How do I compare cloud GPU providers?

Read the unit before the number. Amazon EC2, Google Cloud, Azure, and CoreWeave print an instance hour, and the GPU count is part of that row. Lambda, RunPod, DigitalOcean, Jarvislabs, and E2E print a per-GPU hour. Together prints both a per-GPU dedicated-inference hour and a cluster hour. Yotta prints a per-hour VM and a separate per-GPU cluster hour. Vast.ai does not print a list price. Then check whether the row is on-demand, and whether India or INR is stated.

What is the on-demand H100 price?

There is no single price. Where a vendor printed an on-demand H100 figure in October 2026: EC2 p5.4xlarge is $6.88 per instance hour in US East (N. Virginia), and p5.48xlarge is $55.04. Google Cloud a3-highgpu-8g is $88.490000119 per hour. Azure ND96isr H100 v5 is $98.32 per hour in East US, and NC40ads H100 v5 is $6.98 per hour. Lambda’s 8-GPU H100 SXM row is $3.99 per GPU hour. RunPod Secure Cloud H100 SXM is $3.99 per hour. CoreWeave’s 8-GPU HGX H100 is $49.24 per hour. DigitalOcean’s HGX H100 is $4.41 per GPU hour. Paperspace prints $5.95 per hour as a promo. Together dedicated inference is $5.49 per GPU hour, and GPU clusters print $3.99 hourly. E2E is ₹379.05 per hour. Jarvislabs is $3.49 per GPU hour. Yotta’s 1 x H100 on-demand VM is ₹356 per hour. Vast.ai does not state one.

Which providers publish an A100 or L4 hourly price?

A100 is printed by EC2, Google Cloud, Azure, Lambda, RunPod, CoreWeave, E2E, and Jarvislabs. It is not stated for Vast.ai, DigitalOcean, Paperspace, Together AI, or Yotta. L4 is printed by EC2 (g6.xlarge), RunPod, E2E, and Jarvislabs. Google Cloud prints an L4 virtual-workstation GPU price and does not print a G2 machine total. L4 is not stated for Azure, Lambda, CoreWeave, Vast.ai, DigitalOcean, Paperspace, Together AI, or Yotta.

Which cloud GPU providers have an India region or INR prices?

EC2 lists Asia Pacific (Mumbai) for p5, p4d, and g6, in USD. Azure lists NC40ads H100 v5 in Central India, in USD. E2E names Delhi NCR and Chennai and bills in INR. Jarvislabs names India (Noida). The hourly table checked there is USD, and the page has an INR control without a printed INR amount. Yotta prints INR prices and a Mumbai address. Google Cloud, Lambda, RunPod, CoreWeave, Vast.ai, DigitalOcean, Paperspace, and Together AI do not state an India region or INR billing on the pages checked.

Why does a price say not stated?

The vendor’s own page did not print an on-demand hourly price for that GPU in October 2026. This guide does not estimate a missing number, convert a currency, or copy a marketplace low from another site. Vast.ai’s pricing page describes live rates and does not print an H100, A100, L4, or RTX 4090 figure, so that whole price cell is not stated. A reserved price, a spot price, or a promo is labeled as such and is not used as a stand-in.

Do these hourly prices include tax and storage?

Not as one rule. E2E says the INR and USD lists exclude 18% GST. Lambda says prices are plus sales tax, VAT, or GST. EC2 and Azure figures are Linux compute for a named region, not a storage or data-transfer total. Google Cloud’s accelerator table is USD and does not name a region. DigitalOcean’s H100 card is a per-GPU hour. Open the linked pricing page before you pay. Prices change.

Cloud GPU providers compared

13 cloud GPU providers. Each row gives the on-demand hourly price the vendor prints for an H100, an A100, an L4, or an RTX 4090, whether an India region or INR billing is stated, and what that provider is for. Checked October 2026. Prices change. A cell that says "not stated" was left blank on purpose. There is no star rating.

Comparison

Prices are the public on-demand figures on each vendor site, checked October 2026. Prices change. A reserved rate, a spot rate, and a promo are not substituted for a missing on-demand number. Open the linked page before you pay.

Cloud GPU providers compared by on-demand H100, A100, L4, or RTX 4090 price, India region or INR billing, and what each one is for
Provider	On-demand hourly price	India region / INR billing	Best for
Amazon EC2	H100: $6.88/hr (p5.4xlarge, 1 GPU) and $55.04/hr (p5.48xlarge, 8 GPUs). A100: $21.957642/hr (p4d.24xlarge, 8×40 GB) and $27.44705/hr (p4de.24xlarge, 8×80 GB). L4: $0.8048/hr (g6.xlarge, 1 GPU). RTX 4090: not stated. Linux on-demand, US East (N. Virginia), USD	Yes. Asia Pacific (Mumbai) lists p5, p4d, and g6. Those rows are USD. INR billing: not stated	A Linux on-demand GPU instance, including a Mumbai region, when the bill is for the whole instance.
Google Cloud	H100: $88.490000119/hr for a3-highgpu-8g (8 GPUs). A100: $3.673385/hr for a2-highgpu-1g (1 GPU) and $5.06879789/hr for a2-ultragpu-1g (1 GPU). L4 machine total: not stated. RTX 4090: not stated. USD	not stated	An accelerator-optimized VM when the printed price is the whole machine, in USD.
Microsoft Azure	H100: $98.32/hr for ND96isr H100 v5 (8×80 GB) and $6.98/hr for NC40ads H100 v5 (1×94 GB). A100: $3.673/hr for NC24ads A100 v4 (1×80 GB). L4: not stated. RTX 4090: not stated. Linux consumption, East US, USD	Yes. Central India lists NC40ads H100 v5. The retail rows checked are USD. INR billing: not stated	An 8-GPU H100 training VM, or one A100, on a Linux consumption meter.
Lambda	H100 SXM: $3.99 per GPU/hr on the 8-GPU row (80 GB). A100 SXM: $2.79 per GPU/hr (80 GB, 8-GPU row) and $1.99 per GPU/hr (40 GB, 8-GPU row). L4: not stated. RTX 4090: not stated. USD, plus tax	not stated	A self-serve 1 to 8 GPU instance with a per-GPU hourly card.
RunPod	Secure Cloud cards: H100 SXM $3.99/hr, A100 SXM $1.79/hr, A100 PCIe $1.79/hr, L4 $0.59/hr, RTX 4090 $0.89/hr. Community Cloud in the page offer data: H100 SXM $2.69/hr, A100 SXM $1.39/hr, A100 PCIe $1.19/hr, L4 $0.44/hr, RTX 4090 $0.34/hr. USD	not stated	A single rented GPU, including an L4 or RTX 4090, with two published on-demand rates.
CoreWeave	H100: $49.24/hr for NVIDIA HGX H100, 8 GPUs, 80 GB. A100: $21.60/hr for NVIDIA A100, 8 GPUs, 80 GB. L4: not stated. RTX 4090: not stated. On-demand instance hour, USD	not stated	An 8-GPU HGX node when the printed price is the whole instance hour.
Vast.ai	H100: not stated. A100: not stated. L4: not stated. RTX 4090: not stated	not stated	A marketplace rental, when you can accept that the public pricing page does not print a list rate.
DigitalOcean GPU Droplets	H100: $4.41/GPU/hr on-demand for NVIDIA HGX H100. A100: not stated. L4: not stated. RTX 4090: not stated. USD	not stated	An on-demand H100 droplet with a per-GPU hourly card.
Paperspace	H100: $5.95/hr on-demand, printed as a special promo price. A100: not stated. L4: not stated. RTX 4090: not stated. USD	not stated	The H100 promo rate Paperspace prints, when a 3-year figure is not the price you want.
Together AI	H100: $5.49 per GPU/hr on-demand for dedicated inference, and $3.99 hourly for GPU clusters. A100: not stated. L4: not stated. RTX 4090: not stated. USD	not stated	Dedicated H100 inference, or an H100 cluster hour, from the two tables on one pricing page.
E2E Networks	On-demand, per GPU instance, INR excluding GST: H100 80 GB ₹379.05/hr. A100 80 GB ₹189/hr. A100 40 GB ₹179/hr. L4 ₹49/hr. RTX 4090: not stated. A separate USD list prints H100 $3.99/hr, A100 80 GB $2.10/hr, A100 40 GB $1.98/hr, L4 $0.57/hr	Yes. Delhi NCR and Chennai. INR is the billing currency	An hourly GPU in India with a published INR rate card.
Jarvislabs	H100 SXM $3.49/hr. A100 80 GB $1.49/hr. A100 40 GB $0.89/hr. L4 $0.44/hr. RTX 4090: not stated. On-demand, per GPU, USD	Yes. India (Noida). An INR control is on the page. The printed hourly table is USD, so an INR figure is not stated	A per-GPU on-demand hour, with a Noida region named on the pricing page.
Yotta Shakti Cloud	H100: ₹356 per hour for VM – 1 x H100 On Demand. A100: not stated. L4: not stated. RTX 4090: not stated. INR	Yes. Prices are in INR. The page lists a Mumbai address. A GPU data-center name is not stated	An on-demand H100 virtual machine billed in rupees.

Large clouds

Amazon EC2

The on-demand pricing page is a region selector. The figures here are the Linux lines in Amazon’s public EC2 price list, checked October 2026. The rows say effective 2026-10-01. In US East (N. Virginia) the list prints $6.88 per On Demand Linux p5.4xlarge Instance Hour, with 1 GPU and 80 GB HBM3, and $55.04 per On Demand Linux p5.48xlarge Instance Hour, with 8 GPUs and 640 GB HBM3. The P5 instance page says those sizes are 1 NVIDIA H100 and 8 NVIDIA H100 GPUs. p4d.24xlarge is $21.957642 per instance hour, and the P4 page says P4d has 8 NVIDIA A100 GPUs with 40 GB HBM2. p4de.24xlarge is $27.44705 per instance hour. The price-list row says 8 GPUs and 640 GB HBM2e, and the P4 page says P4de is the 80 GB HBM2e A100. g6.xlarge is $0.8048 per instance hour. The G6 page says g6.xlarge has 1 NVIDIA L4 and 24 GB. RTX 4090 is not stated. Asia Pacific (Mumbai) prints $8.256 for p5.4xlarge, $66.048 for p5.48xlarge, $26.3444 for p4d.24xlarge, and $0.9664 for g6.xlarge, all Linux on-demand. A p4de.24xlarge Mumbai row was not in that file. These are not Capacity Block rates. The currency on the rows is USD. INR billing is not stated.

Google Cloud

The accelerator-optimized price table lists prices in USD and does not name a region. a3-highgpu-8g is Nvidia H100, 8 GPUs, at $88.490000119 / 1 hour. a3-megagpu-8g is also Nvidia H100, 8 GPUs, at $93.400712807 / 1 hour. a2-highgpu-1g is Nvidia A100, 1 GPU, at $3.673385 / 1 hour. The GPU pricing page says A2 Standard machine types have A100 40GB GPUs. a2-ultragpu-1g is Nvidia A100, 1 GPU, at $5.06879789 / 1 hour. The GPU pricing page says A2 Ultra has A100 80GB GPUs. RTX 4090 is not stated. The same accelerator page prints NVIDIA L4 Virtual Workstation at $0.56004024 / 1 hour for 1 GPU. It does not print a G2 (g2-standard) on-demand machine total, so that total is not stated here. Spot rates are a different page and are not copied as on-demand. India region and INR billing are not stated. The page says that if you pay in a currency other than USD, the prices on Cloud Platform SKUs apply.

Microsoft Azure

The Linux virtual-machine pricing page loads prices by region. The numbers here are Azure Retail Prices consumption rows, not Spot and not Low Priority, checked October 2026. In East US, Standard_ND96isr_H100_v5 is $98.32 per hour. Microsoft’s size page says that size has 8 NVIDIA H100 GPUs, 80 GB each. Standard_NC40ads_H100_v5 is $6.98 per hour. The NCads H100 v5 size page says that size has 1 NVIDIA H100 NVL GPU and 94 GB. Standard_NC24ads_A100_v4 is $3.673 per hour, and the retail product name is NCads A100 v4 Series Linux. The size page says that size has 1 NVIDIA A100 PCIe GPU with 80 GB. L4 and RTX 4090 were not in the retail rows checked, so they are not stated. Central India lists Standard_NC40ads_H100_v5 at $9.772 per hour on the Linux consumption meter. Those rows are USD. INR billing is not stated.

GPU clouds

Lambda

Lambda’s instances page prints a price per GPU per hour, plus applicable sales tax, VAT, or GST. On the 8-GPU table, NVIDIA H100 SXM 80 GB is $3.99, NVIDIA A100 SXM 80 GB is $2.79, and NVIDIA A100 SXM 40 GB is $1.99. On the 1-GPU table, NVIDIA H100 SXM is $4.29 and NVIDIA H100 PCIe is $3.29. L4 and RTX 4090 are not stated. India region and INR billing are not stated.

RunPod

RunPod’s pricing page cards, checked October 2026, show one hourly rate per GPU that matches the Secure Cloud offers: H100 NVL $3.19/hr, H100 PCIe $2.89/hr, H100 SXM $3.99/hr, A100 PCIe $1.79/hr, A100 SXM $1.79/hr, L4 $0.59/hr, and RTX 4090 $0.89/hr. The same page’s offer markup also prints Community Cloud on-demand rates: H100 NVL $2.59/hr, H100 PCIe $1.99/hr, H100 SXM $2.69/hr, A100 PCIe $1.19/hr, A100 SXM $1.39/hr, L4 $0.44/hr, and RTX 4090 $0.34/hr. Currency is USD. The page says GPUs are available across 30+ regions and does not name India. INR billing is not stated.

CoreWeave

CoreWeave’s pricing page prints an on-demand price per hour next to a GPU count. In the North America section, NVIDIA HGX H100 is 8 GPUs, 80 GB, at $49.24 / hour. NVIDIA A100 is 8 GPUs, 80 GB, at $21.60 / hour. The Europe section prints the same two on-demand figures, $49.24 and $21.60. Spot prices on those rows are different and are not the on-demand price. L4 and RTX 4090 are not stated. The region headings on the page are North America and Europe. India and INR billing are not stated.

Vast.ai

Vast.ai’s pricing page says prices are set by supply and demand and that on-demand, interruptible, and reserved options exist. The page checked in October 2026 does not print an hourly H100, A100, L4, or RTX 4090 figure. Those prices are not stated. A live marketplace number changes, and this guide does not copy one from somewhere else. India region and INR billing are not stated.

DigitalOcean GPU Droplets

DigitalOcean’s GPU Droplets pricing page prints NVIDIA HGX H100 at $4.41 /GPU/hour and labels it On-Demand Price. The same GPU also shows $3.26 /GPU/hour as a 12 month reserved price. This row uses the on-demand figure. A100, L4, and RTX 4090 are not on that page, so they are not stated. The page does print other on-demand GPUs, including NVIDIA L40S at $1.57 /GPU/hour, which is not one of the four cards in the table. India region and INR billing are not stated.

Paperspace

Paperspace’s pricing page says on-demand pricing for H100 is $5.95/hour under a special promo price. The same note says $2.24/hour is a 3-year commitment, and $1.15/hour is a 3-year commitment. Those commitment figures are not copied here as on-demand. The page also says multi-GPU instances are 2x, 4x, and 8x, and that you multiply the specs and price of each type by the number of GPUs. A100, L4, and RTX 4090 on-demand prices are not stated. India region and INR billing are not stated.

Together AI

Together AI’s pricing page has two H100 figures. The dedicated-inference table says all prices are per GPU per hour. NVIDIA HGX H100 on-demand is $5.49, and the reserved column says contact sales. The GPU-clusters table, under on-demand hourly capacity, prints NVIDIA HGX H100 at $3.99. A100, L4, and RTX 4090 are not stated on that page. India region and INR billing are not stated.

India GPU clouds

E2E Networks

E2E’s pricing page, and the rate card it publishes for that page, says INR is the billing currency and USD is a separate price list, not a conversion. Prices exclude GST of 18%. Regions are Delhi NCR and Chennai. On-demand, per GPU instance: NVIDIA H100 80 GB is ₹379.05 per hour, NVIDIA A100 80 GB is ₹189 per hour, NVIDIA A100 40 GB is ₹179 per hour, and NVIDIA L4 is ₹49 per hour. The USD list on the same card prints H100 $3.99, A100 80 GB $2.10, A100 40 GB $1.98, and L4 $0.57 per hour. RTX 4090 is not stated. The rate card says it was generated on 2026-10-16. The page title also says H200/H100 starting at ₹30/hr. The T4 row is ₹30 per hour, so that “starting” line is not the H100 price.

Jarvislabs

Jarvislabs’ pricing page prints on-demand prices per GPU per hour: NVIDIA H100 SXM $3.49, NVIDIA A100 80 GB $1.49, NVIDIA A100 40 GB $0.89, and NVIDIA L4 $0.44. It says on-demand instances are billed per minute. RTX 4090 is not on that table, so it is not stated. The page says the machines are available in the India (Noida) region, and that customers outside India are billed in USD. It also has a control labeled for Indian rupees. The hourly numbers in the table that was checked are USD. A printed INR amount is not stated.

Yotta Shakti Cloud

Yotta’s Shakti Cloud pricing page prints VM – 1 x H100 On Demand at ₹356. The unit line under that table says per hour, per compute. The same table prints VM – 2 x H100 On Demand at ₹712 and VM – 4 x H100 On Demand at ₹1,423, with the same unit. A separate cluster row prints 8 x HGX H100 at ₹351 and says the unit is per GPU, per hour. A100, L4, and RTX 4090 are not stated. The page lists an address in Powai, Mumbai. It does not name a GPU data-center region beyond that.

How to choose

Match the unit. An EC2 p5.48xlarge hour is eight H100 GPUs. A Lambda or RunPod hour on these pages is one GPU. CoreWeave’s $49.24 is an 8-GPU H100 instance. Yotta’s ₹356 is one H100 virtual machine, and its ₹351 cluster row is per GPU. Dividing an instance price by eight is a calculation this page does not make.

Use the on-demand row. DigitalOcean also prints a 12-month H100 price of $3.26 per GPU hour. Paperspace labels $2.24 and $1.15 as 3-year commitments and prints $5.95 as the H100 on-demand promo. CoreWeave prints a lower spot price beside the on-demand price. The comparison table keeps the on-demand figure.

If the workload is in India, start with a row that says so. E2E names Delhi NCR and Chennai and bills in INR. Jarvislabs names Noida. Yotta prints INR prices. EC2 lists Mumbai prices in USD, and they are higher than the N. Virginia lines above. Azure lists an H100 NVL size in Central India, also in USD. The other providers do not state an India region on the pages checked.

Leave a blank cell blank. Vast.ai does not print a list price. Several providers do not print an A100, an L4, or an RTX 4090. Google Cloud’s only L4 dollar figure on the accelerator page is a virtual-workstation GPU price, not a G2 machine total. A missing number is not a guess.

Uno already has a listing on this topic, Best Cloud GPU Providers In India (2026), and lists QuickPod in AI & Machine Learning. The vendors in the table above are not those directory pages.

FAQ

Uno listings

These directory pages already exist. They are separate from the vendor write-ups above.

List a GPU tool

Uno is a directory page for the product. The submit page explains the free and paid ways to add one.

Submit a tool

Checked October 2026. Prices change. Details checked on: Amazon EC2 on-demand pricing; Amazon EC2 P5; Amazon EC2 P4; Amazon EC2 G6; AWS EC2 price list, US East (N. Virginia); AWS EC2 price list, Asia Pacific (Mumbai); Google Cloud accelerator-optimized pricing; Google Cloud GPU pricing; Azure Linux virtual machine pricing; Azure ND H100 v5 sizes; Azure NCads H100 v5 sizes; Azure NC A100 v4 sizes; Lambda instances; RunPod pricing; CoreWeave pricing; Vast.ai pricing; DigitalOcean GPU Droplets; Paperspace pricing; Together AI pricing; E2E Networks pricing; Jarvislabs pricing; Yotta Shakti Cloud pricing.

Partner Launch Platforms