HostKey

What HostKey actually is

HostKey is a European bare-metal and GPU hosting provider that sits in the middle ground between the big dedicated-server marketplaces and the GPU clouds. It sells three things: instant dedicated servers (pre-built boxes you can have within hours), custom dedicated servers (configured to order, up to dual sockets and 190 cores), and GPU servers — both bare metal and VPS — with professional NVIDIA cards for machine learning, rendering, and HPC. Its flagship facility is in Amsterdam at W. Frederik Hermansstraat 91, and the company also advertises capacity in the USA, Turkey, and Finland, with location pickers on its site also showing France, Germany, and the UK.

The GPU catalog is the reason most buyers look. Pre-configured and custom machines carry RTX A4000, A5000, A6000, 4090, 5090, and 6000 PRO cards, plus Tesla H100 80GB and A100 80GB for training-class workloads, with frameworks such as TensorFlow and PyTorch pre-installed. GPU VPS are also sold — the card is dedicated to the VM, not shared — from €70 per month. For teams that want an A100 without a CoreWeave-style contract negotiation, this is a credible route, and it belongs in any gpu-cloud comparison on price alone.

The rest of the catalog is conventional dedicated hosting done with modern hardware: AMD EPYC, Ryzen, and Intel Xeon builds, IPMI/iDRAC out-of-band management on every server, 1 or 10 Gbps ports, and BYOIP support — you can announce your own IP space or AS at no charge on the instant EPYC line. There is a server control panel and API, pre-installed apps on instant servers, and a "smart remote hands" service that delegates on-site tasks to HostKey engineers in any data center in the Netherlands.

Pricing: what you really pay

HostKey prices in euros and publishes them as "from" figures per server group. The floor is the instant dedicated line at €40 per month for basic pre-built configurations — genuinely cheap for bare metal with IPMI. Above that: previous-generation "sale" servers from €96, Ryzen and Core i9 boxes from €100, GPU servers from €118 (€90 is the advertised GPU entry point), custom single-CPU systems from €160, custom dual-CPU from €205, and hosting-provider bundles from €299 per month.

One concrete configuration published on the GPU pages: an Amsterdam dedicated server with 1 x GTX 1080Ti, Xeon E-2288G (8 cores), 32GB RAM, and 1TB NVMe SSD runs €170 per month, €160 per month on a six-month prepay, or €150 per month annually — a textbook example of the discount-for-commitment structure. The instant EPYC promotion lists a 32-core EPYC 9354 with 192GB RAM, 2 x 1TB NVMe, 10 Gbps and 50TB transfer from €299 per month, with 20% off and discounts up to 12% for BYOIP customers.

Two structural notes. First, billing is hourly and monthly on the instant line — a real advantage for bursty projects and one of the few European dedicated providers offering it. Second, traffic terms vary by product: advanced dedicated servers in the Netherlands get an unmetered 1 Gbps port, VPS get 3TB per month at 1 Gbps, and enterprise configurations are quoted in the €200-to-over-€1,000 range. The bare-metal page also frames three tiers — Economy from $99/month, Business from $299, Premium from $599 — in dollars, which is confusing when everything else is in euros; treat the euro figures as authoritative.

Performance & infrastructure

HostKey claims Tier III data centers and positions its network for European reach, with 1/10 Gbps unmetered ports and free basic DDoS protection (the site specifies that basic protection is free in Europe). Custom builds go deep: up to two CPUs, 190 cores, 4.6TB of DDR5, and 12 NVMe drives — configurations aimed at database, virtualization, and rendering workloads rather than website hosting.

The GPU line is the strongest technical asset. Cards are consumer and professional NVIDIA parts rather than data-center-only SKUs, which keeps prices down — an RTX A5000 build costs a fraction of an equivalent A100 instance at the GPU clouds — while the H100 and A100 options cover the training tier. Pre-installed frameworks (TensorFlow, PyTorch, Caffe) and the option to bring your own hardware configuration (HostKey says it will procure hardware specifically for your project) make it closer to a bespoke integrator than a hyperscaler.

We hold no benchmark results from our own test runs against HostKey hardware, so we will not invent performance numbers. What we note editorially: the price-per-GPU-dollar is the pitch, and the trade is that you get single-tenant hardware in a small number of locations rather than elastic, per-second GPU clusters with managed orchestration.

The remote-hands service is worth highlighting for colocators too: HostKey will send engineers into any data center in the Netherlands to handle physical tasks — swaps, cabling, diagnostics — which turns the company into a small managed-infrastructure partner rather than just a server seller.

Where it falls short

HostKey is a small provider, and it shows. Support is 24/7 but thin compared with larger hosts; the knowledge base exists but documentation depth is limited; and the marketing site mixes eras — location pickers still list Russia alongside the EU, USA, and Turkey, and some GPU pricing tables reference GTX 1080Ti-era cards next to 5090s. Nothing here is disqualifying, but buyers should confirm current stock and location by talking to sales before planning around a listing.

No uptime SLA percentage is published that we could find. "Tier III data centers" and "stable operation for high-load projects" are facility claims, not contractual commitments — for production ML pipelines we would want a written availability figure and a credit schedule.

Geographic concentration is the other limitation. Beyond Amsterdam, the footprint is thin; if your workload needs US West, Asia-Pacific, or multiple EU regions for redundancy, you are building that yourself. There is no free tier, snapshots and backup services are add-ons rather than inclusions, and pricing in mixed currencies (euro lists, dollar tiers) adds friction for finance teams.

A word on the buying process, since it shapes the experience: instant servers are self-service and quick, but anything custom — a specific GPU allocation, a procurement build, a non-standard storage layout — goes through sales, and the site nudges you there explicitly. Budget a few days of email for custom hardware, ask for current stock lists rather than trusting the published tables, and get the SLA question answered in writing before you sign. Providers at this scale will often commit contractually to numbers they do not advertise publicly; you have to ask.

Contract terms deserve the same scrutiny as price. Ask for the SLA, the hardware replacement window, and the escalation path in writing; at this scale the answers are often better than what the website publishes, but only if they are written down.

Who should pick this

Our pick when you need dedicated GPU hardware in Europe at market-beating prices and you are comfortable managing the box. ML teams fine-tuning or running inference on RTX-class cards, rendering studios, agencies doing GPU-accelerated video work, and hosting providers who want white-label dedicated capacity in Amsterdam are the natural audience. The hourly billing on instant servers also fits burst projects and short-term experiments.

It is equally a good fit for European buyers who want bare metal with BYOIP — bringing your own IP space or AS onto a €40-and-up dedicated server is a feature usually reserved for much larger contracts.

If you need managed operations, multi-region deployment, contractual uptime, or elastic GPU clusters billed per minute with an orchestration layer, the hyperscaler-style end of gpu-cloud serves you better. For conventional European dedicated servers without GPUs, compare our dedicated-servers picks, where operators with owned facilities and published SLAs compete on the same euros.

Key takeaways

• Amsterdam-based bare-metal and GPU host (W. Frederik Hermansstraat 91) with capacity in the USA, Turkey, and Finland.
• Instant dedicated servers from €40/month with IPMI, 1/10 Gbps ports, BYOIP, and hourly plus monthly billing.
• GPU catalog spans RTX A4000–6000 PRO and 4090/5090 to Tesla A100/H100, bare metal from €118 and GPU VPS from €70, with frameworks pre-installed.
• Free basic DDoS protection in Europe and unmetered 1 Gbps on advanced NL dedicated servers; VPS get 3TB/month.
• No published uptime SLA percentage; support and docs are small-provider scale; pricing mixes euros and dollar tiers.
• Best for ML/rendering teams and BYOIP users wanting cheap European GPU and bare metal; poor fit for managed, multi-region, or SLA-bound production.

Sources

• http://www.hostkey.com/
• https://hostkey.com/dedicated-servers/bare-metal-dedicated-servers/
• https://hostkey.com/dedicated-servers/powerful/
• https://hostkey.com/dedicated-servers/high-performance-hosting/
• https://hostkey.com/gpu-dedicated-servers/gpu-for-ai/