Fast Lane Access to LUMI AI Factory 

The Fast Lane Access to LUMI AI Factory offers EU-based SMEs and startups accelerated, moderate-scale access to LUMI (up to 50.000 GPU hours). Paired with expert technical support, it enables users with validated ideas to iterate, optimize, and benchmark AI models efficiently. The resources are free of charge for civilian and innovation purposes of eligible companies.   

Service features

Services based on EuroHPC JU’s Fast Lane Access to LUMI AI Factory (for all EU-based SMEs & startups), including:

• Compute: Up to 50.000 GPU hours on the LUMI partition (AMD MI250x GPUs)
• Storage: Flexible and agreed individually; expanded project storage and scratch space allocated during onboarding
• Duration: 1, 2 or 3 months
• Cost: Free of charge for eligible SMEs and startups for innovation use. Free of charge for large businesses if they do open science and publish the results.
• Support:
  • Technical support including standardized onboarding, basic optimization guidance, and troubleshooting
  • Optional additional consultancy through other AI Factory services (e.g., AMD/NVIDIA porting, advanced inference optimization, data pipelines, hyperparameter tuning)

Intended Activities / Supported Workload

This package is suitable for iterative R&D, optimization, and scaling activities such as:

• Fine-tuning and iterative training of domain-specific models (CV, NLP, multimodal, LLMs)
• Performance benchmarking on AMD MI250X hardware across multiple experiments
• Data processing pipelines at scale
• Medium-scale model training and validation
• Porting and optimization from NVIDIA → AMD workflows
• Inference optimization and throughput testing
• Prototyping and early scaling of production AI tools (e.g., enterprise assistants, advanced classification, recommendation engines, real-time analytics)

Reporting Requirements

Standard EuroHPC JU Final Report is required:

• High-level project summary
• Brief description of services used
• Non-confidential technical outline with benchmarks
• Acknowledgement of LUMI AI Factory/EuroHPC resources
  No sensitive business information or trade secrets are required.

Access/Output Requirements

• Access Requirements: short application form based on a template explaining proposal relevance 
• Output requirements: Upon project completion, applicants must submit a Final Report within 3 months after the allocation end date, using the template available on the Fast Lane Access to AI factories – The European High Performance Computing Joint Undertaking (EuroHPC JU)​
• They commit to publishing results (excluding confidential parts) while referencing EuroHPC JU and the AI Factory; data and models remain user-owned for commercial use by SMEs/startups.

Target groups

• SMEs and startups from EU member states and associated countries
• Large enterprises if they use it for doing open science and publishing the results, otherwise available for a fee

Not eligible under this package:

• Academia and public organisations should use “AI for Science and Collaborative Projects” call, or other JU instruments.

Request process

Allocations are granted for a period of one, two or three months depending on duration needs. Resources can be requested on a single system only. The application is a rolling one with no cut-offs (as per: https://www.eurohpc-ju.europa.eu/fast-lane-access-ai-factories_en)

Pricing

The computing resources are provided open and free of charge to eligible AI startups and SMEs for innovation purposes. A short, non‑confidential report (for example a brief abstract of the use case and main outcomes) may be requested for public funding accountability, but detailed technical results and proprietary know‑how remain confidential and do not have to be disclosed.

Questions related to potential state‑aid implications, de minimis rules, or paid access for large companies are handled outside this Fast Track scheme under national calls and contracts, where confidentiality (including NDAs) is ensured according to national and site‑specific procedures.