TL;DR / Key Takeaways
- An on-demand H100 costs about $2.89 to $4.41 an hour at GPU clouds, against about $6.88 on AWS.
- Runpod is the only provider here with an RTX 4090, at $0.89 an hour.
- Modal bills per second and gives $30 of free credit a month.
- Stopped volumes and powered-off machines keep billing storage.
- Vast.ai is a marketplace: lower prices, host-set and more variable.
Short answer: for an H100 on demand, Runpod Secure Cloud and Lambda both list $3.99 an hour for SXM ($2.89 and $3.29 for PCIe), Modal works out to about $3.95 billed by the second with $30 of free credit a month, and AWS is about $6.88. For a cheap consumer GPU, Runpod is the only provider here with an RTX 4090 ($0.89 an hour on Secure Cloud). Vast.ai is a marketplace where hosts set prices, often lower, with more variability. Updated October 30, 2026.
How we checked: prices and billing rules come from each provider's own pricing pages and documentation, read October 30, 2026; AWS from its public price file dated October 29. GPU prices change weekly and availability is never guaranteed, so treat these as a snapshot. The 100-hour table is our arithmetic.
H100 and RTX 4090 per hour, on demand
| Provider | H100 (per GPU-hour) | RTX 4090 | Billing | Free credit |
|---|---|---|---|---|
| Runpod Secure Cloud | $3.99 SXM; $2.89 PCIe | $0.89 | Per second; no ingress or egress fees | None listed |
| Lambda | $3.99 to $4.29 SXM; $3.29 PCIe | Not offered | Per minute; no egress fees | None listed |
| Modal | About $3.95 | Not offered | Per second, serverless | $30 a month on Starter |
| Together AI | $3.99 on demand; $1.99 preemptible | Not offered | Not stated | Not stated |
| Hyperbolic | From $3.19 | Not verified | Not stated | Not stated |
| DigitalOcean GPU Droplets | $4.41 | Not offered | Per second, 5-minute minimum | $5 for new accounts |
| CoreWeave | About $6.16 ($49.24 for 8) | Not offered | Per hour, 8-GPU instances | None listed |
| AWS (p5, us-east-1) | About $6.88 | Not offered | Per second | Free tier does not cover GPUs |
| Vast.ai | Set by each host | Set by each host | Per second, no minimum | None listed |
What 100 H100-hours cost
| Provider | 100 H100-hours (our math) |
|---|---|
| Runpod Secure, PCIe | $289 |
| Lambda PCIe | $329 |
| Modal | About $395, minus $30 of monthly credit |
| Runpod Secure, SXM | $399 |
| DigitalOcean | $441 |
| AWS p5 | About $688 |
Runpod
Runpod is the most flexible option for individuals: per-second billing, no data-transfer fees, consumer GPUs as well as data-center ones, and a serverless product that scales to zero (H100 serverless is $5.39 an hour, billed per second while workers run). Secure Cloud runs in tier 3 and 4 data centers. Its cheaper Community Cloud, where independent hosts rent out machines, no longer accepts new hosts, and Runpod describes its reliability as variable. On-demand launches need at least an hour of credit in the account.
Lambda, Modal and Vast.ai
- Lambda: straightforward data-center GPUs billed by the minute with no egress fees, including B200s and GH200s; capacity is first come, first served.
- Modal: serverless GPUs for code you deploy as functions, billed per second, with $30 of free credit every month on the free Starter plan. Containers stay warm for 60 seconds after the last request by default, and that time is billed.
- Vast.ai: a marketplace where hosts set prices; interruptible instances run at least 50% below on-demand, and bandwidth is charged per byte at each host's rate.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
Which one to choose
- Cheapest dependable H100 by the hour: Runpod Secure PCIe or Lambda.
- Inference that should scale to zero: Modal or Runpod Serverless.
- Fine-tuning a small model or running Stable Diffusion cheaply: an RTX 4090 on Runpod.
- Lowest price and you can tolerate interruptions: Vast.ai interruptible instances.
- Already on AWS with credits or compliance needs: AWS, at a premium.
Frequently asked questions
What is the cheapest cloud H100 per hour?
Among providers with fixed published prices, Runpod Secure Cloud lists an H100 PCIe at $2.89 an hour and Hyperbolic from $3.19. Runpod, Lambda and Together list the H100 SXM at $3.99. Marketplaces such as Vast.ai can be lower but prices are set by each host.
What are the best Runpod alternatives?
Lambda for data-center GPUs billed by the minute, Modal for serverless GPUs with $30 of monthly free credit, Vast.ai for marketplace prices, and DigitalOcean GPU Droplets for a general cloud with GPUs.
Where can I rent an RTX 4090 in the cloud?
Runpod lists the RTX 4090 at $0.89 an hour on Secure Cloud. Lambda, Modal, CoreWeave, DigitalOcean and Together do not offer it; marketplaces such as Vast.ai may, at host-set prices.
Does Runpod charge for data transfer?
No. Runpod charges no ingress or egress fees on pods; it bills compute and storage per second, with stopped volumes at $0.20 per GB-month.