RunPod is the best default for most readers who want single GPUs by the second, with an RTX A5000 at $0.27 per hour on Secure Cloud and an H100 SXM at $3.99 per hour, while Lambda is the best pick for 8-GPU training nodes and CoreWeave suits large teams that rent whole nodes. H100 hourly rates here run from $1.79 on Vast.ai host offers to $6.16 per GPU on a CoreWeave 8-GPU node, so the same card can cost over three times as much. Lambda lists H100 at $4.59 per GPU-hour with new rates from Nov 16, 2026, and Crusoe lists H100 at $3.90 now, rising to $4.50 from Dec 1, 2026, so the headline rates here are dated snapshots.
Turnkey AI compute means you rent ready-to-use GPU machines, with the hardware, drivers, and billing already in place, instead of buying cards and building a cluster. The eight providers below differ in three ways that decide the bill: one GPU or a whole node, a fixed price or a host-set price, and what is charged on top of the GPU hour. Prices here come from each provider’s own pricing page opened on 2026-11-12.
On this page
Best companies offering turnkey ai compute: comparison table
Swipe the table sideways on a phone to see every column.
| # | Name | Best for | Starting price | Free plan / trial | Link |
|---|---|---|---|---|---|
| 1 | Lambda | Teams training on 8-GPU H100 or B200 nodes | $4.59 per GPU/hour | No | Visit |
| 2 | CoreWeave | Large teams renting whole 8-GPU nodes for training | $49.24 per 8-GPU H100 node/hour | No | Visit |
| 3 | RunPod | Solo developers and small teams renting single GPUs by the second | $0.27 per GPU/hour | No | Visit |
| 4 | Nebius | Teams wanting per-GPU billing with a cheaper preemptible tier | $1.55 per GPU/hour | No | Visit |
| 5 | Crusoe | Teams that want listed on-demand rates with no egress fees | $3.90 per GPU/hour | No | Visit |
| 6 | Paperspace | Notebook users who want a free GPU tier and hourly machines | $0.45 per GPU/hour | Yes | Visit |
| 7 | Vast.ai | Budget buyers who accept host-set prices and variable reliability | $1.79 per GPU/hour | No | Visit |
| 8 | Modal | Developers running bursty serverless GPU jobs from Python | $3.95 per GPU/hour | Yes | Visit |
The 8 best best companies offering turnkey ai computes, in detail
Lambda
Best for: Teams training on 8-GPU H100 or B200 nodes
Lambda sells on-demand GPU instances and 1-Click Clusters. The HGX H100 lists at $4.59 per GPU-hour, the HGX B200 at $7.69, the HGX A100 80GB at $3.19, the A100 40GB at $1.99, and the Tesla V100 at $0.79, all before tax. Instances bill per minute with no egress fees.
- The trade-off is that clusters cost more per GPU-hour than single instances.
- A 1-Click Cluster of 16 HGX B200 GPUs lists at $9.86 per GPU-hour, falling to $8.87 at 256 or more, and H100 clusters list at $5.54 to $6.16 on terms
- Choose Lambda when your run fills eight GPUs.
Pros
- H100 at $4.59 per GPU/hour
- 1-Click Clusters scale from 16 to 2,000+ GPUs
- no egress fees, billed per minute
Cons
- H100 SXM sold in 8-GPU nodes, so a small job still rents eight GPUs
- no preemptible tier, so every instance bills at the full on-demand rate
- new rates from Nov 16, 2026
CoreWeave
Best for: Large teams renting whole 8-GPU nodes for training
CoreWeave prices GPU compute by the node, not by the card. The HGX H100 node costs $49.24 per hour on demand, or $6.16 per GPU across eight GPUs, with 128 vCPUs and 2,048 GB of system RAM. The HGX H200 node lists at $50.44 per hour, the HGX B200 at $68.80, and the A100 at $21.60.
- The trade-off is access.
- New accounts go through sales approval, and the newest tiers, including GB300 NVL72 and HGX B300, show no on-demand rate.
- Spot pricing cuts the H100 node to $19.71 per hour, and reserved contracts are quoted at up to 60 percent below on-demand rates.
- Choose CoreWeave for large, steady training jobs on whole nodes.
Pros
- HGX H100 node at $49.24 per hour with 128 vCPUs
- spot H100 node at $19.71 per hour
- no egress fees
Cons
- GPUs sold as fixed 8-GPU nodes, no single H100 on demand
- reserved discounts up to 60 percent need a sales contract
- buying is sales-led
RunPod
Best for: Solo developers and small teams renting single GPUs by the second
RunPod rents single GPUs by the second with no contract. On Secure Cloud, the RTX A5000 costs $0.27 per hour, the RTX 4090 $0.89, the L40S $1.09, the A100 SXM $1.79, and the H100 SXM $3.99. Community Cloud cuts those rates: the RTX 4090 falls to $0.34 per hour and the H100 SXM to $2.69.
- The trade-off is storage.
- Container disk costs $0.10 per GB per month, volume disk costs $0.10 while a Pod runs and $0.20 while it is stopped, so a 100 GB volume left on a stop
- Serverless endpoints bill per second and scale to zero, with an H100 worker at $5.39 per hour.
- Choose RunPod for short single-GPU jobs.
Pros
- RTX A5000 from $0.27 per hour and H100 SXM at $3.99 per hour on Secure Cloud
- Community Cloud cuts H100 SXM to $2.69 per hour
- serverless workers scale to zero
Cons
- Volume disk $0.20 per GB/month on a stopped Pod, double the $0.10 running rate
- Community Cloud runs on third-party hosts
- no free tier
Nebius
Best for: Teams wanting per-GPU billing with a cheaper preemptible tier
Nebius is an AI cloud with a per-GPU rate card and a preemptible tier that undercuts its own on-demand rates. On demand, the L40S lists from $1.55 per GPU-hour, the RTX PRO 6000 at $1.80, the HGX H100 at $4.50, the HGX H200 at $5.40, and the HGX B200 at $8.50. Preemptible rates start from $0.79 per GPU-hour for H100 and from $0.99 for B200.
- The trade-off is that preemptible prices are dynamic spot prices, not a fixed discount, and they can move as often as every 15 minutes based on capaci
- Reserved capacity for multi-month commitments is quoted at up to 35 percent below on-demand rates.
- Choose Nebius for per-GPU billing and a preemptible tier without renting whole nodes.
Pros
- H100 at $4.50 per GPU/hour on demand, preemptible from $0.79
- L40S from $1.55 per GPU/hour
- billing per second, no charge for stopped VMs
Cons
- Preemptible prices are dynamic and can change every 15 minutes
- GB200 and GB300 NVL72 have no listed on-demand price
- $25 minimum first payment
Crusoe
Best for: Teams that want listed on-demand rates with no egress fees
Crusoe lists most of its GPU rate card in the open. The current on-demand HGX H100 lists at $3.90 per GPU-hour and the HGX H200 at $4.29, with the A100 SXM at $2.30 and the L40S at $1.50. New on-demand rates start Dec 1, 2026: H100 at $4.50, H200 at $5.40, B200 at $8.50, and B300 at $9.90.
- The trade-off is that the newest hardware sits behind sales.
- B200, GB200 NVL72, and AMD MI355X show no listed on-demand rate, and spot rates are not listed on the pricing page either.
- General-purpose CPU instances list at $0.04 per vCPU-hour and persistent disks at $0.08 per GiB per month.
- Choose Crusoe for a listed on-demand rate with no egress fees.
Pros
- H100 at $3.90 per GPU/hour on demand
- no charge for network ingress or egress
- billed by the minute with no setup fees
Cons
- H100 rises to $4.50 from Dec 1, 2026
- B200, GB200, and MI355X have no listed on-demand rate
- spot rates not listed
Paperspace
Best for: Notebook users who want a free GPU tier and hourly machines
Paperspace, owned by DigitalOcean, is the only provider here with a free GPU tier. The free tier includes an M4000 machine, and the hourly rate card starts at $0.45 per hour for a GPU+ M4000, then $0.76 for an A4000, $1.38 for an A5000, $2.30 for a V100, $3.09 for an A100, and $5.95 for an H100. The trade-off is the subscription layer.
- Gradient plans run Free at $0, Pro at $8 per month, and Growth at $39 per month, and the A100 tiers require Growth, so the hourly rate is not the whol
- The H100 carries no subscription requirement, but at $5.95 per hour it is the most expensive H100 here.
- Choose Paperspace for notebook work with a free tier.
Pros
- Free GPU tier with an M4000 machine
- H100 at $5.95 per hour with no subscription plan required
- billed by the hour
Cons
- A100 tiers require the Growth plan at $39 per month
- H100 at $5.95 per hour is the highest H100 rate here
- DigitalOcean now lists current GPUs under GPU Droplets
Vast.ai
Best for: Budget buyers who accept host-set prices and variable reliability
Vast.ai is a marketplace, not a fixed-rate cloud. Independent hosts list machines across more than 40 data centers and set their own per-second rates, so there is no single list price to quote. In a Nov 3, 2026 read of 512 offers, an H100 SXM rented on demand from $1.79 per hour with a median of $2.33, and an RTX 4090 from $0.37 with a median of $0.47.
- The trade-off is that the cheapest offer may come from an unverified host, and reliability varies by host.
- Interruptible instances typically run 30 to 50 percent below on-demand from the same host, and reserved rentals with 1, 3, or 6 month pre-payment run
- Storage is billed per GB per hour even when an instance is stopped.
- Choose Vast.ai for the lowest hourly rate.
Pros
- H100 SXM offers from $1.79 per hour, median $2.33, in a Nov 3, 2026 read
- RTX 4090 offers from $0.37
- interruptible rentals 30 to 50 percent below on-demand
Cons
- Prices are set by each host and move with supply and demand
- storage keeps billing after you stop an instance
- bandwidth is priced by the host and varies widely
Modal
Best for: Developers running bursty serverless GPU jobs from Python
Modal is a serverless platform you drive from Python, billing GPU time per second with no minimum usage time. The GPU rate card runs from a T4 at $0.59 per hour and an L4 at $0.80, through an A10 at $1.10, an A100 80GB at $2.50, an H100 at $3.95, an H200 at $4.54, and a B200 at $6.25. The Starter plan costs $0 per month and includes $30 of compute credit per month and 10 GPU concurrency.
- The trade-off is that the GPU rate is only one of three meters.
- CPU costs $0.047 per core-hour and memory $0.008 per GiB-hour, both billed on top of the GPU.
- Pinning a region multiplies the bill by 1.15x for a broad region or 1.75x for a narrow one.
- Choose Modal for bursty workloads.
Pros
- Starter plan $0 per month with $30 of monthly compute credit
- H100 at $3.95 per hour billed per second with scale to zero
- containers spin up in 1 to 2 seconds
Cons
- CPU and memory are billed on top of the GPU rate, at $0.047 per core-hour and $0.008 per GiB-hour
- pinning a region multiplies the bill by 1.15x to 1.75x
- containers stay billed for 60 seconds after the last input
How we chose
We included only providers that sell ready-to-use GPU compute with a listed hourly rate or a live marketplace of host-set rates, and we checked each provider’s own pricing page on 2026-11-12. A provider qualified when a reader could rent a GPU without buying hardware. Prices and plan limits in this article come from official vendor pricing pages and docs opened on 2026-11-12. We did not test any tool hands-on. Where a vendor hid a price behind sales, we said so. Where a price was a marketplace offer rather than a list rate, as with Vast.ai, we gave the date of the read and the median alongside the lowest offer.
Pick by workload and budget
| Workload | Pick | H100 cost per hour | Why |
|---|---|---|---|
| One GPU for a few hours | RunPod | $3.99 Secure Cloud, $2.69 Community Cloud | Single GPUs by the second, no contract |
| One GPU at the lowest rate | Vast.ai | From $1.79, median $2.33 | Host-set offers undercut fixed-rate clouds |
| Bursty inference or jobs | Modal | $3.95 | Scale to zero, $30 monthly credit on Starter |
| 8-GPU training run | Lambda | $4.59 per GPU | Fixed rate card, 8-GPU nodes, no egress fees |
| Whole-node training at scale | CoreWeave | $49.24 per 8-GPU node | Full nodes with 128 vCPUs |
| Preemptible training | Nebius | $4.50 on demand, from $0.79 preemptible | Dynamic spot tier with per-second billing |
| Notebook work with a free tier | Paperspace | $5.95 | Free M4000 tier, hourly machines |
| Listed rate with no egress fees | Crusoe | $3.90 now, $4.50 from Dec 1, 2026 | Minute billing, no setup fees |
Frequently asked questions
Which company is cheapest for an H100?
Vast.ai had the lowest H100 SXM on-demand offer at $1.79 per hour in a Nov 3, 2026 read, with a median of $2.33, but those are host-set offers that move with supply and demand. RunPod Community Cloud lists H100 SXM at $2.69 per hour, and Crusoe lists HGX H100 at $3.90 per hour until Dec 1, 2026.
Which is best for training a large model?
Lambda is the best fit for 8-GPU training runs, with H100 at $4.59 per GPU-hour and 1-Click Clusters from 16 to 2,000+ GPUs. CoreWeave suits larger whole-node training, with an 8-GPU H100 node at $49.24 per hour and reserved contracts quoted at up to 60 percent below on-demand rates.
Do any of these companies offer a free plan or free credit?
Yes. Modal Starter costs $0 per month and includes $30 of compute credit per month, and Paperspace has a free GPU tier with an M4000 machine. The other six have no free tier, and Nebius requires a $25 minimum first payment to enable billing.
What does turnkey AI compute include?
It means the provider supplies the GPU hardware, the drivers, the machine image, and the billing, so you launch a ready machine instead of buying cards and building a cluster. You still bring your own code, model weights, and data, and you pay separately for storage, network egress where charged, and any subscription plan a tier requires.
Why do prices differ so much for the same GPU?
Node size, billing model, and add-ons explain most of the gap. CoreWeave sells only fixed 8-GPU nodes, Lambda sells H100 SXM in 8-GPU nodes, RunPod and Vast.ai rent single GPUs, and Modal adds CPU and memory charges on top of the GPU hour. Marketplace offers move with supply, while fixed-rate cards change on set dates.
Can I rent a single H100 without a sales call?
Yes, on several providers here. RunPod lists a single H100 SXM at $3.99 per hour on Secure Cloud, Nebius lists a single H100 at $4.50 per GPU-hour, and Modal lists H100 at $3.95 per hour billed per second. CoreWeave is the exception, because its on-demand H100 is sold as an 8-GPU node.
Sources
- https://lambda.ai/
- https://lambda.ai/pricing
- https://www.coreweave.com/
- https://www.coreweave.com/pricing
- https://www.runpod.io/
- https://www.runpod.io/pricing
- https://nebius.com/
- https://nebius.com/prices
- https://www.crusoe.ai/
- https://www.crusoe.ai/cloud/pricing
- https://www.paperspace.com/
- https://vast.ai/
- https://vast.ai/pricing
- https://modal.com/
- https://modal.com/pricing
Prices and features were read on each company’s own website in November 2026. They change often, so confirm the current offer on the official site before you sign up or buy.