HUN-REN Magyar Kutatási Hálózat is orienting Hungary’s AI strategy toward knowledge-intensive, algorithmic and energy-efficient solutions rather than trying to compete in training giant models that require massive capital and energy. Jakab Roland, CEO of HUN-REN, laid out this approach ahead of the international AI symposium on May 21–22, which is focused on understanding AI, improving verifiability, and turning AI into a research methodology and economic capability.
Why this strategy?
Jakab Roland argues that building very large language models demands extraordinary amounts of money, energy, data and compute, so a smaller country should manage resources consciously and pick domains where its expertise can be internationally valuable. Hungary can build on strong mathematical traditions and institutions such as HUN-REN’s Rényi Alfréd Mathematical Research Institute and SZTAKI.
The focus is on researching new algorithms, methods and AI architectures that can achieve similar or better outcomes with less energy, less data and lower compute. According to the CEO, the algorithmic advances, mathematical thinking and novel techniques are areas where the domestic research community can make meaningful contributions to global progress.
The symposium and scientific focus
This year HUN-REN holds its international AI symposium for the second time; researchers from Google, Microsoft, ETH Zürich and Nanyang Technological University are among the participants. The event concentrates on four main areas: trustworthy AI, healthcare applications, industrial use and quantum technology.
Organisers aim to go beyond tool- and business-oriented AI conferences and investigate how AI can be harnessed as a scientific methodology: why models work, how to make them more reliable, how to reduce vulnerabilities, and how to place AI at the service of science.
AI-first thinking in research
The "AI-first" approach means redesigning the entire research value chain — from hypothesis generation and literature review to experiment design and interpretation — so it is built on AI tools. It is not about occasional use of an AI application, but about embedding AI into the research methodology from the start.
This amplifies the importance of reliability and verifiability: when AI suggests hypotheses, experimental designs or conclusions, researchers must pre-plan validation steps, tests and checkpoints to determine whether results are robust enough to form the basis for further development.
Institutional tools: AI ambassadors, service centre and scouting team
Under the AI4Science programme HUN-REN first identified researchers already using AI tools in its institutes and formed an AI-ambassador network. These ambassadors help colleagues adopt tools, share best practices and integrate AI into research work.
Alongside the network, HUN-REN created an AI service centre where AI architects and specialists support researchers on a project basis, and a scouting team that continuously monitors new models, methods and publications to identify what is relevant to the domestic community and how to deliver it through validated training and services.
A survey among researchers collected nearly a thousand responses and informed which services, trainings and central expertise are needed. The survey also highlighted that many researchers use personal AI subscriptions, raising data security and cost-sharing concerns.
Agentic Discovery Platform: central access and research-specialist agents
To address these issues, HUN-REN will launch the Agentic Discovery Platform within one to two weeks. The platform has three main goals:
- provide centrally funded access for researchers to the leading AI models they are used to, inside a more secure environment;
- offer an AI system that runs entirely on the internal network for working with sensitive research data or unpublished ideas;
- make available AI agents specialised for research that can access research databases and literature, write and execute code, and run solutions.
The platform will expose more than 2,000 specialist tools usable by AI, all through a unified interface so researchers can easily try and apply the most relevant AI applications.
Connecting to industry and supercomputing
HUN-REN also aims to make AI and supercomputing capabilities accessible to domestic innovative small and medium enterprises. A new supercomputing infrastructure is being built at the HUN-REN Wigner Research Centre for Physics to support compute-intensive modelling and simulations relevant to drug discovery, molecular research, materials science and industrial development.
Through the AI Factory Antenna programme — joined by Hungary under HUN-REN SZTAKI’s coordination as HunAIFA — the plan is to link science, companies, the public sector and the supercomputing infrastructure. Over the coming months and years a 15–20 person expert team will be established to bridge innovative, compute-heavy AI projects and the European supercomputing infrastructure.
That team will help businesses assess whether a business or technological problem is worth modelling or simulating in a supercomputing environment.
Summary
HUN-REN’s emphasis is on research-oriented, efficient and verifiable AI solutions: strengthening international collaborations, offering central platforms for researchers and building supercomputing-enabled links to industry are all part of the strategy. The imminent Agentic Discovery Platform and the HunAIFA-related capabilities aim to make modern AI tools and large-scale compute accessible to Hungarian researchers and companies.
