New B200 spot capacity is live in US East from $1.69 per GPU-hour. See availability
Pricing

Three ways to pay for the same GPU.

Spot is the same hardware sold from capacity we can reclaim with a 2-minute notice, 30–53% below on-demand. On-demand is never reclaimed. Reserved is a fixed monthly rate. All three are billed by the minute, in every region, with no egress fees.

Prices in USD per GPU-hour · updated recently · identical in 3 regions

Most deployed: H100 SXM 96 available
Spot<5% reclaimed over 30 days $0.55$0.0092 / min
On-demandNever reclaimed $0.89−38% on spot
Reserved1 to 12 months $0.65fixed rate
Persistent storageBilled while the volume exists $0.08per GB-month

Every model in the catalogue is priced the same way: three tiers, one rate per GPU-hour, billed by the minute, identical in 3 regions.

Entry price
$0.03/hRTX A4000 on spot, $0.0005 per minute
Price guarantee
−20%Below the cheapest price found across 137 providers
Billing unit
1 minuteHourly rate ÷ 60, no minimum duration
Egress
FreeBandwidth in and out is never billed
The three tiers

Pick the tier by what the job can tolerate, not by the price tag.

Every tier runs on the same nodes, the same images and the same API. The only difference is what we promise about keeping the GPU — and what that promise costs. Prices below are for one H100 SXM; every other model follows the same pattern.

SpotBest price
$0.55/ GPU-hour

$0.0092 per minute · −38% against our own on-demand price

  • Reclaimed with 2 minutes of notice, published on the metadata endpoint, a webhook and the console
  • Instance is stopped, never deleted — disk, data and reserved IP stay attached
  • Reclaim rate published per model: <5% on H100 over 30 days
  • Optional auto-relaunch on the next free capacity, same disk, same image
Launch on spot

For training with checkpoints, batch inference, sweeps, rendering, CI.

On-demandNever reclaimed
$0.89/ GPU-hour

$0.0148 per minute · still 23% below the cheapest on-demand price we found elsewhere

  • The instance runs until you stop it
  • Covered by the 99.9% monthly uptime target
  • Same per-minute billing, same images, same disks as spot
  • Move a disk from spot to on-demand in one click when a deadline appears
Launch on-demand

For production APIs, interactive sessions, anything with a hard deadline.

ReservedFixed rate
$0.65/ GPU-hour

From one month · 27% below on-demand, guaranteed capacity

  • GPUs held for you for 1 to 12 months, never reclaimed
  • Rate fixed for the whole term, billed monthly in advance
  • Spot and on-demand capacity stay available on top of the reservation
  • Multi-node reservations and private networking on request
How reserved works

For steady production load and teams that need capacity on a date.

Full price list

All 47 models, all three tiers.

Prices are per GPU-hour and identical in 3 regions. A node with 2, 4 or 8 GPUs costs the same per GPU as a node with one.

Specs and availability per model

47 GPUs shown ·

GPU Memory Spotper GPU-hour On-demand Reservedmonthly term Spot saving Per minutespot Action
Data center · 21 models
B200 SXM NewNVIDIA · Blackwell 180 GB HBM3e $1.69 $2.49 $1.85 −32% $0.0282 Deploy
H200 SXMNVIDIA · Hopper 141 GB HBM3e $0.75 $1.09 $0.79 −31% $0.0125 Deploy
H200 NVLNVIDIA · Hopper 141 GB HBM3e $0.69 $0.99 $0.69 −30% $0.0115 Deploy
H100 SXM Most popularNVIDIA · Hopper 80 GB HBM3 $0.55 $0.89 $0.65 −38% $0.0092 Deploy
H100 NVLNVIDIA · Hopper 94 GB HBM3 $0.49 $0.79 $0.59 −38% $0.0082 Deploy
H100 PCIeNVIDIA · Hopper 80 GB HBM2e $0.45 $0.69 $0.49 −35% $0.0075 Deploy
GH200NVIDIA · Hopper 96 GB HBM3 $0.49 $0.79 $0.59 −38% $0.0082 Deploy
A100 SXM 80GBNVIDIA · Ampere 80 GB HBM2e $0.19 $0.29 $0.19 −34% $0.0032 Deploy
A100 PCIe 80GBNVIDIA · Ampere 80 GB HBM2e $0.09 $0.19 $0.09 −53% $0.0015 Deploy
A100 40GBNVIDIA · Ampere 40 GB HBM2 $0.09 $0.17 $0.09 −47% $0.0015 Deploy
MI325XAMD · CDNA 3 256 GB HBM3e $1.09 $1.59 $1.19 −31% $0.0182 Deploy
MI300XAMD · CDNA 3 192 GB HBM3 $0.99 $1.45 $1.05 −32% $0.0165 Deploy
L40SNVIDIA · Ada Lovelace 48 GB GDDR6 $0.15 $0.25 $0.15 −40% $0.0025 Deploy
L40NVIDIA · Ada Lovelace 48 GB GDDR6 $0.09 $0.19 $0.09 −53% $0.0015 Deploy
A40NVIDIA · Ampere 48 GB GDDR6 $0.06 $0.09 $0.06 −33% $0.0010 Deploy
A30NVIDIA · Ampere 24 GB HBM2 $0.05 $0.08 $0.06 −38% $0.0008 Deploy
A10NVIDIA · Ampere 24 GB GDDR6 $0.09 $0.15 $0.09 −40% $0.0015 Deploy
L4NVIDIA · Ada Lovelace 24 GB GDDR6 $0.05 $0.08 $0.06 −38% $0.0008 Deploy
T4NVIDIA · Turing 16 GB GDDR6 $0.04 $0.07 $0.05 −43% $0.0007 Deploy
V100 32GBNVIDIA · Volta 32 GB HBM2 $0.06 $0.09 $0.06 −33% $0.0010 Deploy
V100 16GBNVIDIA · Volta 16 GB HBM2 $0.04 $0.08 $0.06 −50% $0.0007 Deploy
Workstation · 12 models
RTX PRO 6000 Blackwell NewNVIDIA · Blackwell 96 GB GDDR7 $0.19 $0.39 $0.29 −51% $0.0032 Deploy
RTX PRO 5000 Blackwell NewNVIDIA · Blackwell 48 GB GDDR7 $0.15 $0.29 $0.19 −48% $0.0025 Deploy
RTX PRO 4500 BlackwellNVIDIA · Blackwell 32 GB GDDR7 $0.09 $0.19 $0.09 −53% $0.0015 Deploy
RTX PRO 4000 BlackwellNVIDIA · Blackwell 24 GB GDDR7 $0.08 $0.15 $0.09 −47% $0.0013 Deploy
RTX 6000 AdaNVIDIA · Ada Lovelace 48 GB GDDR6 $0.09 $0.19 $0.09 −53% $0.0015 Deploy
RTX 5000 AdaNVIDIA · Ada Lovelace 32 GB GDDR6 $0.09 $0.17 $0.09 −47% $0.0015 Deploy
RTX 4500 AdaNVIDIA · Ada Lovelace 24 GB GDDR6 $0.09 $0.15 $0.09 −40% $0.0015 Deploy
RTX 4000 AdaNVIDIA · Ada Lovelace 20 GB GDDR6 $0.06 $0.09 $0.06 −33% $0.0010 Deploy
RTX A6000NVIDIA · Ampere 48 GB GDDR6 $0.09 $0.17 $0.09 −47% $0.0015 Deploy
RTX A5000NVIDIA · Ampere 24 GB GDDR6 $0.06 $0.09 $0.06 −33% $0.0010 Deploy
RTX A4500NVIDIA · Ampere 20 GB GDDR6 $0.06 $0.09 $0.06 −33% $0.0010 Deploy
RTX A4000NVIDIA · Ampere 16 GB GDDR6 $0.03 $0.05 $0.03 −40% $0.0005 Deploy
Consumer · 14 models
RTX 5090 NewNVIDIA · Blackwell 32 GB GDDR7 $0.09 $0.19 $0.09 −53% $0.0015 Deploy
RTX 5080NVIDIA · Blackwell 16 GB GDDR7 $0.06 $0.09 $0.06 −33% $0.0010 Deploy
RTX 5070 TiNVIDIA · Blackwell 16 GB GDDR7 $0.06 $0.09 $0.06 −33% $0.0010 Deploy
RTX 4090NVIDIA · Ada Lovelace 24 GB GDDR6X $0.06 $0.09 $0.06 −33% $0.0010 Deploy
RTX 4080 SUPERNVIDIA · Ada Lovelace 16 GB GDDR6X $0.06 $0.09 $0.06 −33% $0.0010 Deploy
RTX 4080NVIDIA · Ada Lovelace 16 GB GDDR6X $0.06 $0.09 $0.06 −33% $0.0010 Deploy
RTX 4070 Ti SUPERNVIDIA · Ada Lovelace 16 GB GDDR6X $0.06 $0.09 $0.06 −33% $0.0010 Deploy
RTX 4070 TiNVIDIA · Ada Lovelace 12 GB GDDR6X $0.05 $0.08 $0.06 −38% $0.0008 Deploy
RTX 4070NVIDIA · Ada Lovelace 12 GB GDDR6X $0.04 $0.06 $0.04 −33% $0.0007 Deploy
RTX 3090 TiNVIDIA · Ampere 24 GB GDDR6X $0.06 $0.09 $0.06 −33% $0.0010 Deploy
RTX 3090NVIDIA · Ampere 24 GB GDDR6X $0.05 $0.08 $0.06 −38% $0.0008 Deploy
RTX 3080 TiNVIDIA · Ampere 12 GB GDDR6X $0.04 $0.07 $0.05 −43% $0.0007 Deploy
RTX 3080NVIDIA · Ampere 10 GB GDDR6X $0.03 $0.06 $0.04 −50% $0.0005 Deploy
RTX 3070NVIDIA · Ampere 8 GB GDDR6 $0.03 $0.05 $0.03 −40% $0.0005 Deploy
  • Reserved prices apply from a one-month term; longer terms are quoted on request.
  • Persistent storage $0.08 per GB-month. Local NVMe scratch included with every node.
  • No egress fees, no charge for boot time, no charge for a reclaimed instance.
Estimate

What your job will actually cost.

Pick a model, a node size and how long you expect to run. The figures come from the same catalogue as the table above — nothing is rounded in our favour.

Tier
Running rate
$0.55/ hour

1× H100 SXM · $0.0092 per minute

  • Compute, 20 days × 8 h$0.00
  • Persistent storage$0.00
  • Total$0.00
  • Saved against our on-demand price$0.00
  • Saved against the cheapest price found elsewhere$0.00
Configure this instance

Compute is billed per started minute while the instance runs. Storage is billed per GB-month for as long as the volume exists, including while the instance is stopped. No other line appears on your invoice.

What the price covers

One line for compute, one for storage. That is the whole bill.

GPU, CPU, RAM
Included in the hourly rate. A H100 SXM node gives you 26 vCPU and 240 GB of RAM per GPU.
Local NVMe scratch
Included: 2 TB per GPU on that node. Wiped when the instance ends.
Persistent disk
$0.08 per GB-monthBilled while the volume exists, from 10 GB to 4,000 GB. How storage works
Bandwidth in and out
Free. No egress fees, no transfer quota, no charge for pulling your models or pushing your results.
Public IPv4
Included with every instance, kept while the instance exists — including after a spot reclaim.
Templates and images
Included. 9 ready-made environments, or bring any OCI image from a public or private registry.
Support
Included. Tickets answered by the engineers who run the nodes, 24/7.
Minimum credit
$40No top-up fee, no subscription, no expiry on unused credit.

A worked example

spot · H100

A fine-tuning run on one H100 SXM, 6 hours and 40 minutes, with a 200 GB disk kept for the month:

  • 400 minutes × $0.0092$3.68
  • 200 GB × $0.08 per GB-month$16.00
  • Egress of the resulting 40 GB checkpoint$0.00
  • Total for the month$19.68

The same run on-demand would be $21.92. At the cheapest on-demand rate we found elsewhere ($1.15 per GPU-hour) it would be $23.67 before that provider's egress fees.

Spend caps stop the bill, not just the alert.

Set a monthly cap in the console and we stop your instances before the cap is crossed. Disks survive the stop, so nothing is lost — you decide when to start again.

Price guarantee

At least 20% below the cheapest price we can find. Here is how we check.

A claim like this is only worth something if you can see the method. Ours is a script, and it runs before any price is published.

  1. We collect published prices

    Every model in this catalogue is matched against the public price lists of 137 GPU providers — hyperscalers, neoclouds and marketplaces — for the same GPU, in the same three forms: on-demand, spot and reserved. Index of September 2026, from a public price index (getdeploying.com) plus each provider's own page.

  2. We take the lowest, not the average

    For each tier we keep the minimum observed price, which is a far harder number to beat than a median. For H100 on-demand that minimum is $1.15, against a median of $2.69 across 87 providers.

  3. We price at 80% of it, then round down

    Our price for that tier is set at or below 80% of the minimum, then rounded down to a clean figure. Our H100 on-demand is $0.89 — 23% below the cheapest on-demand price in the index.

  4. The rule is enforced by the code, not by us

    The seeding script refuses to publish a catalogue where any node size breaks the rule, where spot is not below on-demand, or where a node size is not cheaper per GPU than the one below it. If the market moves, the prices on this page move with it.

Found a lower published price for the same GPU?

Send us the link through contact. If it is a real, public, comparable rate, we reprice — that is the whole point of the rule. You can also read the index yourself: our prices against the providers people ask about.

Multi-GPU nodes

The per-GPU price does not change when the node grows.

Node sizes for H100 SXM, with the CPU, RAM and local NVMe that come with each. Every model in the catalogue follows the same structure, up to 8 GPUs.

Node vCPU RAM Local NVMe GPU memory Spot On-demand Action
1× H100 26 240 GB 2 TB 80 GB $0.55/h $0.89/h Configure
2× H100 52 480 GB 4 TB 160 GB $1.10/h $1.78/h Configure
4× H100 104 960 GB 8 TB 320 GB $2.20/h $3.56/h Configure
8× H100 208 1,920 GB 16 TB 640 GB $4.40/h $7.12/h Configure

GPUs inside a node are connected by NVLink 4, 900 GB/s; PCIe Gen5 128 GB/s. Sizes vary by model — the exact figures for each one are on its page.

Billing FAQ

Billing questions, answered before you ask

How does per-minute billing work?

Billing starts the minute your instance becomes reachable and stops the minute you stop it. Every started minute costs the hourly price divided by 60 — $0.55 an hour on a spot H100 is $0.0092 a minute. There is no minimum duration, no rounding up to the hour, and boot time is not billed.

What is not included in the hourly price?

Only persistent storage, at $0.08 per GB-month, charged while the volume exists. Local NVMe scratch, CPU, RAM, the public IP, bandwidth in and out, snapshots of your own images and every template are included. There are no egress fees and no charge for a reclaimed instance.

Do I pay while an instance is stopped?

No compute is billed while an instance is stopped — that includes the time after a spot reclaim. The persistent disk attached to it keeps being billed at $0.08 per GB-month because the data is still stored, which is exactly what makes a resume possible.

Is there a minimum spend, a subscription or a commitment?

No. You add credit from $40, and it is drawn down per minute as you run. Nothing renews on its own. Reserved capacity is the only tier with a commitment, and it is explicit: one to twelve months at a fixed rate.

How do you keep prices 20% below everyone else?

We buy capacity that is already installed and not currently sold, and we do not run a sales team. The catalogue builder compares every node size against the lowest published price for the same GPU across 137 providers and prices each tier at or below 80% of it. The check runs again every time the catalogue is seeded, so a price that breaks the rule never reaches this page.

What happens if my credit runs out?

We warn you in the console and by email while there is still time to top up. If the balance reaches zero, running instances are stopped — not deleted — and your persistent disks are kept. Set a monthly spend cap if you prefer a hard ceiling.

Can I get an invoice?

Every top-up and every charge appears in Balance & billing in the console with its date, reference and amount, and can be downloaded. Add your company name and address in Settings and they are printed on the document.

Do you charge for the reclaim itself?

No. The 2 minutes of notice and the stop are not billed. You pay for the minutes the instance actually ran, and nothing else.

Still unsure about a number? Ask us — we would rather explain it now than refund it later.

Get started

Add $40 of credit and get 72 hours on an H100.

Pay as you go — no contracts, no minimum commitment. Add credit, launch, stop whenever you want.

Billed per minute from the moment the instance is reachable. Minimum credit $40, no subscription.