In the global AI race, Estonia cannot afford to give up. Our goal is not to win, but to be competent in all important areas.
As workforce augmentation becomes increasingly common, several major service sectors are no longer able to process data using traditional commercial large-scale language models (LLMs). These include sensitive domains such as healthcare, defense, and national data. Therefore, developing an accurate and effective large-scale language model of Estonian would be of great benefit to Estonian society. However, due to the country’s small size, there is little commercial incentive for other countries to undertake such developments. In order to preserve and protect the Estonian language amid rapid advances in artificial intelligence (AI) and create practical applications for everyday use, the Computer Science Institute at the University of Tartu is training open-source language models to help people communicate more fluently in Estonian and understand Estonian culture more deeply. The model is trained using a computer cluster at the University of Tartu HPC Center (UTHPC).
There is a wide range of innovative, research-based companies with strict requirements for intellectual property protection and data security, and who require absolute clarity about their legal and security environments.
Our solutions are primarily based on investments in knowledge
Today, it is clear that after regaining independence 35 years ago, Estonia’s decision to build an economy based on new technology and education has paid huge dividends.
UTHPC benefits from being part of the University of Tartu, which ranks in the top 1.2% of universities in the world and offers a broad base of expertise, from basic fields to highly specialized fields, including artificial intelligence. This environment provides a unique opportunity to collaborate with leading researchers and allows us to offer University of Tartu students a highly practical training course taught by experts directly involved in cutting-edge HPC and AI operations.
Based on our experience, AI agents are very effective, but only if the user has sufficient expertise to examine and correct the output. Therefore, we believe that as technology advances, the demand for highly skilled IT professionals will increase, not decrease.
Second, our approach focuses on building our own systems.
The University of Tartu High Performance Computing Center (UTHPC) was officially established in 2008. However, the history of scientific computing in Tartu dates back to 1959, when the university’s first scientific computing center was established. In 2008, the Center deployed its first computing cluster, named Aurumasin (“Steam Engine”). The cluster consisted of 42 Sun Fire nodes, each with 8 cores, 32 GB of RAM, and 10 TB of shared storage.
Since then, UTHPC has evolved from a provider of basic computing cycles to a comprehensive service that supports users throughout the research lifecycle, from efficient execution of analyzes to secure data storage and full compliance with relevant regulatory requirements.
In February 2026, local computing power increased by about an order of magnitude. We commissioned new GPU resources, including 12 NVIDIA H200, 24 NVIDIA B200, and 40 NVIDIA RTX Pro 6000 Blackwell Max-Q GPUs. The hardware was procured to accelerate the development of national and scientific AI solutions in collaboration with the Ministry of Justice and the Department of Digital Affairs. As a result, our combined cloud, Kubernetes, and cluster infrastructure now consists of 164 GPUs, 21,520 CPU threads, and 118 TB of system memory, plus over 17 petabytes of usable storage and over 30 petabytes of storage dedicated to archives.
Equally important to this advanced hardware infrastructure is our team of over 50 highly trained professionals. The team actively participates in cyber defense exercises and hackathons, contributing to both resilience against cyber threats and the exploration of future AI-driven solutions.
To illustrate our growth, below are some key milestones.
Shaping the scientific computing landscape in Estonia
As computational methods become an integral part of research across most scientific disciplines, the need for national consortia becomes increasingly apparent. In response, the Estonian Scientific Computing Infrastructure (ETAIS) was established in 2011. The consortium is led by the University of Tartu, in collaboration with partners such as Tallinn University of Technology, the Institute of Chemical and Biophysical Sciences, and the Ministry of Education and Research.
ETAIS supports the competitiveness of Estonia’s science and industry by providing access to cutting-edge computing resources. These services include high-performance computing power, secure data management, user support, and expert consulting. Key user communities include bioinformatics, medicine, data mining, language technology, chemistry, materials science, climate research, and physics.
Currently, ETAIS serves as the official infrastructure provider for Estonia’s research and development sector, and also provides modern AI infrastructure services across the Estonian public sector. Additionally, ETAIS will connect Estonia’s domestic electronic infrastructure with international partners, in particular the EuroHPC LUMI supercomputer and the LUMI AI Factory.
ETAIS represents Estonia within the Nordic Electronic Infrastructure Collaboration (NeIC) and develops the Puhuri service, a specialized resource management and portal system used in several countries, including the LUMI supercomputing environment. Additionally, ETAIS participates in projects focused on the development of quantum computing technologies and secure long-term archiving of personal data.
ETAIS is also a founding member of the LUMI supercomputer consortium. Officially launched on June 13, 2022 in Kajaani, Finland, LUMI provides researchers and entrepreneurs with access to world-class computational resources. This gives even the most demanding users, such as those developing advanced AI models, access to the capabilities needed to make major advances in science and technology. As of early 2026, LUMI continues to rank among the world’s top 10 supercomputers.
Our team is actively contributing to several work packages aimed at building and operating services on and around this great computing platform.
Promoting and protecting Estonia’s genomic data
Estonian Biobank has established a population-based biobank covering more than 200,000 individuals. Considering that approximately 20% of Estonia’s adult population is enrolled in this program, this resource represents a uniquely valuable dataset for the advancement of medical research both domestically and internationally. This data will allow researchers to investigate differences in disease risk and variation in individual responses to healthcare, thereby forming the basis for the implementation of personalized medicine in Estonia.
Due to the sensitive nature of this data, access is only provided through the Secure Sensitive Data Private Research Environment (SAPU). SAPU is the most secure service operated by UTHPC, providing a fully isolated research environment with comprehensive and exception-free logging to ensure traceability and auditability when required.
Obtained ISO/IEC 27001:2022 certification
In 2025, UTHPC obtained ISO/IEC 27001:2022 certification. This internationally recognized standard confirms the Center’s commitment to maintaining the highest level of data protection and security protocols. This certification covers all critical IT operations and services provided by UTHPC and ensures compliance with globally recognized information security standards.
The certification process includes a comprehensive audit of UTHPC’s information security management system, including all aspects of data processing, risk assessment, and implementation of security controls across a wide range of services. Achieving this certification demonstrates UTHPC’s dedication to protecting sensitive research data and maintaining operational excellence across all service domains.
The Center maintains certification through regular audits and continued enhancement of security practices to ensure that information security management systems remain robust against evolving threats in both academic and commercial settings.
Managing resources
Since 2016, part of our team has been involved in developing the Waldur portal, which makes accessing complex computing resources as easy as online banking. Portals allow organizations to give users the tools they need, maintain control over budgets, and make efficient use of expensive technology.
Waldur is a key component of the EuroHPC Federation Platform (EFP), a central portal designed to integrate Europe’s diverse supercomputing and quantum resources into a single, secure and user-friendly ecosystem. Within the framework of CASTIEL 2 and the upcoming CASTIEL 3 project, we are responsible for developing a marketplace for the services offered by the EuroHPC Competence Center.
Everything is made possible through close collaboration
Our Department deserves recognition for its bold and innovative efforts to inspire, enable and support collaboration on projects too ambitious for any partner to tackle alone.
Thirty years ago, the Tiger Leap program was launched, providing computers to all schools and laying the foundations for Estonia’s digital success story. Today, in response to the new challenges of uncontrolled use of AI in education, a new program, AI Leap, is introduced. The initiative brings together thousands of teachers to reimagine education in the age of AI, increase AI literacy across society, and advance educational practices through research-based development.
We believe that these efforts, combined with the infrastructure, expertise and collaboration described above, will provide us with the connections and knowledge necessary to provide comprehensive services. These services allow us to train experts and provide them with the tools to use the latest and future AI technologies independently, safely and effectively.
Please note: This is a commercial profile
This article will also be published in the quarterly magazine issue 25.
Source link
