*[DediGPU](https://dedigpu.com/) — Markdown mirror of [https://dedigpu.com/about](https://dedigpu.com/about) · updated 2026-09-02 · index for LLMs: [llms.txt](https://dedigpu.com/llms.txt) · everything: [llms-full.txt](https://dedigpu.com/llms-full.txt)* About # We rent the machine at what it costs us. That is the whole company. DediGPU is a small infrastructure team with racks of its own in Reykjavík, Helsinki, Amsterdam and Singapore. We buy the cards, sign the power contracts, run the network, and publish the cost sheet of every server as its price. The margin lives in the options, and we say so. [See the servers](https://dedigpu.com/gpu) [How we price](https://dedigpu.com/pricing) - **13** GPU models, B300 to RTX 4090 - **4** regions, racks we own - **0 %** margin on the server - **36 mo** amortisation, straight-line - **99.9%** SLA, credited ×10 The idea ## A rack full at cost beats a rack half-empty at a markup. We came to this from running GPU fleets for other people: render farms, then the first wave of inference clusters. The pattern was always the same. Hardware that cost a known amount per month sat in racks waiting for a customer who would pay three times that, and while it waited it earned nothing. The margin was never the problem; the idle time was. So we inverted it. Every server on the catalogue is priced at its cost sheet: the card and its chassis share over 36 months, power at the rate we actually pay, rack and cooling, network, a reserve for failures, and the people. Rounded up to the next dollar. Nothing else. A tenant who pays that keeps the rack full, the power contract used and the next rack funded. What we earn comes from the things that are optional — extra storage, addresses, faster ports, private networks — and from buying power and hardware at a scale a single tenant could not. The consequence we like most is that we have nothing to negotiate. There is no sales team because there is no discount to give; the 3, 6 and 12-month terms are cheaper only because they cost us less. There is no quote because the price is on the page, per card, with its six lines, checked every month and dated. Principles ## Six things we will not change. - ### The price is the cost sheet Published, per card, six lines, rounded up. Reviewed monthly, with the date printed next to every price and a named cheapest competitor. - ### One tenant per server No hypervisor, no vGPU, no time-slicing. The whole card, the whole chassis, the IPMI console. We do not log in. - ### One term at a time Prepaid from a balance, renewed at a locked price, cancellable by letting the term run out. No contract, no notice period, no setup fee. - ### No identity check An email address and a password. Crypto in, crypto refunded to the address it came from. We hold as little about you as a host can and still bill. - ### Everyone is equal Same hardware, same network, same support, same SLA credits, whether you rent one card or five hundred. No priority tier exists. - ### Clean power, because it is cheaper Geothermal, hydro and wind on long contracts in the Nordics and on the North Sea. It is in the cost sheet, not in a sustainability report. What we own ## The racks, the cards, the contracts. We are not a marketplace and not a reseller. The HGX nodes, the PCIe chassis, the switches and the InfiniBand fabrics are ours, bought on our own invoices, which is what lets us print the purchase price in the hardware line. We lease cage space and power in four carrier-neutral data centres, on multi-year contracts, and run our own network from the cage to the exchange. Spares sit on shelves in every region, because a failed card at two in the morning is a swap, not a support case with a vendor. That reserve is 2 % of the card price per year, and it is a line on every cost sheet. Who we are ## Engineers, on call, in three time zones. A team of infrastructure and network engineers who have run GPU fleets since the render-farm days, spread between the Nordics, Western Europe and Singapore so that someone is awake next to each region. The people who answer tickets are the people who rack the servers. We do not publish names, and we do not ask for yours. We publish the numbers instead: the cost sheet, the status page, the incident history, the prices we found elsewhere. Judge us on those. So far ## Three years, four regions. 1. 2024 **Reykjavík.** First cage on geothermal power, A100 and RTX cards, monthly billing in crypto for a handful of render and inference tenants. The cost-sheet model written down for the first time. 2. 2025 **Helsinki and Amsterdam.** H100 and H200 HGX nodes, the first InfiniBand fabric, the console and the API. The catalogue published at cost with the cheapest competitor named next to every price. 3. 2026 **Singapore, Blackwell, MI300X.** B200 and B300 nodes, the AMD OAM platform, clusters to 512 cards, term discounts, the status page with its 90-day history, and the fourteen templates with model pre-loading. Press and partners ## Logo, facts, contact. Our name is DediGPU, one word, capital D and GPU. The logo below may be used to refer to us. Facts on this site may be quoted with a link; prices should be quoted with the date printed next to them, because they move monthly. - [Mark, 512 px PNG](https://dedigpu.com/assets/img/logo-512.png) · [Logo, 2× PNG](https://dedigpu.com/assets/img/logo@2x.png) - [Live status as JSON](https://dedigpu.com/status.json) · [API reference](https://dedigpu.com/api) - Questions from journalists and partners go through a ticket in the console, like everything else: [how contact works](https://dedigpu.com/contact). - **What we sell:** Dedicated GPU servers by the month, at their published cost sheet - **Where:** Reykjavík, Helsinki, Amsterdam, Singapore - **How to pay:** Bitcoin, Monero, Ethereum, USDT, Litecoin, TRON, Solana - **Identity check:** None, at any amount ## Judge us on the numbers. The cost sheet, the status page, the prices we found elsewhere. Then rent one server. [See the servers](https://dedigpu.com/gpu) [Trust and security](https://dedigpu.com/security)