47s
Median time to first token-serving node
GPU compute · EU-based · SOC 2 in progress
Ship inference on dedicated GPUs, not shared guesswork.
Provision H100, L40S and A100 nodes in under a minute. Predictable per-second billing, private networking, and an API that behaves the same in dev and prod.
99.95%
Rolling 90-day control-plane uptime
3 regions
EU-West, EU-North, EU-Central
/sec
Billing granularity, no minimums
What you get
Built for teams that outgrew a single workstation.
Everything below is on by default. No support ticket to enable a feature you already paid for.
Bare-metal GPUs
Full-node access with NVLink and local NVMe. No noisy neighbours, no vGPU time-slicing between tenants.
Private networking
Every project gets an isolated /24 with wire-speed east-west traffic. Peer nodes without touching the public internet.
Snapshots & images
Freeze a node's disk mid-run and restore it elsewhere. Ship your own base images or start from our CUDA templates.
Autoscaling inference
Point a load balancer at a pool and scale replicas on queue depth. Scale to zero between bursts and stop paying for idle.
Data residency
Workloads and storage stay in the EU region you pick. Signed DPA and per-region audit logs available on request.
One API, two SDKs
REST plus Python and Go clients that mirror it exactly. What you script against staging is what runs in production.
Pricing
Per-second rates. The number you see is the number you pay.
On-demand pricing below. Committed-use and spot discounts apply automatically once your usage qualifies.
Instance
vCPU / RAM
VRAM
On-demand
Tier
gpu.h100.1NVIDIA H100
26 vCPU · 200 GB
80 GB HBM3
€3.90/hr
flagship
gpu.a100.1NVIDIA A100
18 vCPU · 128 GB
80 GB HBM2e
€2.20/hr
training
gpu.l40s.1NVIDIA L40S
12 vCPU · 96 GB
48 GB GDDR6
€1.15/hr
inference
gpu.a10.1NVIDIA A10
8 vCPU · 48 GB
24 GB GDDR6
€0.48/hr
dev
Regions
Close to your users, inside the EU.
Pick a region at launch or pin a project to one permanently. Inter-region private transit is free within the EU zone.
eu-nl-1 · Amsterdam
eu-north-1 · Stockholm
eu-central-1 · Frankfurt
eu-west-2 · Paris · Q3
eu-south-1 · Milan · Q4
Your first 20 GPU-hours are on us.
Spin up a node, run a benchmark, tear it down. No card, no sales call — just an API key and a region.