Launch pricing: every plan costs 30% less than the cheapest offshore competitor we track. See the benchmarkEvery plan 30% under the cheapest offshore host

OffshoreServ
Deploy a server

NVIDIA H100 SXM5 · 80 GB HBM3 · Hopper

Rent an H100 server.
80 GB of HBM3, by the month.

A dedicated NVIDIA H100 SXM5: 80 GB of HBM3 at 3.35 TB/s and the FP8 Transformer Engine, with 24 vCPU, 192 GB of RAM and 2 TB NVMe, for $581.99 a month, about $0.797 an hour around the clock. Need more? Two or four H100 in one server, linked by NVLink. Paid in crypto, no KYC.

  • 80 GB HBM3, 3.35 TB/s
  • 1, 2 or 4 GPUs, NVLink
  • Ready in 1 to 24 hours
  • No KYC, crypto only
See the servers What fits

Configure your H100 serverReady in 1–24 h

JurisdictionMoldova · Chișinău

PlanCompare plans

Billing cycle

$581.99/mo

Monthly · cancel anytime

Cheapest competitor $832.50−30%

No IDBTC, XMR, USDT +2No setup fee

01Plans & pricing

One, two or four H100 per server.

Every GPU is yours alone, with fast NVMe for datasets and checkpoints and an unmetered 1 Gbps port. Quarterly billing saves 5%, yearly 12%.

GPU Servers: 3 plans, monthly prices in USD
GPU VRAM CPU RAM Storage Locations Price Order
H100 80 GB Frontier training, SXM5 80 GB VRAM24 vCPU192 GB RAM2 TB NVMeMoldovaNetherlandsIceland 80 GB 24 vCPU 192 GB 2 TB NVMe MoldovaNetherlandsIceland $581.99/mo Cheapest competitor $832.50 Deploy H100 80 GB
2 × H100 160 GB HBM3, NVLink 160 GB VRAM48 vCPU384 GB RAM4 TB NVMeMoldovaNetherlands 160 GB 48 vCPU 384 GB 4 TB NVMe MoldovaNetherlands $1,096.99/mo Cheapest competitor $1,567.50 Deploy 2 × H100
4 × H100Cluster 320 GB HBM3, NVLink 320 GB VRAM96 vCPU768 GB RAM30 TB NVMeMoldovaNetherlands 320 GB 96 vCPU 768 GB 30 TB NVMe MoldovaNetherlands $2,348.99/mo Cheapest competitor $3,356.00 Deploy 4 × H100
  • RTX 4090, 5090, L40S, A100, H100
  • CUDA drivers preinstalled
  • Single and multi-GPU nodes
  • No KYC · BTC, ETH, XMR, USDT, SOL

Same 80 GB for less? The A100 80 GB costs $355.99/mo.

02Specifications

The H100, in numbers.

NVIDIA’s specifications for the H100 SXM5, and what comes with it in our servers.

The H100, in numbers.
Specification1 × H1002 × H1004 × H100
GPU memory80 GB160 GB320 GB
Memory bandwidth3.35 TB/s3.35 TB/s per GPU3.35 TB/s per GPU
Between the GPUsSingle GPUNVLinkNVLink
FP16 / BF16 tensor (dense)~989 TFLOPS~2 × 989~4 × 989
FP8 tensor (dense)~1,979 TFLOPS~2 × 1,979~4 × 1,979
CPU, RAM, storage24 vCPU · 192 GB · 2 TB NVMe48 vCPU · 384 GB · 4 TB NVMe96 vCPU · 768 GB · 30 TB NVMe
Price$581.99/mo$1,096.99/mo$2,348.99/mo
Per GPU-hour, non-stop$0.797$0.751$0.804

Hopper architecture, 16,896 CUDA cores and 4th-generation tensor cores with the Transformer Engine on every GPU.

03Sizing

What fits on one, two or four H100.

The highest precision at which a dense language model fits with an 8K-token context and 10% headroom, the figures of our VRAM guide.

What fits on one, two or four H100.
Model size1 × H100 (80 GB)2 × H100 (160 GB)4 × H100 (320 GB)
7–8B16-bit16-bit16-bit
13–14B16-bit16-bit16-bit
32B16-bit16-bit16-bit
70B4-bit8-bit16-bit
123BDoes not fit8-bit16-bit

Training needs far more memory than inference: a full fine-tune of a 7–8B model already takes about 16 bytes per parameter, so 2 × H100. How we size models

04Monthly or hourly

When a month beats paying by the hour.

At $581.99 a month, one H100 costs about $0.797 an hour if it runs non-stop. Against an hourly rental, the month pays for itself after this much use:

When a month beats paying by the hour.
If the hourly rate isOur month equalsBreak-even use
$2.00 per GPU-hour291 h40% of the month
$3.00 per GPU-hour194 h27% of the month
$4.00 per GPU-hour145 h20% of the month

Hourly rates are examples, not quotes from any provider. We bill monthly, quarterly (−5%) or yearly (−12%), from your balance.

05Use cases

Where an H100 pays off.

HBM3 bandwidth, FP8 and NVLink matter most for training and heavy serving.

  • Serving 70B models

    A 70B model at 4-bit on one H100, at 8-bit on two, or at full precision on four, behind a vLLM endpoint.

    vLLM guide

  • Fine-tuning

    LoRA and full fine-tunes on private data that never leaves the server.

    2 × H100

  • Training

    Research and training runs across four NVLink-connected H100 in one node.

    4 × H100

  • Batch jobs

    Embeddings, synthetic data and video generation that run for days.

    GPU images

06Jurisdictions

Where H100 servers run.

The plans above show where each server is available; prices are the same everywhere.

Compare jurisdictions

Latency figures are estimates from distance, not measurements. Every jurisdiction has the same prices. How we estimate latency

07Offshore, in writing

The rules, before you pay.

What we promise is written into our policies, not just our marketing.

  • US DMCA noticesNot actioned

    Answered with our policy, never enforced. Only a local court order, or a valid EU notice in EU locations, can require action.

    DMCA policy

  • IdentityEmail only

    No name, address, phone or ID document, ever. A private or disposable address is fine.

    Privacy policy

  • Payment5 cryptocurrencies

    Bitcoin, Ethereum, Monero, USDT and Solana, paid on-chain from any wallet to your balance. No card processor, no chargebacks.

    Crypto payments

  • TransparencySigned canary

    A PGP-signed warrant canary every quarter and a public count of every request we receive.

    Warrant canary

08FAQ

H100 servers, answered.

Another question? The full FAQ answers what people ask before an order: payments, privacy, complaints and support.

Read the full FAQ
H100 or A100?

Both have 80 GB. The H100 has about 1.7 times the memory bandwidth, FP8 support and faster training; the A100 costs less per month and is enough for fine-tuning and steady inference.

Are the GPUs linked by NVLink?

Yes, on the 2 × and 4 × H100 servers. That matters for training and for serving models split across GPUs.

Is this the SXM or the PCIe H100?

The SXM5 version, with 80 GB of HBM3 at 3.35 TB/s.

Can I rent it by the hour?

No. Billing is monthly, quarterly (−5%) or yearly (−12%), paid from your balance. At about $0.797 an hour around the clock, a month costs less than hourly rentals once you use the GPU for a few hours a day.

How fast is delivery?

Between 1 and 24 hours after you order, depending on the jurisdiction. The access details appear on the server’s page in your client area.

Can I get a refund?

GPU servers are not refundable once delivered, because the hardware is set aside for you. Check the specifications above before you order.

09Go further

Guides, policies and articles picked for this page, written by our team.

Hopper-class compute, without a cloud account.

  1. 1Pick one, two or four GPUs
  2. 2Pay in crypto, no ID asked
  3. 3SSH in within 1 to 24 hours
Configure your H100 server All GPU servers

From $581.99/mo · dedicated GPUs · CUDA preinstalled

Welcome back

Sign in to manage your servers and your balance.

No KYCHuman check by Cloudflare TurnstileNo tracking