The first three months will be dedicated to establishing the fundamental structures and processes necessary for the success of the AI factory.
AI Factory
Project background
The SLAIF (Slovenian Artificial Intelligence Factory) project and the investment in the acquisition of a new supercomputer are based on the successful implementation of the HPC RIVR project, which was the basis for the acquisition of the most powerful HPC Vega supercomputer in Slovenia.
Existing supercomputer HPC Vega and SLING
HPC Vega has been operational since 2021 and is expected to remain in use until the acquisition of a new supercomputer in 2027. The maintenance and management of HPC Vega and user support are carried out by IZUM in cooperation with partners SLING (Slovenian National Supercomputing Network), primarily the Jožef Stefan Institute (IJS). SLING, which has included all major organizations involved in the field of high-performance computing (HPC) since 2018, will take over supporting users in using the new supercomputer, which will be intended for the development of AI solutions and services.
Due to the extraordinary success and high utilization of the HPC Vega supercomputer by both Slovenian and European researchers, Slovenia, thanks to the professional competence and expertise of all those involved in the development, maintenance and operation of the supercomputer, has gained the trust of the EuroHPC JU, which is established and funded directly by the European Commission, to keep IZUM as the host of the new supercomputer and to arrange for the registration of the Slovenian AI factory.
Partners
The competences and synergy of the ten project partners in the SLAIF consortium were key in assessing the project application. The SLAIF consortium will consist of IZUM, IJS, Arnes, the Universities of Ljubljana, Maribor, Nova Gorica and Primorska, the Faculty of Information Studies from Novo Mesto, the Technology Park Ljubljana and the Chamber of Commerce and Industry of Slovenia.
Financial support
Of key importance in this regard is the extraordinary support of the Ministry of Higher Education, Science and Innovation, which finances IZUM and covers the operational costs of HPC Vega, as well as the provision of a substantial part of the financial contribution to the SLAIF project, provided by the Ministry of Digital Transformation.
THE NEW SUPERCOMPUTER
The new supercomputer will succeed the existing HPC Vega supercomputer, which will be used to carry out activities within the Slovenian AI Factory until its deployment. HPC Vega maintenance is scheduled to be extended until April 2026. The disk systems and servers for HPC Vega services will be maintained for several more years and will be suitable for storing large amounts of scientific data and archives.
Location in a new data centre
The new supercomputer will be located in the new Arnes data centre in the area of the hydroelectric power plant “Mariborski otok”. The project began in 2023, all documentation was obtained a year later, construction began in 2025 and is expected to be completed in the second half of 2026. Funds of around EUR 14 million are provided from the Recovery and Resilience Plan.
Timeline and financial framework
The tender for the procurement of a new supercomputer will begin in 2025, and the installation of equipment from the first phase, worth around EUR 44 million, will begin at the end of 2026. Equipment in the second phase of procurement worth around EUR 32 million will be installed in 2029.
Infrastructure management and development
The management, maintenance and administration of the new supercomputer will be taken over by IZUM in cooperation with IJS and Arnes within the SLING consortium, which will also carry out joint technical and expert development in the field of infrastructure, which will be optimized to support AI solutions and provide services to the AI factory.
Performance and storage capacities
The new supercomputer will have a capacity of around 100 petaFLOPS, which is around 16 times the capacity of the HPC Vega. For AI needs, the performance will be around 10 exaFLOPS in mixed precision. It will feature an extremely high-performance disk system for read-intensive applications and data storage with a capacity of 10 PB, which is 10 times more than on HPC Vega. It will also have a powerful storage capacity of 100 PB, which is 5 times more than on HPC Vega.
Supercomputer architecture
The new supercomputer will contain the following elements:
- The computing partition with classical processors will be dedicated to existing HPC calculations and data preparation for AI applications and data post-processing.
- A partition for intensive computing based on graphics accelerators (there will be around 2000 of them), which will be intended primarily for intensive loads of AI applications.
- A powerful disk system with sufficient capacity and low latency for the most demanding AI applications, such as large language models.
- The high-capacity storage, based on Ceph technology, provides a shared file system, block and object storage, and is equipped with data exchange services with national and international data centres.
- The services and virtualization partition is designed to manage, monitor, and connect all elements and network services of the supercomputer.
- User login nodes that are similar to the nodes of both partitions.
- Cloud services partition suitable for running (inference) AI applications and any other long-running user services.
Energy consumption and operating costs
The total peak power for all elements combined in both phases is estimated at 4.4 MW, and the expected average power is 3.1 MW at 65% peak capacity. The estimated cost of electricity over seven years is around EUR 32 million, and the total operating costs are estimated at EUR 38 million over seven years.
System software and applications
The system will preferably use open source software and solutions, as they can be easily adapted to the specific requirements of the system. To meet the demands of the AI and HPC communities, the system architecture must be flexible and agile. Cloud service resources will complement traditional HPC architecture with other platform-as-a-service capabilities to efficiently process data-intensive tasks and inference for AI applications. AI-optimized resources will be used to train and simulate AI models, with classical processors being supplemented with GPUs or other accelerators to meet computational demands. Additionally, many users will need a container orchestration system like Kubernetes to provide scalable and manageable application deployments.
Kubernetes installations and configurations include resources in the cloud or on HPC infrastructure, depending on the user’s needs, and are therefore versatile. The container-based technology Kubernetes will be useful for running user tasks on HPC resources and managing cloud services, and it is also possible to run containers on virtual instances in the cloud. Installing directly on physical servers is suitable for direct access to hardware resources, so there is practically no impact on performance. Additionally, Kubernetes can be used to deploy a containerized cloud control layer, which increases maintenance complexity but brings a lot of flexibility in deploying different settings.
AI application workloads typically require programmable access to resources through application programming interfaces (APIs), facilitating seamless integration and automation. This architecture will support the increasing complexity and prevalence of AI applications and provide flexibility and efficiency in resource utilization.
Slovenian Artificial Intelligence Factory – SLAIF
The reason for Slovenia’s own AI factory is primarily that Slovenia has a long tradition in the field of AI development and supercomputing. Slovenia already has a good research base in the field of AI, supported by institutions such as the IJS, the University of Ljubljana, and the University of Maribor. Advanced public research and networking infrastructure, including the Vega supercomputer and the ARNES network, and association in the SLING consortium provide a solid foundation on which SLAIF will build.
The core of the project is the plan and implementation of the involvement of companies and industry to ensure the competitiveness of Slovenian companies in the rapidly developing digital environment and to promote cooperation between industries, research institutions and the public sector. This is in complete alignment with the Slovenian strategy for start-ups.
The project is valued at EUR 11.37 million for three years, starting in the third quarter of 2025.
Ecosystem and project phases
Despite its development in this area, Slovenia needs dedicated supercomputing capacity optimized for AI that could meet the growing computational demands of artificial intelligence research, support industrial AI applications, and enable the development of general-purpose AI models. By aligning with the EuroHPC AI Factories initiative, SLAIF will ensure that Slovenian startups, SMEs and AI researchers can benefit from state-of-the-art supercomputing infrastructure while contributing to the broader European AI ecosystem.
At the heart of the ecosystem is a new supercomputer that will have connections to existing infrastructure and data spaces. Connections with other partners in other countries, with whom close cooperation is planned, will also be key.
Project phases
First phase
Second phase
Third phase
Final phase
First phase
Second phase
The second phase marks the transition from planning to active implementation. The project management team will fully implement service access policies and establish quality assurance systems. Regular reporting to stakeholders will be initiated and procedures will be established to monitor financial expenditures.
Third phase
The third phase focuses on improving and expanding services based on initial feedback and experience. The project management will review and optimize service access policies while improving quality assurance procedures based on user feedback. Stakeholder involvement will be increased and financial management systems will be optimized.
Final phase
The final phase focuses on optimizing all systems and ensuring long-term operation. The project management team will finalize operational procedures after the end of the project. Technical systems will reach maturity in this phase. The infrastructure team will optimize operations and complete the security framework, while also establishing the federation model. The SLAIF artificial intelligence platform will be fully optimized, complete documentation systems and maintenance procedures will be established. In user services, final optimizations of interfaces and support systems will be carried out, and the data storage infrastructure will also be completed.
Project approach
First work package
Second work package
Third work package
Fourth work package
Fifth work package
Sixth work package
Last work package
First work package
The first work package includes project management, which aims to ensure smooth coordination between consortium partners and stakeholders, align activities with the Slovenian AI strategy and European initiatives; establish inclusive policies for access to services, ensure financial transparency and ensure quality for the effective achievement of project milestones.
Second work package
The second work package includes infrastructure for orchestrating workflows and data. The goal of this cluster is to research, develop and deploy a reliable infrastructure that combines HPC and cloud resources and services for scalable implementation of AI workflows. It is about establishing a national infrastructure of data spaces and lakes that enable automated data exchange and provide access to storage infrastructure for data-driven workflows, with the aim of supporting innovation and cross-border collaboration.
Third work package
The third work package is dedicated to the core AI platform with the aim of ensuring the basic workflows of the AI platform. These are general horizontal services, suitable for adaptation to verticals, that provide and maintain a centralized hub with a semantic catalogue of AI components, including databases, models, workflows, and tools, facilitating the adoption of results and the exchange of knowledge between domains.
Fourth work package
The fourth work package is dedicated to user-oriented AI services. It focuses on services to make working with supercomputers and AI services more user-friendly, especially for those using them for the first time. So it’s about providing and maintaining a set of libraries, scripts, and tools that are more focused on first-time users and enable connectivity to data spaces and tools for creating synthetic data. Also important here is the development and maintenance of user interfaces and APIs for seamless connectivity to HPC and AI services, providing user-friendly access to workflows and infrastructure.
Fifth work package
The fifth work package is intended for vertical services for field applications (by sector), where the goal is to demonstrate the impact of AI on sustainability, health, digital society and science. Applications range from agriculture and energy, advances in diagnostics and personalized medicine, support for the processing of Slovenian and other less developed languages for the media and creative industries, to accelerating discoveries in scientific fields of great social importance.
Sixth work package
The sixth work package focuses on AI skills and training, with the aim of developing AI and HPC talent through tailored training programmes that meet the needs of different users, including students, professionals and SMEs. The goal is also to encourage business involvement by creating accessible learning paths and equipping companies with the necessary skills to deploy AI technologies.
Last work package
The last work package is aimed at informing and disseminating information by raising awareness and promoting the services of the Slovenian AI factory through effective communication strategies, strategic collaboration and public engagement. The goal is to highlight success stories, promote partnerships and the ethical deployment of AI, and at the same time enable knowledge exchange in Slovenia and Europe.
Key challenges in skills development
Program usposabljanja Slovenske AI tovarne bo v prihodnosti usklajen s slovenskimi nacionalnimi strateškimi potrebami na področju umetne inteligence (UI) in bo naslavljal ključne izzive. Med ključnimi izzivi so vrzel v digitalnih spretnostih, nezadostna zastopanost na področju IKT, razlike med spoloma in nezadostno sodelovanje z gospodarstvom. Za naslavljanje teh izzivov bodo uvedeni celostni izobraževalni programi, pobude za izpopolnjevanje in prekvalifikacijo na področju UI ter spodbujanje poklicev na področju STEM in IKT. Slovenski program usposabljanja tovarne UI bo vključeval razvoj cevovodov nadarjenosti za UI, pobude za vseživljenjsko učenje na področju UI ter uporabo regionalnega pristopa k usposabljanju na področju UI.
Program usposabljanja Slovenske AI tovarne bo v prihodnosti ponujal celovit nabor pobud za razvoj veščin na področju umetne inteligence (UI). Vključeval bo razvoj nadarjenosti za UI z razširitvijo univerzitetnih učnih načrtov, integracijo izobraževanja o UI na različnih ravneh, praktičnim usposabljanjem in certifikacijo delovne sile. Pobuda za opolnomočenje MSP bo vključevala demistifikacijo UI za MSP, usposabljanje osredotočeno na MSP, individualno uvajanje in spremljanje potreb. Program bo izkoristil regionalno podporno mrežo z decentraliziranim središčem za usposabljanje o UI, sodelovanjem s tehnološkimi parki in inkubatorji ter čezmejnim sodelovanjem. Večstopenjski okvir za usposabljanje bo ponujal osnovno, napredno in sektorsko specifično usposabljanje ter vseživljenjsko učenje in prekvalifikacijo. Specifični programi usposabljanja bodo vključevali usposabljanje administratorjev, delavnice za MSP, programe za učitelje, sodelovanje javne uprave ter sodelovanje raziskovalcev, akademikov in študentov.
Program usposabljanja Slovenske AI tovarne bo v prihodnosti ponujal celovit nabor pobud za različne tipe uporabnikov na začetni, srednji in napredni ravni. Vključeval bo razvoj nadarjenosti za UI z razširitvijo univerzitetnih učnih načrtov, praktičnim usposabljanjem in certifikacijo delovne sile, opolnomočenje MSP z usposabljanjem osredotočenim na MSP, individualnim uvajanjem in spremljanjem potreb, ter regionalno podporo z decentraliziranim usposabljanjem, sodelovanjem s tehnološkimi parki in čezmejnim sodelovanjem. Večstopenjski okvir za usposabljanje bo ponujal osnovno, napredno in sektorsko specifično usposabljanje ter vseživljenjsko učenje in prekvalifikacijo, specifični programi pa bodo vključevali usposabljanje administratorjev, delavnice za MSP, vrstniško učenje ter sodelovanje javne uprave, raziskovalcev, akademikov in študentov.
Companies cooperation and support
The AI Factory envisages close cooperation with the National Competence Center EuroCC SLING and the EuroHPC centres of excellence, as well as cross-border cooperation. At the national level, Slovenia will connect with Austria (Advanced Computing Austria ACA GMBH), Croatia (Ministry of Justice, Public Administration and Digital Transformation), Italy (CINECA), Serbia (Office for Information Technologies and e-Government), Bosnia and Herzegovina (Verlab Research Institute for Biomedical Engineering, Medical Devices and Artificial Intelligence on behalf of NCC BIH).
SLAIF has secured strong support from industry leaders in key sectors, indicating widespread demand for AI transformation in Slovenia. We have thus gained support from the companies Telekom Slovenije, Mikrografija, Špica, Novartis, Siemens, DOMEL, IGEA, Medius, SRC, SRC Infonet, ELES, Operato, TPV Automotive, Alcad, Cosylab, Genialis, Sensum, Better. SLAIF’s involvement in European HPC and AI networks will be strengthened through partnerships with European data spaces and initiatives such as ALT-EDIC, ELIXIR SI, CLARIN ERIC, the Slovenian Open Science Community, DIH Agrifood and SRIP PMIS.