Vast AI alternatives: official A100 prices, September 2026
On a USD comparison basis, RunPod's listed A100 80 GB rate leads five alternatives at $1.59 per GPU-hour. Vast.ai's $0.99 marketplace median is separate. Confirm billing currency, tax and allocation before treating the rate order as a purchase budget.
RunPod lists the lowest A100 80 GB on-demand price among the five ranked Vast.ai alternatives: US$1.59 per GPU-hour, according to official pricing pages checked 28 September 2026.1,2,3,4,5 Crusoe, Modal, CoreWeave and Lambda complete the ranking.2,3,4,5 Vast.ai’s A100 SXM4 marketplace median is lower at US$0.99/hour, but it represents host-set offers and stays outside the on-demand list-price ranking.6
TL;DR
- Compare the allocation as well as the unit rate: CoreWeave’s A100 entry is an eight-GPU node, and Lambda’s lowest A100 80 GB rate belongs to an eight-GPU instance.4,5
- Vast.ai’s marketplace medians describe available offers; they don’t establish the price of a particular on-demand rental.6
- Modal’s converted GPU-task rate needs a separate check of CPU, memory and execution terms.3 Its pricing page lists non-preemptible execution at 3x base prices.3
How this ranking was built
The ranking uses the lowest listed on-demand price per GPU-hour for A100 80 GB, checked 28 September 2026.1,2,3,4,5 The model treats dollar-only figures as USD and leaves quoted amounts unchanged, with no currency conversion or added tax. This is a comparison assumption, not confirmation of every vendor’s billing currency or tax exclusion. Nebius explicitly states USD and excludes applicable taxes; Lambda states that applicable sales tax, VAT or GST is additional.7,5 Confirm other vendors’ currency and tax treatment in the current quote. Node and per-second rates are converted to the same time unit.3,4
The provider comparison starts with eight alternatives; five have an eligible A100 price.1,2,3,4,5,7,8,9 Nebius, Together AI and DigitalOcean have no A100 listed in the checked pricing pages.7,8,9 Each company contributes one row, merging eligible variants and configurations at its lowest rate.1,2,3,4,5 Marketplace listings, spot and reserved prices are excluded; A100 40 GB cards are noted, not ranked.5,6 H100 prices appear alongside, without changing the A100 order.1,2,3,4,5 See our research methodology.
| A100 rank | Alternative | A100 variant | A100 on-demand (US$/GPU-hour) | A100 price basis | H100 on-demand (US$/GPU-hour) | H100 variant and basis |
|---|---|---|---|---|---|---|
| 1 | RunPod | SXM 80 GB; PCIe 80 GB ties | 1.59 | Listed per GPU-hour | 2.89 | PCIe 80 GB; listed per GPU-hour1 |
| 2 | Crusoe | PCIe 80 GB | 2.00 | Listed per GPU-hour | 3.90 | HGX H100 80 GB; listed per GPU-hour2 |
| 3 | Modal | A100 80 GB | 2.50 | GPU-task rate per second, converted | 3.95 | SXM5; GPU-task rate per second, converted3 |
| 4 | CoreWeave | A100 80 GB | 2.70 | Eight-GPU node, divided by eight | 6.16 | HGX H100; eight-GPU node, divided by eight4 |
| 5 | Lambda | SXM 80 GB | 2.79 | Per GPU-hour in an eight-GPU instance | 3.29 | PCIe 80 GB; single-GPU instance5 |
Both price columns are before tax and use the 28 September 2026 snapshot; H100 variants differ by company.1,2,3,4,5 Modal’s entries are base GPU-task rates, with CPU and memory priced separately.3
Which A100 alternatives have different rental terms?
CoreWeave and Lambda attach their ranked A100 rates to eight-GPU allocations, while Modal prices GPU tasks by the second with separate resource charges.3,4,5 A buyer should choose between these configurations before treating the hourly figures as interchangeable.
RunPod is the starting point for the lowest listed A100 80 GB rate in this comparison.1,2,3,4,5 Its PCIe and SXM entries tie, so choosing SXM does not change the ranked rate.1 The source separates Pods, Serverless and Clusters; the winning entry comes from Pods.1 Match the quote to the deployment model you intend to use.
Crusoe makes the A100 variant choice visible in its prices: PCIe 80 GB is US$2.00/GPU-hour and SXM 80 GB is US$2.30/GPU-hour.2 The ranking takes PCIe because the sorting rule selects the lowest eligible rate.2 If your requirement specifies SXM, use that quote when comparing suppliers.
Modal’s A100 80 GB rate is US$0.000694/second, converted to US$2.50/GPU-hour in the table; its H100 SXM5 rate converts from US$0.001097/second to US$3.95/GPU-hour.3 Both conversions multiply the listed rate by 3,600 and round to two decimals.3 Modal describes serverless autoscaling by request volume and lists separate CPU, memory and volume charges.3 Its non-preemptible execution option costs 3x base prices.3 Keep Modal on the shortlist if GPU tasks fit your deployment, then price the required execution mode and accompanying resources.
CoreWeave’s A100 rate comes from US$21.60 per eight-GPU node-hour divided by eight.4 The H100 companion figure similarly divides US$49.24 per node-hour by eight, rounded to US$6.16/GPU-hour.4 The node figures are the allocation prices used in this comparison.4 Use the whole-node quote when budgeting a node; the normalised column does not establish a single-GPU rental offer.4
Lambda lists US$2.79/GPU-hour for A100 SXM 80 GB in its eight-GPU instance configuration.5 Its single-GPU A100 entries in the capture have 40 GB memory, while its lowest H100 rate is a single-GPU PCIe 80 GB instance at US$3.29/GPU-hour.5 Lambda therefore needs a configuration check on both sides of an A100-to-H100 comparison.5
The Lambda allocation associations follow the captured 8x/4x/2x/1x tab order and the proportional resource quantities in those tables.5 Confirm the named configuration in a current quote; the converted GPU-hour figure does not create a single-GPU rental where the source describes a larger allocation.
How does Vast.ai marketplace pricing compare with on-demand?
Vast.ai reports marketplace medians of US$0.99/hour for A100 SXM4 and US$2.16/hour for H100 SXM in the 28 September 2026 capture.6 Each is the median of host-set marketplace offers, not a fixed list price.6
| Marketplace reference, not ranked | GPU | Listed median (US$/hour) | Price basis |
|---|---|---|---|
| Vast.ai | A100 SXM4, 80 GB | 0.99 | Host-set marketplace offers6 |
| Vast.ai | H100 SXM, 80 GB | 2.16 | Host-set marketplace offers6 |
Vast.ai’s price page presents both “from” and median figures and says offers update hourly.6 A median and the lowest on-demand list price answer different questions.1,6 The lower marketplace number alone cannot establish your rental cost.6
Marketplace describes who sets the price; on-demand describes a rental type.10 Vast.ai’s official instance documentation, captured on 1 October 2026, says hosts set prices and lists on-demand, reserved and interruptible instances.10 It describes on-demand as fixed pricing with guaranteed resources, reserved as discounted prepayment, and interruptible as access that may be paused.10 The marketplace model therefore does not mean every Vast.ai rental is interruptible.10
For a Vast.ai-versus-RunPod purchase, compare a specific Vast.ai on-demand offer with the matching RunPod Pod configuration. Include the GPU variant and rental terms; check storage and bandwidth before accepting the cheaper compute quote. Vast.ai’s October documentation separates compute, storage and bandwidth charges, and RunPod’s September price page says storage and deployment choices affect total cost.1,10
Which other Vast.ai alternatives list H100 prices?
DigitalOcean, Nebius and Together AI list H100 on-demand prices in the checked sources, even though none has an A100 entry in the supplied comparison.7,8,9 They remain H100 alternatives without an A100 rank.7,8,9
The following entries are alphabetical, not a second ranking. Prices are US dollars per GPU-hour before tax, checked 28 September 2026.7,8,9
| Alternative | H100 variant | On-demand (US$/GPU-hour) | Qualification |
|---|---|---|---|
| DigitalOcean | HGX H100 | 4.41 | Listed on-demand GPU Droplet rate9 |
| Nebius | HGX H100 | 3.85 | September rate; captured page announces US$4.50 effective 1 October 20267 |
| Together AI | HGX H100 | 3.99 | GPU cluster on-demand rate8 |
Together AI’s cluster rate is separate from its dedicated-inference pricing and reserved-capacity rates.8 DigitalOcean likewise lists separate reserved and spot plans.9 Use the same rental category throughout a comparison. For a GPU-specific comparison, see H100 rental prices by provider.
Why do published GPU prices disagree?
Published GPU prices can refer to different variants, allocation sizes or rental categories, including on-demand and reserved capacity.1,4,5,8 Comparing the labels beside the figures is the first step toward resolving a disagreement.
Lambda illustrates the memory boundary: its A100 40 GB rate starts at US$1.99/GPU-hour, while the ranked A100 80 GB configuration is US$2.79/GPU-hour.5 A100 prices quoted without memory and allocation details can describe different purchases.5
The date can also change the answer. Nebius’s page checked on 28 September already lists an H100 increase from US$3.85 to US$4.50/GPU-hour effective 1 October 2026.7 The H100 companion values retain the September snapshot; a new quote should use the rate that applies when you rent.7
The Vast.ai billing documents used for rental terms and charges were captured on 1 October 2026.10,11 Those later documents explain the billing model; they do not update the 28 September GPU-price snapshot.6,10,11
What the ranking cannot tell you
Glamdring Research ranks reported A100 prices only; the comparison does not measure workload performance, service reliability or the total cost of completing a job.
The evidence consists of company pricing pages, with different GPU variants and allocation bases preserved in the tables.1,2,3,4,5 A listed rate does not establish that the required configuration is available: Lambda describes self-serve instance access as first-come, and Vast.ai says offers change hourly.5,6
For a buying decision, request the required GPU variant, allocation size, region and rental type. Confirm availability and quote the accompanying storage, transfer and resource charges. Run your own workload before treating the rate order as a cost-to-completion order.
Refresh the comparison when an official pricing page changes. Explore more compute infrastructure research, or subscribe to our research.
Frequently asked questions
Why is Vast.ai so cheap?
Vast.ai attributes its pricing to competition between hosts in a supply-and-demand marketplace.10 Its official instance documentation captured on 1 October 2026 says hosts set prices based on regional demand, operating costs, machine specifications and competing offers.10 That explains Vast.ai’s stated pricing mechanism; a specific offer still needs its own configuration and billing check.10
Does stopping a Vast.ai instance stop all charges?
No. Vast.ai’s official payment and billing documentation captured on 1 October 2026 says stopping an instance does not avoid storage costs.11 Active rental is charged while the instance is active/connected; storage is charged while the instance exists and is online, including stopped states.11 The same document exempts offline instances from active-rental and storage charges, while bandwidth is charged for bytes transferred regardless of state.11 Check the instance state and its price breakdown when closing a rental.
Do you need credits before renting on Vast.ai?
Yes. Vast.ai requires prepaid credits for GPU rentals, according to its official billing documentation captured on 1 October 2026.11 It provides one-time credit purchases and autobilling with a saved credit card.11 When the balance reaches zero or below, instances stop automatically and deletion risk depends on the payment method and account history.11 A GPU-hour quote should therefore be checked alongside the account’s payment and storage arrangements.
Sources checked
Company-owned pages establish what a company says. They do not prove a market conclusion. Each source is dated so readers can judge each claim.
