GPU cloud pricing by GPU model, per hour

Compare the same GPU, not just the same price

A GPU model name is the starting point for a cloud GPU comparison, not the whole specification. The same model can come with different memory per card, interconnect, host CPU and RAM, local storage and networking, and some providers sell only multi-GPU machines. The tables keep each instance's own configuration visible and link to its full instance page.

Per GPU-hour is the instance's hourly price divided by its GPU count, rounded to four decimals. It is a derived figure for comparing instance sizes of one model, not necessarily the vendor's own per-GPU price, and it assumes nothing about utilization or performance. Each row is the provider's size with the lowest per-GPU rate for that model; smaller sizes can cost less per hour in total, so open the instance page for every size and region. Prices stay in each provider's native currency and are never ranked across currencies.

GPU models

NVIDIA H100 (4 providers)
NVIDIA A100 (2 providers)
NVIDIA B200 (2 providers)
NVIDIA GH200 (2 providers)
NVIDIA H200 (2 providers)
NVIDIA L4 (2 providers)
NVIDIA L40S (2 providers)
AMD MI300X (1 provider)
AMD MI325X (1 provider)
NVIDIA A10 (1 provider)
NVIDIA A100 PCIe (1 provider)
NVIDIA GB200 (1 provider)
NVIDIA H100 PCIe (1 provider)
NVIDIA L40 (1 provider)
NVIDIA Quadro RTX 6000 (1 provider)
NVIDIA RTX 4000 Ada (1 provider)
NVIDIA RTX A6000 (1 provider)
NVIDIA RTX PRO 6000 Blackwell Server Edition (1 provider)
NVIDIA Tesla V100 (1 provider)

GPU cloud prices by model

NVIDIA H100 cloud pricing

NVIDIA H100: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
Lambda 8x NVIDIA H100 SXM 80 GBLambda public price list; deployment region not specified	8× NVIDIA H10080 GiB per card208 vCPU · 1800 GiB RAM	USD 31.92per instance-hour (derived or composed by the collector; see notes)	USD 3.99derived: USD 31.92 ÷ 8	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Lambda pricing page, "Instances pricing" 8x tab: NVIDIA H100 SXM, 80 GB VRAM per GPU, 208 vCPUs, 1800 GiB RAM, 22 TiB SSD storage. GPU memory is stored as the page's stated 80 GB figure. The 22 TiB SSD is included with the instance, but the page does not state whether it is local or network storage, so no local-storage claim is made. CPU architecture, shared/dedicated allocation, physical cores and network throughput are not stated by the pricing page. Lambda lists $3.99 per GPU per hour; 8 × $3.99 = $31.92 per instance-hour (exact decimal arithmetic). The monthly amount is a 730-hour estimate. On-demand price before applicable sales tax/VAT/GST (page footnote). The page offers self-serve, first-come access, so a listed instance is not guaranteed to have capacity, and it states no deployment region.
DigitalOcean H100 GPU - 1X (gpu-h100x1-80gb)Amsterdam 3	1× NVIDIA H10080 GiB per card20 vCPU · 240 GiB RAM	USD 4.41per instance-hour	USD 4.41derived: USD 4.41 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence A Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Includes 720 GiB local disk allocation; the API does not specify its medium. Includes 15000 GB monthly transfer. Also reports 5120 GiB scratch disk; the API does not specify its medium. Official size price in a region where both API resources report the size available.
CoreWeave NVIDIA HGX H100Europe price table (no data center stated)	8× NVIDIA H10080 GiB per card128 vCPU · 2048 GiB RAM	USD 49.24per instance-hour	USD 6.155derived: USD 49.24 ÷ 8	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source CoreWeave pricing page, "On-demand GPU instances": NVIDIA HGX H100, 8 GPUs, 80 GB VRAM per GPU, 128 vCPUs, 2,048 GB system RAM, 61.44 TB local storage. GPU and system memory are stored as the page's stated GB figures. The 61.44 TB local storage is not converted, because the page's binary-unit footnote is about CoreWeave Cloud storage and does not say which TB the instance column uses. CPU architecture x86-64: CoreWeave's instance docs (https://docs.coreweave.com/platform/instances/gpu/gd-8xh100ib-i128.md, reviewed 2026-10-23) list the CPU model "Intel Sapphire Rapids 8462Y+ (2.80 GHz)" for 128 vCPUs and 2,048 GB RAM, and its Grace CPU guide (https://docs.coreweave.com/products/sunk/run_workloads/run-jobs-on-grace-cpu.md) says Grace instances report aarch64 and Intel or AMD instances report x86_64. Shared/dedicated allocation, physical cores, GPU interconnect and network throughput are not stated in the table. CoreWeave lists $49.24 per instance per hour on demand in its "REGION: EUROPE" table. The monthly amount is a 730-hour estimate. The page does not state whether prices include tax. Spot prices and the inference-platform single-GPU price (footnote 2: inference platform customers only) are not published here.
AWS p5.4xlargeUS East (N. Virginia)	1× NVIDIA H10080 GiB per card16 vCPU · 256 GiB RAM	USD 6.88per instance-hour	USD 6.88derived: USD 6.88 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence A Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source AWS public pricing data identifies this as GPU instance; it reports storage as "1 x 3840 GB SSD" and network as "100 Gigabit". CPU architecture x86-64: AWS EC2 instance-type documentation lists p5 with AMD (x86_64) processors (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html), and the AWS us-east-1 price list version 20261007165408 names the physical processor "AMD EPYC 7R13 Processor" (reviewed 2026-10-08). Physical CPU cores: 8 (16 vCPUs at 2 threads per core), per the EC2 instance-type docs (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html, reviewed 2026-10-08). GPU: 1 x NVIDIA H100 per the EC2 accelerated-computing instance-type docs (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html, reviewed 2026-10-08), with 80 GiB of memory per card (docs). Local storage: 1 x 3800 GB NVMe SSD instance store (capacity as AWS states it, in GB), per the EC2 instance-type docs (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html, reviewed 2026-10-08); the pricing feed lists "1 x 3840 GB SSD", and this record uses the docs value. Instance store data does not persist when the instance stops. Network bandwidth: 100 Gbps sustained, per the EC2 instance-type docs (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html, reviewed 2026-10-08). Official On-Demand hourly rate code K9VKFEBARNNMVRXV.JRTCKXETXF.6YS6EN2CT7 for shared-tenancy Linux; monthly amount is a 730-hour estimate.

NVIDIA A100 cloud pricing

NVIDIA A100: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
Lambda 1x NVIDIA A100 SXM 40 GBLambda public price list; deployment region not specified	1× NVIDIA A10040 GiB per card30 vCPU · 220 GiB RAM	USD 1.99per instance-hour	USD 1.99derived: USD 1.99 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Lambda pricing page, "Instances pricing" 1x tab: NVIDIA A100 SXM, 40 GB VRAM per GPU, 30 vCPUs, 220 GiB RAM, 512 GiB SSD storage. GPU memory is stored as the page's stated 40 GB figure. The 512 GiB SSD is included with the instance, but the page does not state whether it is local or network storage, so no local-storage claim is made. CPU architecture, shared/dedicated allocation, physical cores and network throughput are not stated by the pricing page. Lambda lists $1.99 per GPU per hour; 1 × $1.99 = $1.99 per instance-hour (exact decimal arithmetic). The monthly amount is a 730-hour estimate. On-demand price before applicable sales tax/VAT/GST (page footnote). The page offers self-serve, first-come access, so a listed instance is not guaranteed to have capacity, and it states no deployment region.
CoreWeave NVIDIA A100Europe price table (no data center stated)	8× NVIDIA A10080 GiB per card128 vCPU · 2048 GiB RAM	USD 21.60per instance-hour	USD 2.70derived: USD 21.60 ÷ 8	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source CoreWeave pricing page, "On-demand GPU instances": NVIDIA A100, 8 GPUs, 80 GB VRAM per GPU, 128 vCPUs, 2,048 GB system RAM, 7.68 TB local storage. GPU and system memory are stored as the page's stated GB figures. The 7.68 TB local storage is not converted, because the page's binary-unit footnote is about CoreWeave Cloud storage and does not say which TB the instance column uses. CPU architecture x86-64: CoreWeave's instance docs (https://docs.coreweave.com/platform/instances/gpu/gd-8xa100-i128.md, reviewed 2026-10-23) list the CPU model "Intel Ice Lake 8358 (2.60 GHz)" for 128 vCPUs and 2,048 GB RAM, and its Grace CPU guide (https://docs.coreweave.com/products/sunk/run_workloads/run-jobs-on-grace-cpu.md) says Grace instances report aarch64 and Intel or AMD instances report x86_64. Shared/dedicated allocation, physical cores, GPU interconnect and network throughput are not stated in the table. CoreWeave lists $21.60 per instance per hour on demand in its "REGION: EUROPE" table. The monthly amount is a 730-hour estimate. The page does not state whether prices include tax. Spot prices and the inference-platform single-GPU price (footnote 2: inference platform customers only) are not published here.

NVIDIA B200 cloud pricing

NVIDIA B200: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
Lambda 8x NVIDIA B200 SXM6 180 GBLambda public price list; deployment region not specified	8× NVIDIA B200180 GiB per card208 vCPU · 2900 GiB RAM	USD 53.52per instance-hour (derived or composed by the collector; see notes)	USD 6.69derived: USD 53.52 ÷ 8	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Lambda pricing page, "Instances pricing" 8x tab: NVIDIA B200 SXM6, 180 GB VRAM per GPU, 208 vCPUs, 2900 GiB RAM, 22 TiB SSD storage. GPU memory is stored as the page's stated 180 GB figure. The 22 TiB SSD is included with the instance, but the page does not state whether it is local or network storage, so no local-storage claim is made. CPU architecture, shared/dedicated allocation, physical cores and network throughput are not stated by the pricing page. Lambda lists $6.69 per GPU per hour; 8 × $6.69 = $53.52 per instance-hour (exact decimal arithmetic). The monthly amount is a 730-hour estimate. On-demand price before applicable sales tax/VAT/GST (page footnote). The page offers self-serve, first-come access, so a listed instance is not guaranteed to have capacity, and it states no deployment region.
CoreWeave NVIDIA HGX B200Europe price table (no data center stated)	8× NVIDIA B200180 GiB per card128 vCPU · 2048 GiB RAM	USD 68.80per instance-hour	USD 8.60derived: USD 68.80 ÷ 8	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source CoreWeave pricing page, "On-demand GPU instances": NVIDIA HGX B200, 8 GPUs, 180 GB VRAM per GPU, 128 vCPUs, 2,048 GB system RAM, 61.44 TB local storage. GPU and system memory are stored as the page's stated GB figures. The 61.44 TB local storage is not converted, because the page's binary-unit footnote is about CoreWeave Cloud storage and does not say which TB the instance column uses. CPU architecture x86-64: CoreWeave's instance docs (https://docs.coreweave.com/platform/instances/gpu/b200-8x.md, reviewed 2026-10-23) list the CPU model "Intel Emerald Rapids 8562Y+ (2.80 GHz)" for 128 vCPUs and 2,048 GB RAM, and its Grace CPU guide (https://docs.coreweave.com/products/sunk/run_workloads/run-jobs-on-grace-cpu.md) says Grace instances report aarch64 and Intel or AMD instances report x86_64. Shared/dedicated allocation, physical cores, GPU interconnect and network throughput are not stated in the table. CoreWeave lists $68.80 per instance per hour on demand in its "REGION: EUROPE" table. The monthly amount is a 730-hour estimate. The page does not state whether prices include tax. Spot prices and the inference-platform single-GPU price (footnote 2: inference platform customers only) are not published here.

NVIDIA GH200 cloud pricing

NVIDIA GH200: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
Lambda 1x NVIDIA GH200 96 GBLambda public price list; deployment region not specified	1× NVIDIA GH20096 GiB per card64 vCPU · 432 GiB RAM	USD 2.29per instance-hour	USD 2.29derived: USD 2.29 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Lambda pricing page, "Instances pricing" 1x tab: NVIDIA GH200, 96 GB VRAM per GPU, 64 vCPUs, 432 GiB RAM, 4 TiB SSD storage. GPU memory is stored as the page's stated 96 GB figure. The 4 TiB SSD is included with the instance, but the page does not state whether it is local or network storage, so no local-storage claim is made. CPU architecture arm64: Lambda's On-Demand Cloud docs (https://docs.lambda.ai/public-cloud/on-demand/, reviewed 2026-10-22) footnote the GH200 instance type: "The NVIDIA GH200 Grace Hopper Superchip features an onboard ARM CPU." The same page lists x86-64 and arm64 as the only image architectures. Shared/dedicated allocation, physical cores and network throughput are not stated by the pricing page. Lambda lists $2.29 per GPU per hour; 1 × $2.29 = $2.29 per instance-hour (exact decimal arithmetic). The monthly amount is a 730-hour estimate. On-demand price before applicable sales tax/VAT/GST (page footnote). The page offers self-serve, first-come access, so a listed instance is not guaranteed to have capacity, and it states no deployment region.
CoreWeave NVIDIA GH200Europe price table (no data center stated)	1× NVIDIA GH20096 GiB per card72 vCPU · 480 GiB RAM	USD 6.50per instance-hour	USD 6.50derived: USD 6.50 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source CoreWeave pricing page, "On-demand GPU instances": NVIDIA GH200, 1 GPUs, 96 GB VRAM per GPU, 72 vCPUs, 480 GB system RAM, 7.68 TB local storage. GPU and system memory are stored as the page's stated GB figures. The 7.68 TB local storage is not converted, because the page's binary-unit footnote is about CoreWeave Cloud storage and does not say which TB the instance column uses. CPU architecture arm64: CoreWeave's instance docs (https://docs.coreweave.com/platform/instances/gpu/gd-1xgh200.md, reviewed 2026-10-23) list the CPU model "NVIDIA Grace Arm v9 (3.10 GHz)" for 72 vCPUs and 480 GB RAM, and its Grace CPU guide (https://docs.coreweave.com/products/sunk/run_workloads/run-jobs-on-grace-cpu.md) says Grace instances report aarch64 and Intel or AMD instances report x86_64. Shared/dedicated allocation, physical cores, GPU interconnect and network throughput are not stated in the table. CoreWeave lists $6.50 per instance per hour on demand in its "REGION: EUROPE" table. The monthly amount is a 730-hour estimate. The page does not state whether prices include tax. Spot prices and the inference-platform single-GPU price (footnote 2: inference platform customers only) are not published here.

NVIDIA H200 cloud pricing

NVIDIA H200: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
DigitalOcean Nvidia H200 - 1X (gpu-h200x1-141gb)Atlanta 1	1× NVIDIA H200141 GiB per card24 vCPU · 240 GiB RAM	USD 4.47per instance-hour	USD 4.47derived: USD 4.47 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence A Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Includes 720 GiB local disk allocation; the API does not specify its medium. Includes 15000 GB monthly transfer. Also reports 5120 GiB scratch disk; the API does not specify its medium. Official size price in a region where both API resources report the size available.
CoreWeave NVIDIA HGX H200Europe price table (no data center stated)	8× NVIDIA H200141 GiB per card128 vCPU · 2048 GiB RAM	USD 50.44per instance-hour	USD 6.305derived: USD 50.44 ÷ 8	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source CoreWeave pricing page, "On-demand GPU instances": NVIDIA HGX H200, 8 GPUs, 141 GB VRAM per GPU, 128 vCPUs, 2,048 GB system RAM, 61.44 TB local storage. GPU and system memory are stored as the page's stated GB figures. The 61.44 TB local storage is not converted, because the page's binary-unit footnote is about CoreWeave Cloud storage and does not say which TB the instance column uses. CPU architecture x86-64: CoreWeave's instance docs (https://docs.coreweave.com/platform/instances/gpu/gd-8xh200ib-i128.md, reviewed 2026-10-23) list the CPU model "Intel Emerald Rapids 8562Y+ (2.80 GHz)" for 128 vCPUs and 2,048 GB RAM, and its Grace CPU guide (https://docs.coreweave.com/products/sunk/run_workloads/run-jobs-on-grace-cpu.md) says Grace instances report aarch64 and Intel or AMD instances report x86_64. Shared/dedicated allocation, physical cores, GPU interconnect and network throughput are not stated in the table. CoreWeave lists $50.44 per instance per hour on demand in its "REGION: EUROPE" table. The monthly amount is a 730-hour estimate. The page does not state whether prices include tax. Spot prices and the inference-platform single-GPU price (footnote 2: inference platform customers only) are not published here.

NVIDIA L4 cloud pricing

NVIDIA L4: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
UpCloud GPU-8xCPU-64GB-1xL4Sydney #1 (au-syd1)	1× NVIDIA L424 GiB per card8 vCPU · 64 GiB RAM	EUR 0.580357per instance-hour	EUR 0.5804derived: EUR 0.580357 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Curated gpu representative from the official calculator. CPU architecture and shared/dedicated allocation are not established by the structured plan record. The calculator plan record does not report bundled storage; no storage allocation, medium, or locality claim is made. GPU: 1 x NVIDIA L4 from the structured plan record; 24 GB GDDR6 per card, as UpCloud's GPU server configurations docs state (https://upcloud.com/docs/products/gpu-servers/configurations/, reviewed 2026-10-27). Structured plan record reports 16384 GiB publicTrafficOut; the calculator payload does not establish overage semantics, so this is retained only as source context. Official calculator native EUR hourly list rate and provider 28-day (672-hour) monthly cap. The calculator payload does not state a tax basis, so tax remains unknown and USD normalization is intentionally skipped. A positive plan price in the zone payload is treated as current public orderability.
AWS g6.xlargeUS East (N. Virginia)	1× NVIDIA L424 GiB per card4 vCPU · 16 GiB RAM	USD 0.8048per instance-hour	USD 0.8048derived: USD 0.8048 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence A Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source AWS public pricing data identifies this as GPU instance; it reports storage as "1 x 250 GB NVMe SSD" and network as "Up to 10 Gigabit". CPU architecture x86-64: AWS EC2 instance-type documentation lists g6 with AMD (x86_64) processors (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html), and the AWS us-east-1 price list version 20261007165408 names the physical processor "AMD EPYC 7R13 Processor" (reviewed 2026-10-08). Physical CPU cores: 2 (4 vCPUs at 2 threads per core), per the EC2 instance-type docs (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html, reviewed 2026-10-08). GPU: 1 x NVIDIA L4 per the EC2 accelerated-computing instance-type docs (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html, reviewed 2026-10-08), with 24 GB of memory per card as AWS states it (https://aws.amazon.com/ec2/instance-types/g6/); the docs' accelerator-memory column lists 22 GiB. Local storage: 1 x 250 GB NVMe SSD instance store (capacity as AWS states it, in GB), per the EC2 instance-type docs (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html, reviewed 2026-10-08). Instance store data does not persist when the instance stops. Network bandwidth: 2.5 Gbps baseline, bursting to 10 Gbps, per the EC2 instance-type docs (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html, reviewed 2026-10-08); this record uses the baseline. Official On-Demand hourly rate code TPKUC6XYSUQ3VFK2.JRTCKXETXF.6YS6EN2CT7 for shared-tenancy Linux; monthly amount is a 730-hour estimate.

NVIDIA L40S cloud pricing

NVIDIA L40S: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
AWS g6e.xlargeUS East (N. Virginia)	1× NVIDIA L40S48 GiB per card4 vCPU · 32 GiB RAM	USD 1.861per instance-hour	USD 1.861derived: USD 1.861 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence A Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source AWS public pricing data identifies this as GPU instance; it reports storage as "1 x 250 GB NVMe SSD" and network as "Up to 20 Gigabit". CPU architecture x86-64: AWS EC2 instance-type documentation lists g6e with AMD (x86_64) processors (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html), and the AWS us-east-1 price list version 20261007165408 names the physical processor "AMD EPYC 7R13 Processor" (reviewed 2026-10-08). Physical CPU cores: 2 (4 vCPUs at 2 threads per core), per the EC2 instance-type docs (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html, reviewed 2026-10-08). GPU: 1 x NVIDIA L40S per the EC2 accelerated-computing instance-type docs (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html, reviewed 2026-10-08), with 48 GB of memory per card as AWS states it (https://aws.amazon.com/ec2/instance-types/g6e/); the docs' accelerator-memory column lists 44 GiB. Local storage: 1 x 250 GB NVMe SSD instance store (capacity as AWS states it, in GB), per the EC2 instance-type docs (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html, reviewed 2026-10-08). Instance store data does not persist when the instance stops. Network bandwidth: 2.5 Gbps baseline, bursting to 20 Gbps, per the EC2 instance-type docs (https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html, reviewed 2026-10-08); this record uses the baseline. Official On-Demand hourly rate code HEU7PA78QPB8SDYY.JRTCKXETXF.6YS6EN2CT7 for shared-tenancy Linux; monthly amount is a 730-hour estimate.
CoreWeave NVIDIA L40SEurope price table (no data center stated)	8× NVIDIA L40S48 GiB per card128 vCPU · 1024 GiB RAM	USD 18.00per instance-hour	USD 2.25derived: USD 18.00 ÷ 8	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source CoreWeave pricing page, "On-demand GPU instances": NVIDIA L40S, 8 GPUs, 48 GB VRAM per GPU, 128 vCPUs, 1,024 GB system RAM, 7.68 TB local storage. GPU and system memory are stored as the page's stated GB figures. The 7.68 TB local storage is not converted, because the page's binary-unit footnote is about CoreWeave Cloud storage and does not say which TB the instance column uses. CPU architecture x86-64: CoreWeave's instance docs (https://docs.coreweave.com/platform/instances/gpu/gd-8xl40s-i128.md, reviewed 2026-10-23) list the CPU model "Intel Sapphire Rapids 8462Y+ (2.80 GHz)" for 128 vCPUs and 1,024 GB RAM, and its Grace CPU guide (https://docs.coreweave.com/products/sunk/run_workloads/run-jobs-on-grace-cpu.md) says Grace instances report aarch64 and Intel or AMD instances report x86_64. Shared/dedicated allocation, physical cores, GPU interconnect and network throughput are not stated in the table. CoreWeave lists $18.00 per instance per hour on demand in its "REGION: EUROPE" table. The monthly amount is a 730-hour estimate. The page does not state whether prices include tax. Spot prices and the inference-platform single-GPU price (footnote 2: inference platform customers only) are not published here.

AMD MI300X cloud pricing

AMD MI300X: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
DigitalOcean AMD MI300X - 1X (gpu-mi300x1-192gb)Atlanta 1	1× AMD MI300X192 GiB per card20 vCPU · 240 GiB RAM	USD 2.59per instance-hour	USD 2.59derived: USD 2.59 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence A Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Includes 720 GiB local disk allocation; the API does not specify its medium. Includes 15000 GB monthly transfer. Also reports 5120 GiB scratch disk; the API does not specify its medium. Official size price in a region where both API resources report the size available.

AMD MI325X cloud pricing

AMD MI325X: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
DigitalOcean AMD MI325X - 1X (gpu-mi325x1-256gb)New York 2	1× AMD MI325X256 GiB per card20 vCPU · 160 GiB RAM	USD 3.80per instance-hour	USD 3.80derived: USD 3.80 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence A Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Includes 720 GiB local disk allocation; the API does not specify its medium. Includes 15000 GB monthly transfer. Also reports 5120 GiB scratch disk; the API does not specify its medium. Official size price in a region where both API resources report the size available.

NVIDIA A10 cloud pricing

NVIDIA A10: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
Lambda 1x NVIDIA A10 24 GBLambda public price list; deployment region not specified	1× NVIDIA A1024 GiB per card30 vCPU · 226 GiB RAM	USD 1.29per instance-hour	USD 1.29derived: USD 1.29 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Lambda pricing page, "Instances pricing" 1x tab: NVIDIA A10, 24 GB VRAM per GPU, 30 vCPUs, 226 GiB RAM, 1.3 TiB SSD storage. GPU memory is stored as the page's stated 24 GB figure. The 1.3 TiB SSD is included with the instance, but the page does not state whether it is local or network storage, so no local-storage claim is made. CPU architecture, shared/dedicated allocation, physical cores and network throughput are not stated by the pricing page. Lambda lists $1.29 per GPU per hour; 1 × $1.29 = $1.29 per instance-hour (exact decimal arithmetic). The monthly amount is a 730-hour estimate. On-demand price before applicable sales tax/VAT/GST (page footnote). The page offers self-serve, first-come access, so a listed instance is not guaranteed to have capacity, and it states no deployment region.

NVIDIA A100 PCIe cloud pricing

NVIDIA A100 PCIe: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
Lambda 1x NVIDIA A100 PCIe 40 GBLambda public price list; deployment region not specified	1× NVIDIA A100 PCIe40 GiB per card30 vCPU · 225 GiB RAM	USD 1.99per instance-hour	USD 1.99derived: USD 1.99 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Lambda pricing page, "Instances pricing" 1x tab: NVIDIA A100 PCIe, 40 GB VRAM per GPU, 30 vCPUs, 225 GiB RAM, 512 GiB SSD storage. GPU memory is stored as the page's stated 40 GB figure. The 512 GiB SSD is included with the instance, but the page does not state whether it is local or network storage, so no local-storage claim is made. CPU architecture, shared/dedicated allocation, physical cores and network throughput are not stated by the pricing page. Lambda lists $1.99 per GPU per hour; 1 × $1.99 = $1.99 per instance-hour (exact decimal arithmetic). The monthly amount is a 730-hour estimate. On-demand price before applicable sales tax/VAT/GST (page footnote). The page offers self-serve, first-come access, so a listed instance is not guaranteed to have capacity, and it states no deployment region.

NVIDIA GB200 cloud pricing

NVIDIA GB200: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
CoreWeave NVIDIA GB200 NVL72Europe price table (no data center stated)	4× NVIDIA GB200186 GiB per card144 vCPU · 960 GiB RAM	USD 42.00per instance-hour	USD 10.50derived: USD 42.00 ÷ 4	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source CoreWeave pricing page, "On-demand GPU instances": NVIDIA GB200 NVL72, 4 GPUs, 186 GB VRAM per GPU, 144 vCPUs, 960 GB system RAM, 30.72 TB local storage. Page footnote: each instance is 2 Grace Blackwell superchips, each with 1 Grace CPU and 2 Blackwell GPUs, so 4 GPUs. GPU and system memory are stored as the page's stated GB figures. The 30.72 TB local storage is not converted, because the page's binary-unit footnote is about CoreWeave Cloud storage and does not say which TB the instance column uses. CPU architecture arm64: CoreWeave's instance docs (https://docs.coreweave.com/platform/instances/gpu/gb200-4x.md, reviewed 2026-10-23) list the CPU model "2x NVIDIA Grace Arm v9 (3.10 GHz)" for 144 vCPUs and 960 GB RAM, and its Grace CPU guide (https://docs.coreweave.com/products/sunk/run_workloads/run-jobs-on-grace-cpu.md) says Grace instances report aarch64 and Intel or AMD instances report x86_64. Shared/dedicated allocation, physical cores, GPU interconnect and network throughput are not stated in the table. CoreWeave lists $42.00 per instance per hour on demand in its "REGION: EUROPE" table. The monthly amount is a 730-hour estimate. The page does not state whether prices include tax. Spot prices and the inference-platform single-GPU price (footnote 2: inference platform customers only) are not published here.

NVIDIA H100 PCIe cloud pricing

NVIDIA H100 PCIe: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
Lambda 1x NVIDIA H100 PCIe 80 GBLambda public price list; deployment region not specified	1× NVIDIA H100 PCIe80 GiB per card26 vCPU · 225 GiB RAM	USD 3.29per instance-hour	USD 3.29derived: USD 3.29 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Lambda pricing page, "Instances pricing" 1x tab: NVIDIA H100 PCIe, 80 GB VRAM per GPU, 26 vCPUs, 225 GiB RAM, 1 TiB SSD storage. GPU memory is stored as the page's stated 80 GB figure. The 1 TiB SSD is included with the instance, but the page does not state whether it is local or network storage, so no local-storage claim is made. CPU architecture, shared/dedicated allocation, physical cores and network throughput are not stated by the pricing page. Lambda lists $3.29 per GPU per hour; 1 × $3.29 = $3.29 per instance-hour (exact decimal arithmetic). The monthly amount is a 730-hour estimate. On-demand price before applicable sales tax/VAT/GST (page footnote). The page offers self-serve, first-come access, so a listed instance is not guaranteed to have capacity, and it states no deployment region.

NVIDIA L40 cloud pricing

NVIDIA L40: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
CoreWeave NVIDIA L40Europe price table (no data center stated)	8× NVIDIA L4048 GiB per card128 vCPU · 1024 GiB RAM	USD 10.00per instance-hour	USD 1.25derived: USD 10.00 ÷ 8	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source CoreWeave pricing page, "On-demand GPU instances": NVIDIA L40, 8 GPUs, 48 GB VRAM per GPU, 128 vCPUs, 1,024 GB system RAM, 7.68 TB local storage. GPU and system memory are stored as the page's stated GB figures. The 7.68 TB local storage is not converted, because the page's binary-unit footnote is about CoreWeave Cloud storage and does not say which TB the instance column uses. CPU architecture x86-64: CoreWeave's instance docs (https://docs.coreweave.com/platform/instances/gpu/gd-8xl40-i128.md, reviewed 2026-10-23) list the CPU model "Intel Sapphire Rapids 8462Y+ (2.80 GHz)" for 128 vCPUs and 1,024 GB RAM, and its Grace CPU guide (https://docs.coreweave.com/products/sunk/run_workloads/run-jobs-on-grace-cpu.md) says Grace instances report aarch64 and Intel or AMD instances report x86_64. Shared/dedicated allocation, physical cores, GPU interconnect and network throughput are not stated in the table. CoreWeave lists $10.00 per instance per hour on demand in its "REGION: EUROPE" table. The monthly amount is a 730-hour estimate. The page does not state whether prices include tax. Spot prices and the inference-platform single-GPU price (footnote 2: inference platform customers only) are not published here.

NVIDIA Quadro RTX 6000 cloud pricing

NVIDIA Quadro RTX 6000: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
Lambda 1x NVIDIA Quadro RTX 6000 24 GBLambda public price list; deployment region not specified	1× NVIDIA Quadro RTX 600024 GiB per card14 vCPU · 46 GiB RAM	USD 0.69per instance-hour	USD 0.69derived: USD 0.69 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Lambda pricing page, "Instances pricing" 1x tab: NVIDIA Quadro RTX 6000, 24 GB VRAM per GPU, 14 vCPUs, 46 GiB RAM, 512 GiB SSD storage. GPU memory is stored as the page's stated 24 GB figure. The 512 GiB SSD is included with the instance, but the page does not state whether it is local or network storage, so no local-storage claim is made. CPU architecture, shared/dedicated allocation, physical cores and network throughput are not stated by the pricing page. Lambda lists $0.69 per GPU per hour; 1 × $0.69 = $0.69 per instance-hour (exact decimal arithmetic). The monthly amount is a 730-hour estimate. On-demand price before applicable sales tax/VAT/GST (page footnote). The page offers self-serve, first-come access, so a listed instance is not guaranteed to have capacity, and it states no deployment region.

NVIDIA RTX 4000 Ada cloud pricing

NVIDIA RTX 4000 Ada: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
Akamai Connected Cloud RTX4000 Ada x1 SmallFrankfurt 2, DE	1× NVIDIA RTX 4000 Ada20 GiB per card4 vCPU · 16 GiB RAM	USD 0.52per instance-hour	USD 0.52derived: USD 0.52 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence A Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source No monthly transfer included (the API lists 0 GB). Includes 512 GiB disk allocation; the API does not specify storage medium or locality. Published outbound network limit: 16000 Mbit/s. GPU: 1 x NVIDIA RTX 4000 Ada (count from the API gpus field); 20 GB GDDR6 per card, as Akamai's GPU compute instances docs state (https://techdocs.akamai.com/cloud-computing/docs/gpu-compute-instances, reviewed 2026-10-27). CPU allocation dedicated: Akamai's plan types page gives NVIDIA RTX 4000 Ada GPU Linodes dedicated compute isolation (https://techdocs.akamai.com/cloud-computing/docs/compute-instance-plan-types, reviewed 2026-10-23). Official base type price in an API-supported region.

NVIDIA RTX A6000 cloud pricing

NVIDIA RTX A6000: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
Lambda 1x NVIDIA A6000 48 GBLambda public price list; deployment region not specified	1× NVIDIA RTX A600048 GiB per card14 vCPU · 100 GiB RAM	USD 1.09per instance-hour	USD 1.09derived: USD 1.09 ÷ 1	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Lambda pricing page, "Instances pricing" 1x tab: NVIDIA A6000, 48 GB VRAM per GPU, 14 vCPUs, 100 GiB RAM, 512 GiB SSD storage. GPU memory is stored as the page's stated 48 GB figure. The 512 GiB SSD is included with the instance, but the page does not state whether it is local or network storage, so no local-storage claim is made. CPU architecture, shared/dedicated allocation, physical cores and network throughput are not stated by the pricing page. Lambda lists $1.09 per GPU per hour; 1 × $1.09 = $1.09 per instance-hour (exact decimal arithmetic). The monthly amount is a 730-hour estimate. On-demand price before applicable sales tax/VAT/GST (page footnote). The page offers self-serve, first-come access, so a listed instance is not guaranteed to have capacity, and it states no deployment region.

NVIDIA RTX PRO 6000 Blackwell Server Edition cloud pricing

NVIDIA RTX PRO 6000 Blackwell Server Edition: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
CoreWeave NVIDIA RTX PRO 6000 Blackwell Server Edition (High Memory)Europe price table (no data center stated)	8× NVIDIA RTX PRO 6000 Blackwell Server Edition96 GiB per card128 vCPU · 1024 GiB RAM	USD 20.00per instance-hour	USD 2.50derived: USD 20.00 ÷ 8	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source CoreWeave pricing page, "On-demand GPU instances": NVIDIA RTX PRO 6000 Blackwell Server Edition (High Memory), 8 GPUs, 96 GB VRAM per GPU, 128 vCPUs, 1,024 GB system RAM, 7.68 TB local storage. GPU and system memory are stored as the page's stated GB figures. The 7.68 TB local storage is not converted, because the page's binary-unit footnote is about CoreWeave Cloud storage and does not say which TB the instance column uses. CPU architecture x86-64: CoreWeave's instance docs (https://docs.coreweave.com/platform/instances/gpu/rtxp6000-8x.md, reviewed 2026-10-23) list the CPU model "Intel Emerald Rapids 8562Y+ (2.80 GHz)" for 128 vCPUs and 1,024 GB RAM, and its Grace CPU guide (https://docs.coreweave.com/products/sunk/run_workloads/run-jobs-on-grace-cpu.md) says Grace instances report aarch64 and Intel or AMD instances report x86_64. Shared/dedicated allocation, physical cores, GPU interconnect and network throughput are not stated in the table. CoreWeave lists $20.00 per instance per hour on demand in its "REGION: EUROPE" table. The monthly amount is a 730-hour estimate. The page does not state whether prices include tax. Spot prices and the inference-platform single-GPU price (footnote 2: inference platform customers only) are not published here.

NVIDIA Tesla V100 cloud pricing

NVIDIA Tesla V100: each provider's instance with the lowest collected per-GPU hourly rate, native currency. Per GPU-hour is derived (instance price ÷ GPUs); smaller sizes may cost less per hour in total.
Provider / instance / region	GPUs	Instance price per hour	Per GPU-hour (derived)	Vendor pricing and sources
Lambda 8x NVIDIA Tesla V100 16 GBLambda public price list; deployment region not specified	8× NVIDIA Tesla V10016 GiB per card88 vCPU · 448 GiB RAM	USD 6.32per instance-hour (derived or composed by the collector; see notes)	USD 0.79derived: USD 6.32 ÷ 8	View vendor pricing Source as of 2026-11-02Freshness deadline: 2026-11-09 · confidence B Data sources and collection notes Technical evidence used to build this comparison, not a shopping page. Original pricing source Lambda pricing page, "Instances pricing" 8x tab: NVIDIA Tesla V100, 16 GB VRAM per GPU, 88 vCPUs, 448 GiB RAM, 5.8 TiB SSD storage. GPU memory is stored as the page's stated 16 GB figure. The 5.8 TiB SSD is included with the instance, but the page does not state whether it is local or network storage, so no local-storage claim is made. CPU architecture, shared/dedicated allocation, physical cores and network throughput are not stated by the pricing page. Lambda lists $0.79 per GPU per hour; 8 × $0.79 = $6.32 per instance-hour (exact decimal arithmetic). The monthly amount is a 730-hour estimate. On-demand price before applicable sales tax/VAT/GST (page footnote). The page offers self-serve, first-come access, so a listed instance is not guaranteed to have capacity, and it states no deployment region.

Providers in this guide

Akamai Connected Cloud VM pricing on compute.info: every collected Akamai Connected Cloud instance type with its lowest price per billing term.
AWS VM pricing on compute.info: every collected AWS instance type with its lowest price per billing term.
CoreWeave VM pricing on compute.info: every collected CoreWeave instance type with its lowest price per billing term.
DigitalOcean VM pricing on compute.info: every collected DigitalOcean instance type with its lowest price per billing term.
Lambda VM pricing on compute.info: every collected Lambda instance type with its lowest price per billing term.
UpCloud VM pricing on compute.info: every collected UpCloud instance type with its lowest price per billing term.

Selection, freshness and limitations

Only active GPU instances marked available, with a known GPU model and count and a positive on-demand hourly price with no commitment, are listed. Spot, reserved and committed prices, stale providers and stale rows are excluded. For each GPU model, provider and currency, the instance with the lowest per-GPU hourly rate is shown, then the lower instance price; ties use stable SKU and region identifiers. Expired price evidence is excluded unless no GPU offer has fresh evidence: then the page shows the last-known offers under the same rules, says so at the top, and labels every row. Fresh and expired offers are never mixed.

Availability describes the collected catalog, not live capacity, quota or account eligibility. Storage, networking, egress, support and tax can change the bill. Hourly prices marked as derived or composed come from the collector's documented arithmetic on the vendor's own rates; the notes under each row say how. See the comparison methodology and the full Cloud VM catalog.