What makes Chapek unique?
Moving data costs nothing extra. There are no bandwidth charges and no egress fees: data in, data out and retrieval are included in the server price, with no monthly allowance to run out of. On AWS, Google Cloud and Azure, data sent to the internet costs $0.087 to $0.12 per GB above a free allowance of at most 100 GB a month (list prices, October 2026).
The rest is the combination. You rent a whole dedicated server with eight NVIDIA B300 or H200 NVL GPUs, 2 TB of memory and 120 TB of local NVMe, not a slice of one. It runs in the EU, under an EU contract, or a US contract if your organisation is established, incorporated or headquartered in the United States. Storage, 24/7 support and Chapek AI Studio come with every server, at one fixed price per term.
What is included in the price?
The price is per server and per term, fixed for the term. Storage, bandwidth and egress are included. Data transfer is unlimited: there is no charge for data in, data out or retrieval, no monthly allowance and no charge by volume, and nothing is invoiced in arrears. Fair use under the Acceptable Use Policy applies: for example, no use that goes beyond what you ordered, overloads the shared network or storage, or degrades the service for other customers. Creating, resizing or deleting virtual machines never changes what you pay.
Support is included as well, under the Service Level Agreement, and so is the platform: the portal, GPU telemetry and AI Studio.
See: Pricing & Servers · Service Level Agreement · Acceptable Use Policy
What exactly am I renting?
One whole, dedicated server, for a term you choose. Each server has eight NVIDIA GPUs (B300 or H200 NVL), 2 TB of system memory, 120 TB of local NVMe (about 90 TB usable) and a 100 GbE front-end connection. It is rented by one organisation at a time; no other customer's workload runs on it during your rental.
How you use the eight GPUs is up to you. Split the server into up to eight virtual machines, each with whole GPUs assigned to it: one machine with all eight, eight machines with one each, or any mix such as 4 + 2 + 1 + 1. Every virtual machine gets its own operating system, its own SSH login and its own public IPv4 address.
See: Pricing & Servers
Where do my servers and my data run, and who do I contract with?
Your server runs at a Chapek site in the Czech Republic, in the EU. Your virtual machines, and everything inside them, stay on your dedicated server and storage in that facility. Portal account data, usage records and GPU telemetry are kept in the EU. Payments for EU contracts go through a payment processor established in the EU.
When you create your account, you tell us where your organisation is established, incorporated and headquartered, and that decides who you contract with. If it is not established, incorporated or headquartered in the United States, your contract is with DCXPS SE, a European company based in Prague, under Czech law, and you pay in EUR; the terms that apply, including the data processing terms for personal data you process on your server, are published on this site. If it is, you contract with DCXPS Inc. under US law and pay in USD. Your server runs in the EU either way.
See: Sovereignty, Security and Assurance Annex · Data Processing Agreement
What can Chapek see inside my server?
Very little, by design. The layer Chapek keeps on the server creates and deletes your virtual machines, hands the GPUs to them and reports health. Chapek can see that a machine exists and whether it is running, the GPUs' own counters (utilisation, memory, temperature, power) and bytes in and out.
Chapek does not look at your data except in a few set cases. Under the contract, our staff access your servers only when you ask us to, when it is strictly necessary to provision, maintain, secure or support them, or when Czech or EU law requires it. Every such access is ticketed, logged and limited in time, happens only from the EU, and you can ask for the log.
Your private SSH key never reaches Chapek; you upload only the public half. The telemetry service inside a machine sends GPU counters and service metrics only. It does not read process lists, files, datasets, prompts or model outputs.
See: Data Processing Agreement · Sovereignty, Security and Assurance Annex · Privacy Notice
What happens if something breaks at 3 a.m.?
Support works in four layers that run in parallel, and monitoring and human engineers are on duty around the clock, on every day of the year. Autonomous agents monitor the platform continuously and restart or rebalance what they can. Human engineers are staffed 24/7 and commit to resolving a production-affecting issue in the software we operate (the operating system image, drivers, AI Studio, the platform, network and storage) within one hour of the ticket or alert. For a hardware fault, an on-site team arrives at the facility by the end of the next business day after the fault is confirmed and restores the server from spares held on site or, where one is available, moves you to an equivalent replacement server, with vendor parts replacement running in parallel on a next-business-day basis.
The availability commitment is 99.5% per month for each order. Downtime counted under the Service Level Agreement is credited automatically, without a claim from you, and a month below the commitment earns an additional credit. Planned maintenance, downtime you cause and the other exclusions listed in the Service Level Agreement are not counted. Credits go to your prepaid balance or against your next fees.
Can I run my own scheduler and software stack?
Yes. You have root on every virtual machine. A new machine boots with Ubuntu, the NVIDIA driver and your SSH key in place, so nvidia-smi works at first login. From there install what you need the way you would on any Linux server: your CUDA toolkit, PyTorch or JAX, Docker, SLURM, Kubernetes or Ray. The GPU fabric carries GPU-to-GPU traffic between your GPUs without configuration on your side.
If you would rather not run a stack at all, AI Studio deploys open models as an OpenAI-compatible endpoint on your server, and runs fine-tuning and batch jobs, without Kubernetes.
What is the minimum term, and how do I pay?
The minimum is one server for one month. You can also commit for a quarter or a year; longer terms get lower rates. We do not sell GPU-hours.
Each term is paid in advance: by card or by SEPA transfer. Reserving a server starts your order. The order is confirmed once your payment has arrived and the checks on your account are complete, and the term starts on the day the portal tells you the server is ready. If the capacity you ordered is no longer available, we decline the order and refund you within 14 days. Prices exclude VAT.
What happens to my data when I release a server?
When an order ends at the end of its term, your data stays available for you to copy off for at least 30 days, at no charge. If you end an order early and ask for your data to be erased rather than moved, it is erased within 30 days after the order ends, so copy anything you want to keep first. Either way, every disk that held your data is wiped by cryptographic erasure before the server goes to another customer. The portal will not release a server that still has virtual machines on it.
Each GPU then goes through Chapek's own checks before it is handed to the next customer: ownership is cleared, the device's identity is confirmed against our records and a diagnostic reads the counters a tenant cannot see from inside a machine. A GPU that fails or does not complete a check stays out of use until an engineer has looked at it.
Fees already paid for the current term are generally not refunded when you end an order early. The Terms of Service and the Sovereignty, Security and Assurance Annex list the cases in which we do refund.
See: Platform Terms of Service · Sovereignty, Security and Assurance Annex
When can I start, and what happens after I reserve?
Onboarding starts in December 2026. Until then, book a call with us, with an engineer or with sales, and get on the list; for larger capacity or multi-server setups, write to sales@chapek.io.
Chapek is for businesses. Once the portal opens you create an account for your organisation and tell us where it is established, incorporated and headquartered, which decides whether you contract in the EU or the US. Your organisation and its users are screened against sanctions lists after you create the account, before each order is provisioned, and periodically after that. Then you turn on two-factor authentication, add an SSH key and choose a server and a term. Reserving starts the order; the rental starts on the day the portal tells you the server is ready, after your payment has arrived, and you can create virtual machines straight away.
What certifications do you hold?
None yet, and we do not claim any. ISO/IEC 27001 certification of the information security management system covering the facility, the platform and the operation of the services is committed in the Sovereignty, Security and Assurance Annex, with a target of 31 December 2027. We do not claim a SOC 2 report. The facility is designed to be Tier II compliant and holds no Uptime Institute certification.
Until certificates exist, you can audit us instead: once a year, and again after a security incident or a government access request that affects you, you or an independent auditor can audit our compliance with the sovereignty criteria, the Annex and the data processing terms, on 20 business days' notice and at your own cost.
See: Sovereignty, Security and Assurance Annex · Master Services Agreement