34.Privacy Policy - Lambda (opens in a new tab) · Official docs · accessed
  2. 35.Terms of Service (Cloud Terms of Service) - Lambda (opens in a new tab) · Official docs · accessed

Status

  1. 36.Lambda GPU Cloud Status (opens in a new tab) · Official status page · accessed

Secondary sources

  1. 37.A100 Cloud Pricing in Europe (2026) - GPU Finder (opens in a new tab) · Secondary source · accessed

Alternatives to Lambda

Providers that share a hosting category with Lambda, the most relevant first, spread so that every provider is listed on several provider pages. Not a ranking.

  • Hyperstack profileB200 · H200 · +4 more · None in the EU · Per minute
    $0.50≈ €0.447 (estimated conversion from $0.50)/GPU-hrChecked hyperstack.cloud
  • Modal profileB200 · H200 · +7 more · None in the EU · Per second
    $0.59≈ €0.527 (estimated conversion from $0.59)/GPU-hrChecked modal.com
  • OVHcloud profileH100 · L40S · +3 more · 4 in the EU · Per second
    ≈ $0.798 (converted from the EUR list price, €0.70)€0.70/GPU-hrChecked eu.api.ovh.com

Also sharing categories: Nebius profile

Compare all 4 side by side

Related comparisons and guides

Categories that list Lambda