01Dedicated NVIDIA B200
Dedicated NVIDIA B200, not shared
Current-generation accelerators on bare metal. No shared tenant, so performance does not depend on your neighbours.
AI infrastructure from Spain:LLM API, sandboxes anddedicated GPUs, pay as you go.
Call open-source models through an API, run coding agents in a cloud sandbox, or reserve an NVIDIA B200 of your own.Sandboxes and GPUs in Spain, billed in euros, no contracts.
Activate with a card and you are up in under a minute. Use it for three minutes, pay for three minutes.
- cert.ISO 27001 and ENS Medium
- data centreSpain
- invoicein EUR
- computeNVIDIA B200
products
Three products, one account: which one do you need?
Each one solves a different problem. You can use one, two or all three from the same account, billed in euros.
Token Factory
An LLM API your app calls
For adding AI answers, summaries, document reading or text generation to a product you already have. Works with the OpenAI and Anthropic SDKs: change the base URL and the key.
Real example
a clinic answering appointment questions at any hour.
- The answer arrives piece by piece as it is generated, without waiting for it to finish.
- Open-weight models from several providers, all behind the same API key.
- You top up your balance in advance and it is drawn down by actual usage. No fees, no surprises.
from 0.06 €/ M input tokenscredit from 10 €
Sandbox
A cloud computer with Claude Code ready
For building and testing with AI agents without installing anything on your laptop. It pauses itself when you stop and keeps your files.
Real example
a team building and testing without depending on its own hardware.
- Both coding agents come installed and set up against Token Factory. You pick one when you create the Sandbox.
- Connect from your terminal over SSH with your key, with nothing else to install on your machine.
- When you are not using it, it suspends itself and stops counting active hours. When you come back, it starts in seconds with your files where you left them.
from 0.15 €/ active hour+ volume from €5/month
Dedicated GPU
An NVIDIA B200 of your own
For training, fine-tuning or running your own models. A hardware-isolated slice or the whole 180 GB card, shared with nobody, by the hour or by the month.
Real example
a company training its own model on private data.
- MIG splits the B200 into hardware-isolated slices of 23, 45 or 90 GB. You rent a slice or the whole card.
- NVIDIA's most advanced compute chip. It is the engine that runs the AI.
- The GPU's memory: the more there is, the larger the model that fits. FP4 is the number format that lets it be served faster.
from 1.10 €/ hourmonthly reserved −10%
The path
From your first call to production in an afternoon
If your code already talks to OpenAI or Anthropic, it already talks to GPU Flow. Then you grow without switching providers: one account covers all three stages.
- 01
Test the idea
Token Factory
Call a model from your app through the API and test the idea with the welcome credit.
- 02
Build it
Sandbox
A persistent environment to build in, with agent templates ready to go.
- 03
Take it to production
Dedicated GPU
Dedicated power for your real workload, by the month or by the year.
Three lines change.
Python
from openai import OpenAI
client = OpenAI(base_url="https://api.gpuflow.ai/v1", api_key="GPUFLOW_API_KEY")
r = client.chat.completions.create(
model="glm-5.3-nvfp4",
messages=[{"role": "user", "content": "Hello"}],
)use cases
What customers build with it
Answering customers around the clock
Clinics, shops and training centres answer questions about hours, bookings and orders at any time.
500 answers a day: about 1.43 € a month
Reading invoices and paperwork
Accountants and back offices hand it a scan and get the data back in order.
1,000 invoices: about 0.49 €
Writing in the brand's voice
Marketing and retail teams generate product descriptions, emails and posts that sound like them.
300 descriptions: about 0.03 €
Building with agents, without a server room
Developers and startups run Claude Code in a Sandbox and ship from there.
from 0.60 € an afternoon
Two questions
Which one do I need?
Whether you run infrastructure for a living or have never touched AI, two answers tell you where to start.
infrastructure
Why teams in Europe choose GPU Flow
Six reasons to bring your AI here: what hardware runs it, where it sits, how it connects, where your data lives, which certifications it holds and how it is invoiced.
02Machines in Spain
Machines in Spain, under GDPR
Sandbox and dedicated GPU run in a data centre on Spanish soil. Your prompts are never used to train any model.
03RDMA internal network
RDMA internal network
Low latency between nodes for multi-GPU and sustained throughput in production.
04DDN EXAScaler
DDN EXAScaler
High-speed parallel storage for datasets, checkpoints and persistent volumes.
05ISO 27001 and ENS Medium
ISO 27001 and ENS Medium
Operational certifications you can download. Standard data processing agreement ready to sign.
06Invoice in euros
Invoice in euros, with VAT
Deductible for EU companies and freelancers, valid for intra-EU reverse charge. No dollar conversion, no month-end surprises.
models
Open-source models, ready to call
Qwen, GLM, Gemma, Nemotron, ALIA, Whisper, Qwen3-TTS and ACE-Step. One API key for all of them.
Text and code
Chat, writing and code, with tool calling.
Reasoning
Models that think before they answer, with adjustable effort.
Vision and documents
Reading images, PDFs and screenshots, with OCR and document understanding.
Voice
Speech to text and text to speech, including voice cloning.
Music
Full songs from lyrics and a style prompt.
Spanish and co-official languages
ALIA, Spain’s public model from the Barcelona Supercomputing Center, trained in Spanish, Catalan, Basque and Galician.
Support
People on support
You send an email and a person replies within four working hours, in Spanish or English. For critical incidents, a call with an engineer.
Write to us- Machines in Spain
- A data centre on Spanish soil.
- Invoiced in euros
- Ready for your accountant, with VAT included and no currency conversion.
- ISO 27001 and ENS Media
- Downloadable certifications and a DPA ready to sign.
- Human support
- In Spanish and English, in under 4 working hours.
frequently asked questions
Questions before you start
Do I need a developer to use GPU Flow?
To connect the API to your own website or app, yes, for about an afternoon. You create a Sandbox from the dashboard in a few clicks; to get in you only need a terminal with SSH.
Is it compatible with OpenAI and Anthropic?
Yes. The API exposes the same endpoints; you change the base URL and the key and your code stays the same.
How does the welcome credit work?
Every new account gets welcome credit to spend on Token Factory. You activate it with a card, which is required to sign up; the €1 verification charge is refunded.
How do I pay, and do I get an invoice?
You top up credit in euros by card and it is drawn down by use. Every top-up comes with a Spanish invoice, valid across the EU. Prices include VAT; customers in the Canary Islands, Ceuta, Melilla or outside the EU, and EU businesses with a valid VAT number, pay without VAT.
What happens if my credit runs out?
The Token Factory API stops answering, with a 402 error, until you top up. Nothing is charged to your card automatically.
What happens if the environment goes idle?
We only charge for the hours the sandbox is active. After the idle time you configured (from 10 min to 4 h, or manual stop only) with no SSH activity or user processes, it suspends automatically and stops billing compute. Your persistent volume survives intact and is available again in 5-8 seconds on reconnect.
Can I cancel?
There is nothing to cancel: no subscription, no commitment. Stop using it and you stop paying, except for any disk or GPU reservation you keep. If you delete a Sandbox, its disk is not erased: it stays so you can reuse it in another Sandbox, and it is billed while it exists.
Do you have a public SLA?
Yes, we publish real-time status at gpuflow.ai/status. The contractual SLA with support response times is agreed in enterprise plans.
What does it mean that GPU Flow is in beta?
That the service is already in production and billed on real usage, but we are still polishing the platform: there may be the occasional interruption and the interface changes from one week to the next. Prices and your balance do not change because of the beta. The state of every service is published on the status page, and if something breaks you write to us at [email protected] and we sort it out.
Your AI infrastructure, sovereign and in euros.
Create your account, activate your welcome credit and make your first call in under a minute.
Building a European startup? Apply for up to €5,000 in credits