Back to directory
Infrastructure Clouds

Lyceum

EU-built GPU cloud for AI training and inference with per-second billing and data residency.

What makes Lyceum different

Lyceum is purpose-built for European AI teams that need GPU compute without the overhead and lock-in of hyperscalers. The platform operates exclusively from EU data centres with full GDPR compliance and data residency guarantees—a critical differentiator for organizations subject to European data sovereignty regulations.

The platform uniquely combines multiple execution models (serverless, dedicated, and VM-based) on a single infrastructure, allowing teams to optimize cost and latency for each workload. Unlike AWS or GCP’s opaque capacity allocation, Lyceum emphasizes transparent per-second billing with no minimum commitments and visible queue times and GPU utilization metrics (e.g., 76% average utilization, under one minute time-to-first-job).

Docker-native execution eliminates vendor lock-in—any container runs unmodified without proprietary SDKs. This design appeals to infrastructure teams managing heterogeneous workloads and enterprises requiring portability across providers.

Pricing model

Serverless Inference (pay-per-token):

  • Llama 3.1 8B: $0.10/1M input, $0.10/1M output tokens
  • Llama 3.1 70B: $0.35/1M input, $0.40/1M output tokens
  • Llama 3.1 405B: $1.00/1M input, $1.00/1M output tokens
  • Mixtral 8x22B: $0.50/1M input/output

On-Demand GPU VMs (per-second billing):

  • H100 (80GB): $2.49/hr
  • H200 (141GB): $3.19/hr
  • B200 (192GB): $4.29/hr
  • L40S (48GB): $1.19/hr
  • T4 (16GB): $0.39/hr

Long-term Contracts: Custom pricing from 3-month minimums.

Standout features: per-second (not per-hour) billing means jobs finishing early cost proportionally less; no hidden fees; visible pricing across 7+ competing providers via their GPU Pricing Calculator tool.

When it fits

  • EU-regulated AI teams: Organizations requiring GDPR compliance and data residency in Europe.
  • Cost-conscious ML teams: Per-second billing and transparent pricing reduce waste vs. hourly hyperscaler models.
  • Multi-GPU training at scale: 8–8,000 GPU clusters with InfiniBand for distributed training.
  • Development-to-production workflow: Notebooks for prototyping, serverless for quick iteration, dedicated endpoints for production inference.
  • Workload portability: Docker-native execution minimizes vendor lock-in for teams managing infrastructure across providers.

When it doesn’t

  • Non-EU data residency requirements: Lyceum operates exclusively from European data centres; no US or APAC regions available.
  • Non-GPU workloads: The platform is GPU-focused; general-purpose compute, databases, or networking services are out of scope.

Inclusion criteria

✓ Transparent pricing: Per-token and per-second rates published publicly with GPU Pricing Calculator comparing costs across competitors. (https://lyceum.technology/resources/pricing)

✓ Self-service signup: Direct account creation and dashboard access available at https://dashboard.lyceum.technology/.

✓ Public SLA/status: Platform uptime 99.9%, dedicated endpoints uptime 99.99% with visible metrics (latency 12ms, requests/min 2,847, GPU utilization 83%).