Lyceum Cloud

On-demand GPUs, billed per second.

Access NVIDIA B200, H200, H100, A100, and L40s GPUs instantly. No contracts, no idle spend - just submit your job and Lyceum handles the rest.

Start running jobs Talk to sales

// Why Lyceum Cloud

GPU infrastructure without the infrastructure headaches.

Instant availability

No procurement queues, no capacity planning. Select a GPU type or let Lyceum's AI choose the optimal one, and your job starts within seconds.

Per-second billing

Pay only for what you use, down to the second. Jobs that finish early don't cost you for time you didn't need.

Automatic hardware selection

Not sure which GPU to pick? Lyceum analyses your workload at the compiler level and recommends - or automatically selects - the most cost-efficient configuration.

Docker-native

Submit any Docker container. No proprietary SDKs, no vendor lock-in. If it runs in Docker, it runs on Lyceum.

Secure by default

Every job runs in an isolated environment. Your data and code never touch other tenants. Optional confidential computing with Intel TDX for regulated workloads.

Global availability

GPUs across multiple data centres in Europe, with expansion planned. Low-latency access for EU-based teams with full GDPR compliance.

// Available GPUs

Choose your hardware.

From cost-efficient inference on L40s to frontier training on B200s - or let Lyceum pick for you.

GPU	VRAM	Price / hour	Best for
NVIDIA B300	288 GB HBM3e	$8.49	Frontier model training, large-scale inference
NVIDIA B200	192 GB HBM3e	$8.89	Large-scale training, frontier models
NVIDIA H200	141 GB HBM3e	$4.99	Training, large inference workloads
NVIDIA H100	80 GB HBM3	$3.59	Training, fine-tuning, high-throughput inference
NVIDIA A100 80GB	80 GB HBM2e	$2.00	Fine-tuning, medium inference, general ML
NVIDIA L40s	48 GB GDDR6X	$1.69	Inference, image/video generation

Per second billing Multi-GPU and multi-node configurations available for all GPU types. InfiniBand interconnect available on H100, H200, and B200 clusters.

// How it works

From code to running job in three steps.

Submit your job

Push a Docker container via the CLI, Python SDK, VSCode extension, or web dashboard. Specify your requirements - or don't, and let Lyceum decide.

lyceum docker run my-training:latest --machine gpu.h100

Lyceum matches hardware

Our AI analyses your workload's compiler representation to predict runtime, memory needs, and optimal GPU configuration. You get a cost and time estimate before the job starts.

Results delivered

Your job runs on optimally matched hardware. Monitor progress in real-time via the dashboard. Outputs are stored and accessible immediately on completion.

// Scale out

Multi-node training, handled for you.

Need more than one GPU? Lyceum provisions multi-GPU and multi-node clusters with high-speed InfiniBand interconnects. Distributed training frameworks like DeepSpeed, FSDP, and Horovod work out of the box.

Automatic InfiniBand/RDMA configuration
Up to 8x GPU per node, multi-node scaling on demand
NCCL optimised networking
No manual cluster setup - just specify the number of GPUs

Node 1

8x H100

Node 2

8x H100

Node 3

8x H100

Node 4

8x H100

// Integrations

Works with your existing tools.

CLI

lyceum command line tool for scripting and CI/CD pipelines

Python SDK

Submit jobs programmatically from notebooks or scripts

VSCode Extension

Run and monitor GPU jobs without leaving your editor

Docker

Any Docker container, any base image, no modifications required

Web Dashboard

Visual job management, monitoring, and cost tracking

Start running GPU jobs in minutes.

No credit card required. Sign up and pay only for what you use.