Launch

Launch pricing: up to 74% below on-demand, fixed for your term — ends Oct 31, 2026 · 42d left Launch pricing ends Oct 31 · 42d left

See pricing

Pricing

GPU server pricing, per GPU per month

10 NVIDIA GPUs, from the L4 to the Blackwell Ultra B300, each as a dedicated bare-metal server at one flat monthly price. The host, the NVMe, 20 TB of transfer and 24/7 support are in the number; the number is fixed for your term; and every GPU listed is in stock.

  • From$169NVIDIA L4, 24 GB GDDR6, per month
  • Most popular$1,249NVIDIA H100 SXM, 80 GB HBM3, per month
  • vs hourly on-demandup to −74%−78% on a 12-month term
  • USD, excl. VAT · no egress fees · no hourly meter
  • Live in under 10 minutes after payment
  • No KYC · BTC, ETH, USDT
  • Stock counters updated 18 September 2026

All 10 GPUs, one flat price each

Filter by class, switch the term, and every price updates. Click a GPU for its full page: specs, price by term, market comparison and FAQ.

Show
Billing term
Launch pricing Launch rates, fixed for your whole term — ends Oct 31, 2026 (42d left). List prices apply from Nov 1. Inventory updated 18 Sep 2026, 21:00 UTC
GPU Memory Dedicated host & network Per GPU / month Deploy
NVIDIA B300New5 left Blackwell Ultra · NVLink · Frontier training & long-context inference 288 GB HBM3e 8 TB/s · 2.25 PFLOPS FP16 32 vCPU · 512 GB RAM · 4 TB NVMe25 Gbps uplink · 20 TB egress $3,790−34% $5.19/h effective $5,760 hourly on-demand Deploy NVIDIA B300
NVIDIA B2008 left Blackwell · NVLink · Frontier-scale training 180 GB HBM3e 8 TB/s · 2.25 PFLOPS FP16 32 vCPU · 384 GB RAM · 4 TB NVMe25 Gbps uplink · 20 TB egress $2,690−48% $3.68/h effective $5,220 hourly on-demand Deploy NVIDIA B200
NVIDIA H2009 left Hopper · NVLink · Large-model inference & training 141 GB HBM3e 4.8 TB/s · 989 TFLOPS FP16 24 vCPU · 256 GB RAM · 3 TB NVMe25 Gbps uplink · 20 TB egress $1,690−48% $2.32/h effective $3,270 hourly on-demand Deploy NVIDIA H200
NVIDIA A10021 left Ampere · NVLink · Proven workhorse for training 80 GB HBM2e 2 TB/s · 312 TFLOPS FP16 16 vCPU · 128 GB RAM · 2 TB NVMe25 Gbps uplink · 20 TB egress $649−49% $0.89/h effective $1,285 hourly on-demand Deploy NVIDIA A100
NVIDIA L40S17 left Ada Lovelace · Inference, video & graphics 48 GB GDDR6 864 GB/s · 362 TFLOPS FP16 12 vCPU · 96 GB RAM · 1 TB NVMe10 Gbps uplink · 20 TB egress $429−62% $0.59/h effective $1,124 hourly on-demand Deploy NVIDIA L40S
NVIDIA L426 left Ada Lovelace · Efficient inference & transcoding 24 GB GDDR6 300 GB/s · 121 TFLOPS FP16 6 vCPU · 48 GB RAM · 500 GB NVMe10 Gbps uplink · 20 TB egress $169−74% $0.23/h effective $657 hourly on-demand Deploy NVIDIA L4
NVIDIA RTX PRO 6000 BlackwellNew6 left Blackwell · 96 GB inference & rendering 96 GB GDDR7 ECC 1.6 TB/s · 125 TFLOPS FP16 16 vCPU · 128 GB RAM · 2 TB NVMe10 Gbps uplink · 20 TB egress $749−53% $1.03/h effective $1,606 hourly on-demand Deploy NVIDIA RTX PRO 6000 Blackwell
NVIDIA RTX 509011 left Blackwell · Fast single-GPU inference 32 GB GDDR7 1.79 TB/s · 105 TFLOPS FP16 8 vCPU · 64 GB RAM · 1 TB NVMe10 Gbps uplink · 20 TB egress $279−47% $0.38/h effective $526 hourly on-demand Deploy NVIDIA RTX 5090
NVIDIA RTX 4090Best value19 left Ada Lovelace · Diffusion, small LLMs, dev boxes 24 GB GDDR6X 1 TB/s · 83 TFLOPS FP16 8 vCPU · 64 GB RAM · 1 TB NVMe10 Gbps uplink · 20 TB egress $189−42% $0.26/h effective $329 hourly on-demand Deploy NVIDIA RTX 4090

Rolling month-to-month. Cancel with 30 days' notice. Prices per dedicated GPU per month in USD, excl. VAT. Savings compare the typical published hourly on-demand rate for the same GPU over 730 h. Need 8 or more GPUs? See 8-GPU clusters. “Left” counters are the unallocated units of the current hardware batch, updated 18 Sep 2026.

Terms

Every GPU, every term

The monthly rate, then what it becomes on a 3-, 6- or 12-month commitment. Per GPU, per month, in USD.

GPUMonthly3 months · −5%6 months · −10%12 months · −15%12-month, hourly equivalent
NVIDIA B300288 GB HBM3e $3,790 $3,601 $3,411 $3,222 $4.41/h
NVIDIA B200180 GB HBM3e $2,690 $2,556 $2,421 $2,287 $3.13/h
NVIDIA H200141 GB HBM3e $1,690 $1,606 $1,521 $1,437 $1.97/h
NVIDIA H100 SXM80 GB HBM3 $1,249 $1,187 $1,124 $1,062 $1.45/h
NVIDIA A10080 GB HBM2e $649 $617 $584 $552 $0.76/h
NVIDIA L40S48 GB GDDR6 $429 $408 $386 $365 $0.50/h
NVIDIA L424 GB GDDR6 $169 $161 $152 $144 $0.20/h
NVIDIA RTX PRO 6000 Blackwell96 GB GDDR7 ECC $749 $712 $674 $637 $0.87/h
NVIDIA RTX 509032 GB GDDR7 $279 $265 $251 $237 $0.32/h
NVIDIA RTX 409024 GB GDDR6X $189 $180 $170 $161 $0.22/h

Term prices are rounded to the dollar. Term plans are invoiced monthly at the discounted rate and run to the end of the term; monthly plans roll over and cancel with 30 days' notice.

Clusters

8-GPU nodes, per node per month

Eight GPUs on NVLink in one dedicated host, priced below eight single GPUs. Details, workloads and FAQ on the clusters page.

NodeGPU memoryHostPer node, per monthvs 8 single GPUs
8× H100 SXMNVLink · InfiniBand-ready 640 GB HBM3 NVLink 4 · 3.2 Tb/s InfiniBand · 192 vCPU · 2 TB RAM · 30 TB NVMe $7,990 −20%
8× H200NVLink · InfiniBand-ready 1.1 TB HBM3e NVLink 4 · 3.2 Tb/s InfiniBand · 192 vCPU · 2 TB RAM · 30 TB NVMe $10,990 −19%
8× B200NVLink · InfiniBand-ready 1.4 TB HBM3e NVLink 5 · 3.2 Tb/s InfiniBand · 224 vCPU · 3 TB RAM · 30 TB NVMe $18,900 −12%
8× B300NVLink · InfiniBand-ready 2.3 TB HBM3e NVLink 5 · 3.2 Tb/s InfiniBand · 224 vCPU · 4 TB RAM · 30 TB NVMe $27,900 −8%

Add-ons — all three of them

The base price is meant to be the whole price. Three things are optional because not everyone needs them.

Block storage$25 / TB / monthNetwork volumes in 1–8 TB steps, on top of the local NVMe
Extra IPv4$4 / address / monthUp to 4 per server; one IPv4 and a /64 IPv6 are included
Windows Server$29 / month2022 or 2025 with the RTX driver, on workstation-class GPUs

Included in every price

  • Single-tenant bare metal. Your GPU, CPU cores, RAM and NVMe are yours alone. No noisy neighbours, no oversubscription.
  • 10–25 Gbps uplink. Dedicated bandwidth per server with 20 TB of outbound transfer included every month. No egress fees.
  • Local NVMe storage. Fast scratch space for datasets and checkpoints, sized to the GPU. Expand with attached block volumes.
  • Ready-to-train images. Ubuntu with NVIDIA drivers, CUDA 12, cuDNN and Docker — or PyTorch, TensorFlow, vLLM, Kubernetes and Windows Server images at checkout.
  • Public IPv4 & IPv6. A dedicated IPv4 address and a /64 IPv6 block on every server, with reverse DNS on request.
  • DDoS protection. Always-on volumetric mitigation at the network edge, included at no charge.
  • Hardware replacement. Failed GPUs or drives are swapped within 4 hours, backed by our 99.9% uptime SLA with credits.
  • 24/7 engineer support. Reach an infrastructure engineer around the clock, with a 15-minute response target for incidents.

How billing works

  1. Pick a GPU, a term and a regionThe deploy page shows the exact monthly amount before you create an account. No card, no ID — an email address.
  2. Fund your balance in cryptoBitcoin, Ether or USDT (ERC-20 or TRC-20) to a fresh deposit address. The first invoice settles from the balance the moment the payment confirms on-chain.
  3. Live in under 10 minutesThe server is imaged, secured with your SSH key and handed over automatically. The clock on your term starts then, not at order time.
  4. Renewals settle themselvesThe renewal invoice is issued seven days ahead and charged to your balance on the day. Unpaid three days past due: suspended, never deleted. Auto-renew is per server, in the console.

Why monthly beats hourly for a busy GPU

An hourly rate is priced for a machine you switch off, and it includes the idle capacity the provider keeps ready for you. A monthly rate is committed usage, so the certainty comes back as a lower price: across the catalogue, up to 74% below the market-median on-demand rate over 730 hours, and up to 78% on a 12-month term.

The crossover for an H100 is around 373 hours a month. If your GPUs run less than that, use an hourly provider — our comparison pages say which one. If they run more, the monthly bill is smaller and, unlike a meter, known in advance.

Compare 12 GPU clouds on price

Pricing questions, answered

Is the price per GPU or per server?

Per GPU, per month — and the server comes with it. Each price on this page includes a dedicated bare-metal host sized for the GPU (CPU cores, RAM, NVMe), a 10–25 Gbps uplink with 20 TB of outbound transfer, a public IPv4, DDoS protection and 24/7 support. There is no separate host, storage or bandwidth line.

Do the prices include VAT?

No. Prices are in US dollars and exclude VAT. Invoices are issued in USD and settled from your prepaid crypto balance at the exchange rate at the moment you pay.

What happens when launch pricing ends?

Orders confirmed before Oct 31, 2026 keep their launch rate for the whole term, renewals included. From Nov 1, new orders are billed at list price — about 15% higher, shown on each GPU page. Existing servers are never repriced mid-term.

How do the term discounts work?

A 3-month term takes 5% off the monthly rate, 6 months 10%, 12 months 15% — for the whole term, with the same server and the same support. Monthly plans roll over month to month and can be cancelled with 30 days' notice; term plans run to the end of the term in exchange for the discount.

What add-ons are there?

Only three: block storage volumes at $25 per TB a month, extra IPv4 addresses at $4 a month each (up to 4), and Windows Server images on workstation-class GPUs at $29 a month for the licence. Everything else — NVMe, bandwidth, IPv6, DDoS protection, support — is included.

How is the on-demand comparison calculated?

We take the median published hourly on-demand rate for the same GPU across the public GetDeploying index (30 to 40 providers per model, read 18 September 2026) and multiply it by 730 hours — a machine kept on for a month. Below roughly 373 hours of use a month for an H100, hourly billing is cheaper and we say so.

Lock in your monthly GPU rate

Every GPU on this page is in stock. The price you see is the price you pay, every month, for the length of your term.