--- title: About GPUBank: the company, the GPUs and how we price them url: https://gpubank.io/about description: GPUBank is a GPU cloud run by GPU Compute OÜ, Tallinn. Ten GPU models in Paris, Frankfurt, Ashburn and Singapore, billed per second, all-in. generated: 2026-11-02T00:25Z publisher: GPUBank (GPU Compute OÜ, Tallinn, Estonia) summary_for_ai: https://gpubank.io/llms.txt full_text_for_ai: https://gpubank.io/llms-full.txt --- > Markdown version of https://gpubank.io/about, generated from the live page. GPUBank is a GPU cloud billed by the second at one all-in price per GPU (vCPU, RAM, NVMe, a public IPv4 and traffic included), in Paris, Frankfurt, Ashburn and Singapore. Prices are in US dollars, excluding VAT. About # We rent GPUs at the price we show. GPUBank is a GPU cloud run by GPU Compute OÜ, a company registered in Tallinn, Estonia. We rent ten data-center and workstation GPUs in four regions, billed by the second at one all-in price per GPU, and we publish how that price is built. 10 GPU models, from the RTX 4090 to the B200 SXM 4 regions: Paris, Frankfurt, Ashburn, Singapore 1,506 GPUs you can launch right now 63 GPU clouds compared on our site, with dated sources 1 s billing increment on every instance type 01 / Why we started ## We started as customers whose GPU bills ran 40 % over the list price. We started GPUBank after a year of renting GPUs for our own training jobs. The hourly price on the page was rarely what we paid. CPU and RAM came as extra lines, the public IP cost something, traffic was metered, a disk kept billing after the machine stopped, and a ten-minute test was rounded up to an hour. Each of those charges was documented somewhere, spread across five pages of pricing notes. So we built the opposite: one price per GPU that already covers the machine around it, a meter that counts seconds and starts only when the machine works, and a receipt that shows every one of those seconds. 1.4× what our own H100 bills averaged against the list price in 2025, once CPU, IP, traffic, idle disks and rounding were added. 02 / What we hold ourselves to ## Commitments you can check yourself. ### One price per GPU vCPU, RAM, local NVMe, a public IPv4 and traffic are in the hourly price. The persistent disk is the only other line. [See a full receipt](https://gpubank.io/pricing#included) ### Billing starts when the machine works The meter starts after the health check. A failed launch costs nothing. [How the meter works](https://gpubank.io/docs#quickstart) ### Credits without asking A host failure credits the interrupted session in full. A missed interruption notice credits 30 minutes. [Credits on the status page](https://console.gpubank.io/status) ### The stock you see is stock you can launch The board lists allocatable cards per region. When a region runs out, the line says queue. [The live board](https://gpubank.io/#board) ### Figures you can check Over 400 open models priced on each GPU, with the precision, the card count, the method and the command behind every figure. [The benchmarks](https://gpubank.io/#benchmarks) ### Contracts you read before paying Terms, privacy policy, DPA, sub-processors and company details are public and dated. [The legal pages](https://gpubank.io/legal) 03 / Security and reliability ## How your data and your uptime are protected. Reliability - 99.9 % monthly availability target per region for on-demand and reserved machines. - A host failure restarts your machine on a healthy host with the same disk, and credits the session in full. - Every regional incident is published with a post-mortem and the credits applied. - Planned maintenance is announced 7 days ahead and never stops a running on-demand machine without notice. - Uptime per region over the last 90 days is public on the status page. Security - TLS on every connection; our servers expose no public web port, traffic arrives through an encrypted tunnel. - Persistent disks encrypted at rest; customer-managed keys for reserved capacity. - GPU memory and local NVMe wiped between customers; no GPU is shared at the same time. - Production access limited to named engineers, logged and reviewed every month. - A breach affecting your data is notified to business customers within 48 hours. [Privacy policy, security section](https://gpubank.io/privacy#security) · [Data Processing Agreement](https://gpubank.io/dpa) · [Sub-processors](https://gpubank.io/subprocessors) 04 / Our business model ## Our business model. We earn the difference between what a GPU-hour costs us at our datacenter partners and the price on the board. That margin pays for the team, the control plane, support and the deposit bonuses, which is why a bonus arrives as credit and not as cash. We do - sell GPU time by the second - sell reserved capacity at a fixed price - publish our prices in a public JSON feed - invoice in US dollars, with your VAT number We do not - sell or rent your data - run advertising or third-party trackers - take referral fees from the providers we compare - charge for a launch that failed 05 / Infrastructure ## Where the GPUs are. We rent capacity from one datacenter partner per region and run everything above the hardware ourselves: the control plane, the machine images, the meter, the console, the API and support. The partner sees servers. Your account, your billing and your support tickets stay with us. Paris eu-paris 7 GPU models · 100 Gbps per node Data stays in the EU. Frankfurt eu-frankfurt 6 GPU models · 100 Gbps per node Data stays in the EU. Ashburn us-ashburn All ten GPU models · 100 Gbps per node Data stays in the United States. Singapore ap-singapore 5 GPU models · 100 Gbps per node Data stays in Singapore. Account data, billing and backups live on our servers in France, and emails are delivered by Namecheap Private Email. Disks stay in the region you choose and move only when you move them. 06 / The company ## The company behind GPUBank. **Company** GPU Compute OÜ, trading as GPUBank **Registered** 12 September 2026, Estonian Commercial Register **Registry code** 17984370 **VAT number** EE103096524 **Registered office** Müürivahe 62, 10135 Tallinn, Estonia **Management board** Thomas Begout **Team** Nine people: infrastructure and platform engineers, billing, and support working Tallinn and Paris hours **Support** Tickets from the console, answered seven days a week [Check the registration on the e-Business Register](https://ariregister.rik.ee/eng/company/17984370) · [Full legal notice](https://gpubank.io/legal) 07 / Milestones ## Milestones. 1. Mar 2026 First H100 SXM nodes running in Paris, used by the founding team for its own training jobs. 2. Jun 2026 Frankfurt and Ashburn added for a private test, with the H100 SXM at $3.29 an hour. 3. 5 Aug 2026 GPU Compute OÜ registered in Tallinn, Estonia. 4. 27 Aug 2026 Disks can move between regions from the console or the CLI. 5. 11 Sep 2026 Per-second billing on every instance type. 6. 22 Sep 2026 The console opens to everyone, first customers included. 7. 23 Sep 2026 The H100 SXM drops to $1.85 an hour, its fourth price cut since June. 8. Oct 2026 63 GPU clouds compared on our site, each figure with its source and date. 9. 4 Oct 2026 Over 400 open models sized and priced on all ten GPUs, from chat and code to video, speech and 3D. 08 / Press kit ## Press kit. GPUBank.io GPUBank.io #010E1A #4C8DFF #7FB0FF #FFFFFF GPUBank is a GPU cloud that rents NVIDIA and AMD GPUs by the second at an all-in hourly price, from $0.23 for an RTX 4090 to $4.15 for a B200 SXM, in Paris, Frankfurt, Ashburn and Singapore. It is operated by GPU Compute OÜ, Tallinn, Estonia. ## See the price, then see the receipt. [Open the console](https://console.gpubank.io/register) [View pricing](https://gpubank.io/pricing)