--:--:--
Forssa local time
Twelve megawatts connected.Thirty-six on the map.
A Finnish-owned, modular GPU data centre. Air and liquid-cooled colocation, and custom bare-metal GPU contracts — on Finnish soil, under Finnish law.
Forssa, Kanta-Häme, Finland60.8145°N 23.6214°E
Site masterplan
0.82 MW per MDC, drawn over the live site. Drag the timeline.
12 MW
Grid capacity
5 / 30
Modules on site
4.1 MW
Hosted today, 5 modules
Today
Timeline position
--
Outside air, monthly normal — free cooling for most of the year
Path A — 200 Gbit/s upstream
Path B — 200 Gbit/s upstream, diverse route
7.9 MW
Available for deployment now — 4.1 of the 12 MW connection already hosted
Traces are illustrative — the uplinks are real, and both are upgradeable on request.Status, Sep 2026 — December-phase modules in production, on schedule.
Two ways to take space
Every contract is quoted against a real module and a real power allocation. Nothing is oversold, and nothing is a slice of a shared floor.
Bare-metal GPU contracts
Dedicated GPU servers, specified around your workload and rented on your terms. Single tenant, no hypervisor, root access from the first hour.
- Built to your accelerator and node count
- Air or liquid-cooled platforms
- Terms from 6 months to 3 years
- Private VLANs and dedicated uplink
- Priced per build — tell us what you need
Reference node: 8× RTX PRO 6000 Blackwell — from €1.20 / GPU-hour, depending on contract duration
Get in touch
Colocation
Bring your own hardware into Finnish jurisdiction. Take a rack, a cage, or a dedicated 0.82 MW module with its own power feed and containment.
- Rack, cage or whole-module deployments
- Air-cooled and liquid-cooled module options
- Remote hands on site
- Cross-connects to both upstream paths
From €89 / kW / month — metered power at cost
Get in touch
The reference node, in full
from €1.20 / GPU-hour, by contract duration
- GPU
- 8× NVIDIA RTX PRO 6000 Blackwell Server Edition — 96 GB GDDR7 each, 768 GB per node
- CPU
- 2× AMD EPYC 9005-series, 64 cores per node
- Memory
- 1.5 TB DDR5 ECC
- Storage
- 2× 960 GB NVMe system, 4× 7.68 TB NVMe scratch — about 30 TB per node
- Network
- 2× 200 GbE east-west with RoCE, 25 GbE management, dedicated IPMI
- Power
- ≈ 7 kW per node — about ten nodes to a 72 kW air rack
- Access
- Bare-metal root from the first hour; Ubuntu with the CUDA stack preinstalled, or your own image
The most requested build. Other accelerators, CPU platforms and storage mixes are quoted per deployment.
Air modules. Liquid modules.
The yard runs two kinds of MDC side by side: air-cooled modules for conventional GPU fleets, and liquid-cooled modules for the densest racks. Your deployment lands in the type that fits it — and can step over to a liquid module as densities climb, without changing site, contract or address space.
Air-cooled modules
Sealed units with contained hot-aisle airflow, for conventional GPU servers and everything around them — storage, networking, head nodes. Proven, simple to service, ready for hardware you already own.
Up to 72 kW per rack on air
Liquid-cooled modules
Units plumbed for direct-to-chip liquid at Blackwell-class density, with in-row CDUs on the facility water loop. Rear-door heat exchangers available as a middle step.
Up to 220 kW per rack on liquid Coming Q4 2026
≤ 1.15Design PUE, annualised — free cooling does most of the year's work
99.9%Power availability SLA — 99.99% with generator-backed feeds on request
99.99%Network availability across the two diverse 200G paths
The matching towers are dry coolers — no water is consumed in cooling — and in Forssa the outside air does most of the year's work for free.
Capacity that arrives on a truck
The site is built from prefabricated 0.82 MW modules. Each one lands with switchgear, integrated UPS, containment and fire suppression already inside; the cooling unit is lifted onto the roof once the module is set. Commissioning takes weeks, not the years a conventional shell takes.
That is how a 12 MW connection grows to 24 MW inside eighteen months, with a 36 MW option behind it — and it is why we can hand you a dedicated block with its own power envelope instead of a share of a common floor.
Site facts
- Location
- Forssa, Kanta-Häme — roughly an hour and a half from Helsinki, an hour from Tampere and Turku. Off the coastal congestion, still inside the southern Finnish grid and fibre corridor.
- Power
- 12 MW of grid capacity connected today, with 4.1 MW running and hosted across five modules. The 12 MW upgrade goes live in December 2026, the full 24 MW is connected by June 2027, and the site holds an option to expand to 36 MW by the end of 2028. All of it on one of Europe's lowest-carbon electricity mixes.
- Resilience
- UPS is integrated into every MDC module, conditioning the feed and carrying the load through grid events; standby generators are available on request per deployment. Cooling and internet both run N+1 — a cooling train or an uplink can fail with no loss of service.
- Network
- Dual 200 Gbps internet on physically diverse paths, entering the site from opposite sides, so either can fail without taking the site with it. Higher capacity and private transport available on request. The meet-me room is carrier-neutral — bring your own carrier, or take blended transit from ours. Typical round-trip times from the site: Helsinki2 ms Stockholm8 ms Frankfurt24 ms Amsterdam26 ms London30 ms New York92 ms Averages on current routes; per-uplink figures are measured and published at contract.
- Cooling
- Air-cooled modules take racks to 72 kW today; liquid-cooled modules with direct-to-chip cooling to 220 kW per rack arrive in Q4 2026. Forssa's climate keeps mechanical cooling off for most of the year.
- Sustainability
- The cooling towers are dry — the site consumes no water for cooling. Power comes from one of Europe's lowest-carbon grids, typically under 100 gCO₂ per kWh, and waste-heat recovery into the local network is under study for the later phases.
- Build
- Thirty MDC units at full build, 0.82 MW each, deployed in either liquid or air-cooled configurations. Each unit operates with independent power distribution and containment, engineered for GPU rack densities rather than mixed enterprise IT.
- Jurisdiction
- Finnish-owned and Finnish-operated, so there is no foreign parent and no route for a non-EU authority to reach into the estate. Data stays in Finland, and so does the hardware it runs on.
- Operator
- The site is developed and operated by NCM Computing Oy (Business ID 3549914-7), Lavossuontie 191, 30420 Forssa — a Finnish-registered, Finnish-owned company.
- Compliance
- ISO/IEC 27001 — certification in progress. Operations run to the standard's controls today, with the audit ahead. Under GDPR we act as a processor for tenant workloads, with a data processing agreement available at contract, and operations are being aligned with NIS2. All processing sits under Finnish and EU law.
- Security
- The entire site sits inside a fully fenced, monitored perimeter under 24/7 on-site security — access control at the gatehouse, CCTV across the yard, remote hands inside.
Delivered with
Built and run with established partners — Forssa companies for the ground, grid and electrical work, and national names for energy supply and site security.
Questions we get
What is the minimum commitment?
Bare-metal GPU contracts run from 6 months to 3 years. Colocation starts at 12 months for rack and cage deployments; shorter pilots are quoted case by case.
How fast can you deploy?
Reference nodes from stock are typically configured, burned in and handed over within 2–4 weeks of contract. Whole-module deployments follow the module delivery schedule — additional 0.82 MW units commission in weeks, and the ramp to 24 MW is designed to stay ahead of tenant growth.
Who owns the hardware?
Either model works. On bare-metal contracts we own and operate the servers and you get single-tenant root access; in colocation you bring your own hardware into your own locked containment.
How is power billed?
Colocation is priced per kW of committed capacity, with metered power passed through at cost. Bare-metal GPU pricing is all-inclusive per GPU-hour — power, cooling and connectivity included.
What does remote hands cover?
Reboots, component swaps, cabling, media handling and visual inspections, around the clock through the on-site team. Response targets are agreed per contract.
Where does our data sit, legally?
In Finland, under Finnish and EU law. The site is Finnish-owned and Finnish-operated, and ISO/IEC 27001 certification is in progress.
From enquiry to energised
01Tell us the shapeAccelerator, node count or kilowatts, and term — the form below takes two minutes, and a person reads it the same day.Day 0
02Priced proposalA per-deployment quote with power, cooling and network laid out — nothing off a list.Within 2 working days
03Contract and buildNodes are configured and burned in, or your module, racks and cross-connects are allocated.On signature
04EnergisedRoot access handed over, or your racks powered and on the monitoring floor.Typically weeks 2–4
Get a quote
Tell us the shape of the deployment — accelerator, node count or kilowatts, and term — and a priced proposal comes back within two working days. 7.9 MW of connected capacity is ready for deployment now. If the shape is still fuzzy, send it anyway; scoping it with you is part of the job.
- Hardware, fastWe supply the servers as well as the space. Bare-metal GPU nodes arrive configured, burned in and racked on short lead times — no second vendor to wait on.
- Sharp pricingLow-cost Nordic power and a lean modular build go straight into the number — colocation from €89 per kW a month, Blackwell-class bare metal from €1.20 per GPU-hour depending on contract duration. Every quote is priced per deployment.
- Room to growNew modules commission in weeks, not years. Start with a rack, grow into a dedicated 0.82 MW block — the ramp to 24 MW stays ahead of you.
- Secure groundA fully fenced, monitored site under 24/7 security, on Finnish soil and under Finnish law.
Opens your mail app with the details filled in. A person reads it the same day — no ticket queue.