Questions we are asked most, answered in full.

Typically within 5 business days after signing and the first payment, subject to availability. We use that time for a clean install and full hardware diagnostics.

Tell us the workload, the term and the preferred start date through the form on our homepage, by email at [email protected], or by phone or WhatsApp at +359 882 490 689 at any time. We usually reply within one business day.

Terms are flexible, and longer commitments get better rates. Tell us your plans and we will suggest the options that fit.

The workload, the term and the start date, your SSH public keys, the operating system you want and your company details for the agreement and the invoices. The full checklist is on the How it works page.

From $1.70 per GPU per hour, or about $9,930 a month for the whole server with eight GPUs. There are no extra costs: we pay for power, cooling, internet (1 Gbps, unmetered) and rack space, and support is included. The price is fixed for the term, longer commitments get better rates, and prices exclude VAT where it applies.

Power at any load, cooling, rack space, a dedicated 1 Gbps uplink with no traffic charges, a dedicated static public IPv4 address, replacement of failed parts and 24/7 response to critical incidents. There are no fees for setup, reboots or reinstalls.

We invoice monthly in advance and you pay by bank transfer. The price is fixed for the term and includes all operating costs. Prices are quoted without VAT, where VAT applies. Details are on the pricing page.

Eight NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs with 96 GB of GDDR7 each, 768 GB in total. Two AMD EPYC 9754 CPUs with 256 cores and 512 threads, 1.5 TB of DDR5-4800 ECC memory and two 4 TB Samsung 990 PRO NVMe drives. The full list is on the server page.

No. The RTX PRO 6000 Server Edition connects over PCIe 5.0 x16. For inference we recommend running models on 1, 2 or 4 GPUs and adding copies. Splitting one model across all eight GPUs works, but scales less efficiently than on NVLink systems.

Yes. Each GPU supports Multi-Instance GPU (MIG) with up to four isolated instances, and you are free to run containers or virtual machines on the host.

Any open-weight model that fits in 768 GB of GPU memory, for example Llama, Qwen, Mistral, DeepSeek or gpt-oss, as well as your own fine-tuned versions.

For longer commitments, we can build and host additional nodes to your specification together with our hardware partner, Aura Digital. Lead times depend on GPU supply, so ask us for a current estimate.

Yes. Mount your ISO through the BMC virtual media and install what you need. By default we deliver Ubuntu Server 24.04 LTS with NVIDIA drivers. Network boot (PXE) is not available, so virtual media is the recommended route.

Yes, through the remote console. For example, you can enable AMD SEV-SNP yourself, or we can do it before handover. We only ask you to leave the fan and cooling profile unchanged, as it protects the GPUs under full load.

A 10 GbE interface on a dedicated 1 Gbps uplink with no traffic charges, and a dedicated static public IPv4 address. The server firewall is yours to manage, and we open ports at the network level on request.

Yes. vLLM and SGLang both provide OpenAI-compatible endpoints, so existing clients and SDKs work after a change of the base URL.

99.5% monthly availability for the whole server, including its hardware. If it falls short, a service credit of 5, 10 or 25% of the monthly fee is applied automatically. See the SLA page.

Open a ticket and share the relevant logs. Once the fault is confirmed, we start the replacement within one business day. Until the GPU is replaced, you receive a pro-rata credit for it.

Within 4 hours, around the clock, when the server is down or unreachable. You get a direct phone line that is answered 24/7.

By email at [email protected], or by phone or WhatsApp at +359 882 490 689, at any time.

No. We have no accounts on your operating system. We monitor the hardware (power, temperatures, fans and hardware events) through the BMC. We use the remote console only at your request or to restore an unresponsive server, and we always tell you when we do.

This is an AMD platform, so Intel TDX is not available. AMD SEV-SNP is supported by the CPUs and the BIOS. A TPM 2.0 module can be fitted on request before handover.

The server is in Bulgaria, inside the EU, and you decide what data is processed and stored on it. Data processing terms under Article 28 of the GDPR are part of our agreement.

The drives are purged following NIST SP 800-88 before the server is used again. Failed drives never leave the data centre.

Yes. GPU clouds and resellers are welcome. We only ask that your terms with your customers include the same acceptable use and export control rules as ours.

NEOXIS Corp. is a Bulgarian joint-stock company founded in 2026. We own the hardware our customers rent, and we run it ourselves from Sofia. Our registered office is at 44 Hristo Botev Blvd., 6400 Dimitrovgrad, Bulgaria, UIC 208772721.

In Neterra's data centre SDC Stolnik, near Sofia, Bulgaria, inside the European Union. Neterra also provides our dedicated internet uplink.