Skip to main content

Coming Soon. Free GPU Tier

50 GPU hours a month. Free, every month.

A 24GB slice of an RTX 6000 Pro Blackwell GPU, running on EU sovereign infrastructure. Prototype, fine-tune and run real models. Renews monthly.

Join the Waitlist
Sovereign European Cloud that Heats Homes

How It Works

Three simple steps from signup to building

Step 1

Verify once
Sign up and complete a €0 card check or 1-cent iDEAL verification from a non-disposable card. No charge, no subscription. Just proof you're a person, not a mining farm.

Step 2

Deploy in minutes
Pick a one-click image or start from a clean server. SSH in, pull your model, and go. Your 50 free hours only count while a slice is provisioned.

Step 3

Build every month
Your allowance renews monthly. Stay inside it and pay nothing, ever. Outgrow it and keep working at €0.49 per hour, or move up to dedicated GPUs.

Free? What's the catch?

Fair question. Our servers live inside Dutch apartment buildings, care homes, and swimming pools, where the heat they produce warms real showers. So while an idle GPU is still producing useful heat, we would rather you were building on it.

Build something real, convert your compute into hot water, and if you outgrow the free hours we will be delighted. Everyone's shower stays warm either way.

Run Open Models

Llama 3.1 70B, Mistral 7B, Qwen 2.5—run inference at full speed

Generate Images

SDXL, Flux.1 dev—create production-quality visuals

Fine-Tune Models

LoRA fine-tuning for your custom datasets

Real Server Access

SSH, Docker, full root—it's your machine

Start with 50 free hours

Join the waitlist, first 100 signups get priority access as soon as we launch.

Join Waitlist

One-click images we're building for launch and beyond

Ready at Launch

Open WebUI

Open WebUI + Ollama

Your private ChatGPT. Llama 3.1 preloaded; pull Hermes 3, Qwen, Mistral with one command

vLLM

vLLM

An OpenAI-compatible API on your own server; point your existing SDK at it

JupyterLab

Notebook (JupyterLab + PyTorch)

Colab, except it's your server: no disconnects, disk that persists, root underneath

On the Horizon

ComfyUI

ComfyUI

Soon

SDXL + Flux image generation, workflows and LoRAs that stay put

Unsloth

Unsloth

Soon

QLoRA fine-tuning of 8B models. On a free tier. Yes really

Something else?

Tell us on Mastodon or LinkedIn Help decide what we build first!

What you get
What you get Good to know
50 GPU hours, every month. Wall-clock metered while your slice is provisioned, just like on-demand billing. The allowance resets monthly and never expires mid-cycle.A card check is required from a non-disposable card. A €0 card verification or 1-cent iDEAL payment. This is what keeps crypto miners out and the slices free for everyone else.
A dedicated 24GB VRAM slice of an RTX 6000 Pro Blackwell. Big enough for 7B to 13B models, LoRA fine-tuning, and SDXL or Flux image generation.Usage past 50 hours is billed automatically at €0.49 per GPU hour to the card on file. New accounts start with a low spend cap that rises after your first payment clears, so a forgotten job cannot run up a large bill.
Real server access. SSH, Docker, and root. It behaves like your machine, not a notebook that disconnects.Free slices are preemptible. There is no SLA on the free tier. If your slice sits idle we warn you, then reclaim it. We reclaim idle slices rather than billing them.
One-click images. Open WebUI + Ollama, vLLM, and JupyterLab + PyTorch ready at launch. ComfyUI, Unsloth, and Whisper on the roadmap.It is a playground, not production hosting. 50 hours is real development time, but a full month is 730 hours. If you need always-on capacity, our paid GPUs are one click away.
EU sovereign by design. Dutch-owned infrastructure in Amsterdam. Your data and workloads stay under EU jurisdiction, outside the reach of the US CLOUD Act.Our freemium model is a pilot, to test if it provides real value for our users. We retain the right to end this model at any time. All standard Leafcloud T&Cs apply.

Common Questions

Everything you need to know about our free GPU tier

Our servers don't live in traditional datacenters. They're installed inside Dutch residential buildings where their waste heat feeds the hot-water systems for showers and heating. The heat is metered, TNO-audited, and measured as the gas equivalent of heat delivered. Your GPU workload literally heats Dutch homes. Learn more about how we make cloud computing truly green.

Yes. 50 hours on a 24GB GPU slice lets you run Llama 3.1 8B at full precision, generate images with SDXL, or fine-tune models with LoRA. You get a real server with SSH access and persistent storage—not a notebook environment. For workload specifics and what fits in 24GB VRAM, see our RTX 6000 Blackwell GPU specifications.

No. Leafcloud is a Dutch company running on Dutch infrastructure with no US parent entity. Unlike EU regions of US-parented clouds (AWS, Azure, GCP), there is no CLOUD Act path to your data. Your AI workloads and model weights stay under EU jurisdiction. Learn more about our approach to digital sovereignty.

Free-tier GPU slices are preemptible: we can reclaim them with notice when paying customers need capacity. You get no SLA or uptime guarantee. If you need guaranteed capacity, reserved instances, or production SLAs, our paid GPU tiers provide those guarantees on the same infrastructure.

Once you reach 50 hours, your instance pauses. Billing never starts without your explicit go-ahead. If you choose to continue, you pay €0.49 per GPU-hour beyond the free tier. Important: you get one account per person, and free hours don't stack or carry over between months.

We require a €0 card check or 1-cent iDEAL verification to unlock your free tier. This one-time verification prevents abuse and ensures fair access for everyone. No charge, just verification. Once verified, you get 50 free GPU-hours every month, automatically renewed.