National Reference Architectures for Science, Data Management & Scientific Computing

A comparative survey of governmental and national-level reference architectures across 9 countries and the European Union — scope, timescales, strategic themes, similarities, and key differences, with thematic analysis of cybersecurity, AI, quantum computing, and international cooperation.

Compiled: March 2026 14 Architectures Surveyed 9 Countries + EU Timespan: 2015–2030
§ 1 — Introduction

Why National Reference Architectures Matter

As research increasingly depends on large-scale data, high-performance computing, and AI, national governments have turned to reference architectures — formal strategic blueprints defining how scientific computing, data management, and digital infrastructure should be organised, governed, and evolved at a national level.

These documents differ in character: some are technical blueprints (Finland's TiLa), others are investment roadmaps (Netherlands' LSRI), national infrastructure programmes (Germany's NFDI), or supranational federation frameworks (EOSC). Together they reveal a global convergence toward FAIR data, open science, federated infrastructure, and AI-readiness — while also reflecting national priorities, governance cultures, and timescales.

This report surveys 14 architectures across 9 countries and the EU. Finland's two architectures — Tieteellisen laskennan viitearkkitehtuuri 2025 (TiLa) and Tutkimuksen datanhallinnan viitearkkitehtuuri 2030 (DAHA) — serve as the reference baseline. The final section analyses four emerging themes — cybersecurity and research security, artificial intelligence, quantum computing, and international cooperation — with dedicated focus on what TiLa and DAHA say about each topic and what is currently missing from them.

Scope note

This survey covers national or governmental-level documents relating to scientific computing, research data management, and adjacent AI/e-infrastructure strategies. Sectoral plans are noted where they form part of a broader national framework.

1 of 7
§ 2 — Inventory of Architectures

The Reference Architectures

🇫🇮

Finland

Tieteellisen laskennan viitearkkitehtuuri 2025 (TiLa)
Finland
2021–2025Scientific ComputingHPCData Management

Published January 2021, five-year horizon. Developed by the Ministry of Education and Culture's Scientific Computing Cooperation Forum (YTF), supported by CSC. Defines the reference architecture for scientific computing resources, data management, and ancillary services for Finnish universities and research institutions. Emphasises interoperability, service development, and alignment with European e-infrastructure. The 2025–2027 YTF is now evaluating whether to produce a successor architecture, with AI, geopolitical security, and quantum identified as top priorities.

Tutkimuksen datanhallinnan viitearkkitehtuuri 2030 (DAHA)
Finland
2023–2030Research Data ManagementOpen ScienceFAIR

Version 1.0 published 2024, horizon year 2030. Developed under OKM mandate by the YTF 2022–2024 term, with CSC support. Describes the target state of research data management in Finland by 2030, covering the full data lifecycle, sensitive data, long-term preservation, and concrete institutional actions. Builds on TiLa and the Open Science reference architecture (AVOTT). Recipient of the Finn-ARMA award for best research conditions development work.

🇪🇺

European Union

European Open Science Cloud (EOSC) — Architecture & Interoperability Framework
EU
2018–ongoingFederated Science CloudFAIROpen Science

The EU's flagship open science infrastructure, providing European researchers a trusted multi-disciplinary environment for publishing, finding, and reusing data. The EOSC EU Node launched October 2024 as the first Federation node. Governed tripartitely by the European Commission, EOSC Association (200+ members), and Member State representatives. Horizon Europe funds services through 2027; post-2027 sustainability is under active negotiation.

EuroHPC Joint Undertaking — European HPC, AI & Quantum Infrastructure
EU/Nordic
2018–ongoingSupercomputing / AI / QuantumAI FactoriesQuantum

Coordinates pan-European HPC investment including LUMI in Finland (9th globally, hosted by CSC), JUPITER in Germany (Europe's first exascale, 4th globally), and LEONARDO in Italy. The 2024 mandate extension launched 19 AI Factories including Finland's LUMI AI Factory, plus a network of 13 AI Factory Antennas. EuroHPC has also procured six quantum computers integrated with HPC systems across Europe, with Finland participating in the LUMI-Q consortium and hosting the LUMI-IQ experimental quantum platform.

🇩🇪

Germany

Nationale Forschungsdateninfrastruktur (NFDI)
Germany
2018–2028+Research Data Infrastructure27 ConsortiaFAIR

Established by Federal Government–Länder agreement November 2018. A decentralised, consortium-based national infrastructure for managing scientific data across all disciplines. 27 consortia funded across three rounds (2020–2022). An Overall Architecture Working Group was established April 2025 to develop a shared federated blueprint. The Wissenschaftsrat published a structural evaluation July 2025 recommending continuation. Germany's funding runs to 2028.

🇬🇧

United Kingdom

UKRI Digital Research Infrastructure (DRI) Programme
UK
2020–ongoingDigital Research InfrastructureHPC/AISkills

Launched 2020 by UK Research and Innovation. Creates a coherent national digital research infrastructure spanning data, compute, software, networks, and skills. Includes BioFAIR, DARE UK (Trusted Research Environments), and the AI Research Resource (AIRR). In December 2025 UKRI set out a £38.6 billion four-year allocation, complementing DSIT's £1 billion UK Compute Roadmap targeting a 20-fold expansion of compute capacity over five years.

UK National Data Library (NDL)
UK
2024–2030National Data AccessHealth DataFederated TRE

A developing national data infrastructure to make UK public data assets accessible for research and innovation. The Health Data Research Service (HDRS), backed by up to £600 million from Government and Wellcome Trust, will be fully operational by December 2030. Architecture is federated and distributed, with strong emphasis on safe-access Trusted Research Environments.

🇳🇱

Netherlands

National Roadmap for Large-Scale Research Infrastructure (LSRI)
Netherlands
5-yr cycles (2016→)Research InfrastructureDigital & Physical

Published every five years by NWO, covering all forms of large-scale research infrastructure including supercomputers, data collections, and biobanks. The 2021 edition established nine thematic groups; the 2024 round awarded €197 million to eleven projects. A third roadmap is planned for 2026. The Netherlands participates in the LUMI-Q quantum consortium alongside Finland.

🇸🇪

Sweden

NAISS — National Academic Infrastructure for Supercomputing in Sweden
Sweden
2023–ongoingHPC / SupercomputingAI ComputingQuantum

NAISS (from 2023) replaced SNIC as Sweden's national HPC infrastructure. In 2024 NAISS published its strategic report covering AI-dedicated hardware (MIMER initiative) and the Arrhenius supercomputer (Linköping, Q1 2026). Sweden is one of the first seven AI Factory countries (December 2024, MIMER AI Factory) and participates in the LUMI-Q quantum computing consortium. Sweden's 2024 research bill designates quantum as a strategic research area.

🇦🇺

Australia

National Research Infrastructure Roadmap 2021 & Draft National Digital Research Infrastructure Strategy
Australia
2021–2028 (10–15 yr)Digital Research InfrastructureNCRISOpen Science

Australia's 2021 NRI Roadmap outlines national collaborative research infrastructure to 2028 under the NCRIS programme. A Draft National Digital Research Infrastructure Strategy (2023) extends this to a 10–15 year vision. The ARDC leads major programmes including the Nectar Research Cloud, HASS Research Data Commons, and the 2024 Australian National PID Strategy. Australia published a National Quantum Strategy in May 2023 targeting global leadership by 2030.

🇺🇸

United States

NASA SMD Data and Computing Architecture Study
USA
FY22–2024Space Science Data & ComputingHybrid CloudOpen Science

NASA's Science Mission Directorate Data and Computing Architecture Study (final report August 2024) addressed computing needs of 150+ active missions generating over 100 PB of open data, projected to exceed 500 PB by 2030. Recommends a hybrid on-premises/cloud architecture in support of SPD-41a open science mandates.

Federal Big Data R&D Strategic Plan (NITRD Update 2024)
USA
2024Federal R&D Data EcosystemAI/MLCross-agency

NITRD published an updated Federal Big Data R&D Strategic Plan in 2024, covering AI, generative AI, data quality, and sustainable computing. Coordinates federal investment across DOE, NSF, NIST, NASA, and NIH. The DOE operates Frontier (world's first exascale, USA) and Aurora at national labs. A parallel NITRD Cybersecurity R&D Strategic Plan was published in 2023.

🇨🇦

Canada

Data Strategy for the Federal Public Service 2023–2026 + Tri-Agency RDM Policy
Canada
2023–2026Federal Data GovernanceRDM PolicyIndigenous Data

The Treasury Board of Canada Secretariat leads the whole-of-government Data Strategy 2023–2026, covering data standards, CDO accountabilities, data literacy, and indigenous data sovereignty. Canada's three federal research agencies (SSHRC, NSERC, CIHR) require mandatory Data Management Plans from 2023. Canada's National Quantum Strategy (2023) and National Security Guidelines for Research Partnerships (2021) address emerging technology risks.

🇯🇵

Japan

Comprehensive Data Strategy & NII Research Data Cloud (NII RDC)
Japan
2021–ongoingNational Data InfrastructureOpen ScienceAI

Japan's Cabinet-level Comprehensive Data Strategy (June 2021, updated annually) directed a 21st-century digital infrastructure. The NII operates the NII Research Data Cloud: GakuNin RDM, WEKO3, and CiNii Research. Japan's 6th STI Basic Plan mandated open science; 70 university data policies were published by June 2024. Japan's Q-LEAP quantum initiative (2018) and RIKEN quantum computing programme have built significant national capacity.

2 of 7
§ 3 — Comparative Analysis

Side-by-Side Comparison

The table below compares key dimensions of each architecture: issuing body, primary scope, horizon year, governance model, FAIR alignment, and international framework linkages.

CountryDocument / ProgrammeHorizonPrimary ScopeGovernanceFAIR / OpenInt'l Alignment
🇫🇮 Finland (TiLa)Scientific Computing Ref. Arch. 20252021–2025HPC, data mgmt, servicesMinistry + YTF + CSCPartial FAIREOSC, EuroHPC/LUMI
🇫🇮 Finland (DAHA)Research Data Mgmt Ref. Arch. 20302023–2030Full data lifecycle, RDMMinistry + YTF + CSCFAIR-centredEOSC, AVOTT, TiLa
🇪🇺 EU (EOSC)Architecture & Interoperability Framework2018–2027+Pan-European science cloudEC + EOSC Assoc. + Member StatesFAIR-nativeRDA, OpenAIRE, national nodes
🇪🇺 EU (EuroHPC)EuroHPC Joint Undertaking2018–ongoingPan-European HPC / AI / QuantumEU JU + member statesOpen accessEOSC, Horizon Europe
🇩🇪 GermanyNFDI — National Research Data Infrastructure2018–2028+Research data, all disciplinesDFG + 27 consortiaFAIR-nativeEOSC (mandatory), RDA
🇬🇧 UK (UKRI DRI)Digital Research Infrastructure Programme2020–2030Compute, data, skills, softwareUKRI + DSITFAIR-orientedEOSC-adjacent, EuroHPC
🇬🇧 UK (NDL)National Data Library2024–2030Public data access, health dataGovernment + Wellcome + UKRIPrivacy-first + openHDR UK, DARE UK, TRE network
🇳🇱 NetherlandsNational Roadmap for LSRI5-yr cyclesAll research infrastructureNWO + OCW ministryFAIR in digital pillarESFRI, EuroHPC, EOSC
🇸🇪 SwedenNAISS2023–ongoingHPC, storage, AI, quantumSwedish Research Council + universitiesOpen accessEuroHPC, EOSC, LUMI-Q
🇦🇺 AustraliaNRI Roadmap + Draft NDRI Strategy2021–2028 (10–15 yr)Research infrastructure, digitalDept of Education + NCRIS + ARDCFAIR + open dataRDA, international PIDs
🇺🇸 USA (NASA)SMD Data & Computing Arch. StudyFY22–2030Science mission data + computingNASA SMDOpen science mandateNITRD, DOE, NSF
🇺🇸 USA (NITRD)Federal Big Data R&D Strategic Plan2024Federal data + AI ecosystemNITRD cross-agency (OSTP)AI/ML data qualityG7/G20 data principles
🇨🇦 CanadaFederal Data Strategy + Tri-Agency RDM2023–2026Gov data governance + research RDMTBS whole-of-government + 3 agenciesOpen by defaultG7/G20, OECD principles
🇯🇵 JapanComprehensive Data Strategy + NII RDC2021–ongoingNational data infra + open scienceCabinet Office + NII + Digital AgencyFAIR-compliant (NII)G7/G20, data spaces
3 of 7
§ 4 — Cross-Cutting Themes

Convergent Themes Across All Architectures

Despite significant differences in governance, scale, and national context, a striking degree of convergence is visible across all surveyed architectures.

🔄
FAIR Data Principles
Every architecture references FAIR (Findable, Accessible, Interoperable, Reusable). Some treat it as an aspiration (TiLa), others as a native design principle (EOSC, NFDI, DAHA). FAIR has become the de facto global interoperability standard.
🌐
Federated Architecture
No country has chosen a purely centralised model. Data sovereignty, GDPR, and practical data volumes all drive federation. EOSC's Federation model is becoming the European template.
🤖
AI Integration
AI and ML have shifted from research workload to first-class infrastructure. EuroHPC AI Factories, Sweden's MIMER, UK AIRR, and US DOE exascale+AI all reflect this transition.
🔒
Sensitive Data & Security
Handling sensitive and personal data is a universal challenge. Solutions include TREs (UK DARE UK), secure enclaves (NAISS), and sensitive data workflows (DAHA, CSC SD Services). Geopolitical risks are escalating.
♻️
Sustainability
Long-term financial sustainability is a recurring tension. NFDI faces post-2028 uncertainty; EOSC post-2027 governance is unresolved. Research infrastructure is politically hard to fund on a recurring basis.
🎓
Skills & Capacity
Workforce development appears in every architecture. Data stewards, RSEs, HPC specialists, and AI literacy are uniformly identified as the limiting factor regardless of infrastructure investment.
🛤️
Open Science Mandate
All architectures endorse open science. The gap is in enforcement: mandates (NASA, Canadian Tri-Agency), recommendations (TiLa, DAHA), or incentive-based norms (ARDC). EU policy pressure is raising the bar across Europe.
🔗
Cross-Domain Interoperability
Breaking down disciplinary and institutional silos is universal. EOSC, NFDI cross-consortia sections, Australia's PID Strategy, and Japan's CiNii/WEKO3 platforms all target this goal.
Governance variation

Finland and Sweden use ministry-mandated advisory forums (YTF, SRC) with operational agencies (CSC, NAISS). Germany chose a bottom-up science-driven consortium model (NFDI). UK operates through UKRI as a large mission-oriented funder. USA is fragmented by agency mission with NITRD cross-agency coordination. Japan is top-down from Cabinet level with NII as the operational node.

Finland's distinctive approach

Finland's TiLa/DAHA pair is notable for its explicit architectural layering — from principles through capability mapping to services — and for the tight coupling between CSC, OKM, and YTF. DAHA uniquely treats Data Management Plans, long-term preservation, and sensitive data workflows as co-equal concerns alongside compute and storage. The 2025 YTF forum flagged AI, geopolitical data security, and quantum as the top forward-looking priorities.

4 of 7
§ 5 — Chronological View

Timeline of Major Milestones

2015–2016
First-generation national roadmaps
Netherlands publishes its first LSRI Roadmap. Swedish Research Council conducts first international review of SNIC. Finland publishes its first scientific computing architecture. EU Quantum Technologies Flagship preparations begin.
2018
NFDI agreement, EOSC foundation, quantum flagships
Germany's Federal Government–Länder NFDI agreement signed November 2018. EOSC governance established. EuroHPC Joint Undertaking created. EU Quantum Flagship launched (€1B, 10 years). US National Quantum Initiative Act enacted.
2020–2021
Pandemic-accelerated wave of new architectures
Finland's TiLa 2025 published (January 2021). NFDI first consortia funded. UKRI DRI Programme launched. Japan's Comprehensive Data Strategy and 6th STI Basic Plan. Australia's NRI Roadmap 2021. Netherlands LSRI Roadmap 2021.
2022–2023
Consolidation, quantum strategies, RDM mandates
Canada Tri-Agency RDM Policy mandatory from March 2023. Sweden transitions SNIC → NAISS. Australia Draft NDRI Strategy published. UK National Quantum Strategy (£2.5B / 10 years). NASA SMD Architecture Study chartered. NIS2 Directive adopted by EU.
2024
AI Factories launched, quantum systems deployed
EOSC EU Node launched (October 2024). EuroHPC selects first 7 AI Factories including Finland's LUMI AI Factory (December 2024). NASA SMD report released (August 2024). NITRD Big Data Plan updated. Finland DAHA 2030 v1.0 published. NIS2 transposition deadline (October 2024). NIST post-quantum cryptography standards published (August 2024).
2025
Evaluation cycles, quantum deployment, AI Factory expansion
Finland YTF 2025–2027 begins; evaluates TiLa successor. Germany NFDI Wissenschaftsrat evaluation (July 2025). NFDI Overall Architecture Working Group established (April 2025). EuroHPC deploys VLQ quantum computer in Czechia (September 2025). LUMI AI Factory opens first services (October 2025). EuroHPC selects 13 AI Factory Antennas.
2026–2030
Target horizons
Sweden Arrhenius supercomputer operational (Q1 2026). Netherlands 3rd LSRI Roadmap (2026). LUMI AI Factory LUMI-AI system procurement underway. UK HDRS fully operational 2030. Finland DAHA target year 2030. EOSC post-2027 sustainability model to be determined.
5 of 7
§ 6 — Summary Findings

Main Findings & Observations

1

A Global Wave, Accelerated Post-2020

Most architectures were published or revised 2020–2024. COVID-19 was a forcing function, exposing gaps in interoperability and data sharing. TiLa, NFDI, Japan's Comprehensive Data Strategy, Australia's NRI Roadmap, and UKRI DRI all reflect this wave.

2

FAIR Data as the Global Standard

FAIR principles have achieved near-universal adoption. Germany and the EU have built FAIR into foundational governance and technical design; others are working toward systematic implementation. Finland's DAHA 2030 sits in a mature implementation position, treating FAIR as a design requirement rather than an aspiration.

3

Federated Over Centralised

No country has chosen a purely centralised architecture. The dominant model is federated central coordination with distributed data sovereignty, driven by GDPR, cultural norms, and data volumes. Smaller countries with single national nodes (CSC, NAISS) navigate this considerably more easily than larger, fragmented systems.

4

AI Has Moved from Aspiration to Infrastructure

In 2021-era documents, AI appeared as a research workload to be supported. In 2024–2025 documents, AI is first-class infrastructure. EuroHPC's 19 AI Factories across Europe — including Finland's LUMI AI Factory in Kajaani — represent the most architecturally integrated public AI response globally.

5

Skills Remain the Critical Bottleneck

Every architecture identifies data stewards, research software engineers, HPC specialists, and AI literacy as the limiting factor — regardless of infrastructure scale. Training programmes and career pathways are universally emphasised.

6

Sustainability and Post-Horizon Governance Are Unresolved

Several architectures approach their nominal horizon years without fully resolved successors: TiLa 2025, NFDI 2028, EOSC 2027. Research infrastructure is politically difficult to fund on a long-term recurring basis, yet scientific research requires decade-scale stability.

7

Timescale Variation Reflects Governance Culture

Horizons range from 3-year operational plans (Canada) to 5-year roadmaps (Netherlands, TiLa) to decade-long visions (DAHA 2030, UK NDL, Australia NDRI). The most effective architectures combine a short-term operational plan with a longer-term aspirational vision — precisely the model Finland's TiLa/DAHA pairing demonstrates.

8

Open Science Is Mainstream — Implementation Varies

All architectures endorse open science in principle. The gap is in enforcement mechanisms: mandates (NASA, Canadian Tri-Agency), EU policy pressure, or recommendation-based models (Finland, Sweden). Countries with centralised research funding bodies can more easily attach open science conditions to grants.

6 of 7
§ 7 — Emerging Thematic Areas

Four Emerging Topics: Global Landscape & Finland's Position

Four topics have emerged as cross-cutting priorities that challenge or extend the scope of most national architectures. For each topic this section covers the international landscape, then provides a dedicated analysis of what TiLa and DAHA currently say and what is missing from those documents.

🔐
A — Cybersecurity & Research Security
Protection of research infrastructure, data, and intellectual assets against state-sponsored and criminal threats

The threat landscape for research infrastructure has intensified markedly since 2020. State-sponsored actors have increasingly targeted universities, research databases, and HPC centres for espionage, IP theft, and disruption. Most earlier national architectures treated security as a compliance matter; 2024–2025 documents increasingly treat it as a strategic design principle. NIS2 (transposition deadline October 2024) expanded mandatory cybersecurity controls to digital infrastructure operators across the EU, and the January 2026 EU Cybersecurity Package proposes horizontal ICT supply chain security provisions that will directly affect EuroHPC operators including CSC.

🇪🇺 EU (EOSC / EuroHPC)
AdvancedEOSC explicitly incorporates trusted data environments and NIS2/GDPR compliance. The EOSC Federation requires security certification for Candidate Nodes. The January 2026 EU Cybersecurity Package proposes a supply chain security framework covering NIS2-regulated entities, including research digital infrastructure operators.
🇬🇧 UK (UKRI DRI)
LeadingUKRI-EPSRC funds four National Cyber Security Centre research institutes. DARE UK built Trusted Research Environments as a core architectural component with security by design. The UK Cybersecurity and Resilience Bill — among the most comprehensive globally — was under active parliamentary consideration in early 2026.
🇩🇪 Germany (NFDI)
PartialNFDI handles security at the consortia level with variable maturity. The emerging Overall Architecture Working Group is expected to address federated security governance. Germany's BSI provides national guidance separately from NFDI documents.
🇸🇪 Sweden (NAISS)
PartialNAISS operates within Sweden's NCSC framework. Sensitive HPC workloads are segregated on dedicated secure enclaves. Security considerations appear in the 2024 NAISS strategic report but are not architecturally central.
🇺🇸 USA (NITRD)
AdvancedNITRD operates a dedicated Cybersecurity R&D Strategic Plan (2023) parallel to the Big Data Plan. NIST published finalised post-quantum cryptography standards in August 2024. NASA SPD-41a open science mandates are explicitly balanced against data security requirements.
🇨🇦 Canada
PartialThe National Security Guidelines for Research Partnerships (2021) provide a framework for identifying and mitigating foreign interference risks — one of the most developed such frameworks globally for academic research.
🇫🇮
TiLa 2025 & DAHA 2030 — Cybersecurity & Research Security
What the documents say · What is missing
What TiLa & DAHA address
TiLa — Access controls & authenticationTiLa defines an access management capability layer, referencing HAKA federation and institutional authentication services as part of the service architecture. It treats identity and access as a solved infrastructure component.
TiLa — Data security classificationTiLa briefly covers data security levels and the need to match service offerings to data sensitivity tiers, distinguishing between openly available, restricted, and confidential research data.
DAHA — Sensitive data servicesDAHA dedicates a full capability chapter to sensitive data management, describing CSC's SD Desktop, SD Connect, and SD Apply services for processing personal and sensitive research data in compliance with GDPR. This is DAHA's most operationally developed security area.
DAHA — Legal compliance frameworkDAHA maps the Finnish legal landscape for research data, including the Act on the Openness of Government Activities, GDPR obligations, and sector-specific legislation (health data, criminal records). Security is treated primarily through the lens of regulatory compliance.
DAHA — Data lifecycle securityDAHA references security requirements at each stage of the data lifecycle (collection, storage, processing, publication, preservation), including data destruction procedures for sensitive datasets that must not be retained.
What is missing
Threat modelling & adversarial riskNeither document contains any adversarial threat analysis. State-sponsored espionage, ransomware targeting research infrastructure, and IP theft from HPC systems are not addressed. The 2025 YTF flagged this gap explicitly — noting geopolitical security as a new priority — but no architectural response exists yet.
NIS2 compliance architectureFinland transposed NIS2 in October 2024. CSC as a digital infrastructure operator is within scope. Neither TiLa nor DAHA maps the NIS2 obligations (incident reporting, supply chain due diligence, board-level accountability) onto research infrastructure architecture. This is a material documentation gap given CSC's regulatory obligations.
Post-quantum cryptography transitionDAHA covers long-term data preservation to 2030 and beyond, but contains no reference to the post-quantum cryptography migration challenge. Data encrypted today using RSA/ECC may become vulnerable to future quantum computers; a migration plan for long-term archives is absent.
LUMI security governanceLUMI is operated by CSC but used by researchers from all 11 LUMI consortium countries plus EuroHPC access countries. Neither TiLa nor DAHA addresses the security governance of a multi-country HPC system — screening of users from high-risk countries, dual-use research controls, or export-controlled compute workloads.
Research security (foreign interference)Unlike Canada's National Security Guidelines for Research Partnerships or UK guidance on protecting sensitive research, neither TiLa nor DAHA addresses the risk of foreign state actors targeting Finnish universities through research partnerships, personnel recruitment, or data exfiltration.
🤖
B — Artificial Intelligence
AI as infrastructure tier, as research workload, and as governance challenge

Artificial intelligence has undergone the most dramatic repositioning of any theme in this survey. In most 2020–2022 architectures AI appeared as a research application domain. By 2024–2025 it is treated as a first-class infrastructure tier requiring dedicated hardware, data pipelines, governance frameworks, and sovereign model development programmes. The LUMI AI Factory — one of the first seven EuroHPC AI Factories, opened October 2025 — places Finland at the centre of European AI infrastructure, making the absence of AI from TiLa's architecture particularly visible.

🇪🇺 EU (EuroHPC)
LeadingEuroHPC's mandate was extended to AI Factories by Council Regulation 2024/1732. 19 AI Factories and 13 Antennas are selected across Europe by late 2025. Finland's LUMI AI Factory (opened October 2025) and Germany's JUPITER AI Factory are flagship sites. The EU AI Act (2024) provides a risk-classification governance framework for AI systems in research.
🇬🇧 UK (UKRI DRI)
AdvancedUKRI's AI Research Resource (AIRR) provides dedicated AI compute for UK researchers. The £1B UK Compute Roadmap targets 20-fold expansion. The UK AI Safety Institute (2023) addresses AI governance across research and public sector applications.
🇸🇪 Sweden (NAISS)
AdvancedSweden is among the first seven EuroHPC AI Factory countries (MIMER AI Factory, December 2024). The MIMER initiative provides dedicated AI hardware within NAISS. A network of AI Factory Antennas including HEARTS (Switzerland) and AIFA-LAT (Latvia) link to MIMER.
🇺🇸 USA (NITRD)
AdvancedNITRD's 2024 Big Data Plan explicitly addresses generative AI, AI/ML data quality, and AI governance. The NSF National AI Research Resource (NAIRR) pilot launched 2024, providing federated AI compute access across the US. DOE exascale systems (Frontier, Aurora) are designed for AI+HPC convergence.
🇩🇪 Germany (NFDI)
AdvancedGermany hosts two EuroHPC AI Factories (JUPITER at Jülich and HammerHAI at HLRS). BMBF funds separate AI4Science initiatives. NFDI consortia increasingly integrate ML/AI tools and workflows into research data management.
🇦🇺 Australia
DevelopingAustralia's NRI Roadmap and Draft NDRI Strategy do not systematically address AI as infrastructure. Australia's National AI Strategy (2021) runs separately from research infrastructure planning without formal integration.
🇫🇮
TiLa 2025 & DAHA 2030 — Artificial Intelligence
What the documents say · What is missing
What TiLa & DAHA address
TiLa — Computing services for AI workloadsTiLa's service architecture covers batch and interactive computing resources, including GPU-accelerated services. AI/ML research is implicitly covered as a high-performance computing workload, addressed through CSC's compute service portfolio (Puhti, Mahti, and the LUMI system added to the service catalogue in 2022).
TiLa — Data services enabling AITiLa covers research data storage and the IDA (research data storage service) that underpins data-intensive workflows, including those used for AI training datasets. Data movement pipelines to compute are part of the service architecture.
DAHA — AI in research data management (future priority)DAHA v1.0 explicitly lists "AI in research data management" as one of the key priorities for the 2025–2027 YTF term. The document notes that AI tools will increasingly affect all stages of the data lifecycle, from data collection and quality control to discovery and analysis.
DAHA — Automated metadata and discoveryDAHA references automated metadata generation as a capability gap to be addressed, which implicitly includes AI-assisted metadata tools. The Fairdata services (Qvain, Etsin) are described as targets for future AI-assisted enhancement.
What is missing
LUMI AI Factory as a national AI infrastructure tierTiLa 2025 was written before the AI Factory era. The LUMI AI Factory — which opened October 2025, offers 13 services, and is described as Finland's primary national AI infrastructure for research — does not exist in TiLa's architecture. Any successor document must explicitly define the AI Factory as a distinct layer alongside HPC and data.
AI model management and lifecycleNeither document addresses the management of AI models as research outputs — model versioning, provenance, reproducibility, and sharing. As AI models become primary research artefacts (as in biomedicine and climate science), this is a significant gap in DAHA's data lifecycle framework.
EU AI Act complianceThe EU AI Act (2024) introduces binding risk-classification and transparency obligations for AI systems, including those deployed in research contexts and those integrated into public administration. Neither TiLa nor DAHA maps these obligations onto research infrastructure or CSC's service portfolio.
AI governance and research integrityThe rise of AI-generated content and AI-assisted research creates new research integrity challenges (reproducibility, hallucination, undisclosed AI use). Neither document addresses AI governance frameworks for research — a gap that several peer architectures (NITRD 2024, UK AI Safety Institute) are beginning to fill.
Training data infrastructureLarge AI models require curated, large-scale training datasets. Neither TiLa nor DAHA addresses the infrastructure needs of AI training data — curation pipelines, licensing, quality assurance, or national repositories for research-grade training corpora. This is distinct from standard research data management and requires new architectural elements.
⚛️
C — Quantum Computing
Integration of quantum systems into national research infrastructure and post-quantum security planning

Quantum computing has transitioned from a pure research topic to an infrastructure planning concern. EuroHPC has taken the most architecturally integrated approach, with six quantum computers deployed at HPC sites by late 2025. Finland participates in the LUMI-Q Consortium (eight countries), co-funding the VLQ quantum computer in Czechia, and is developing the LUMI-IQ experimental hybrid HPC-quantum-AI platform in Kajaani. Finland's Quantum Technology Strategy 2025–2035 sets ambitious national targets, creating a sharp contrast with TiLa's and DAHA's silence on the topic.

🇪🇺 EU (EuroHPC)
LeadingEuroHPC's 2024 mandate extension covers all quantum technologies. Six quantum computers are deployed by late 2025. All 27 EU member states signed the EuroQCI Declaration for quantum-safe communication. A Quantum Act is expected Q2 2026.
🇬🇧 UK (UKRI)
AdvancedUK National Quantum Technologies Programme 2024–2035 with £2.5B committed. Five UKRI quantum hubs funded. The National Quantum Computing Centre at Harwell provides assured capability. UK leads on quantum-safe cryptography transition planning.
🇦🇺 Australia
AdvancedNational Quantum Strategy (May 2023) targets global leadership by 2030. AU$893M total public investment by January 2024. Partnership with PsiQuantum to build a utility-scale fault-tolerant quantum computer in Brisbane.
🇺🇸 USA (NITRD)
AdvancedNational Quantum Initiative Act (2018) coordinates federal R&D. NIST published finalised post-quantum cryptography standards August 2024. DARPA/DOE Quantum Benchmarking Initiative aims to determine utility-scale feasibility by 2033.
🇩🇪 Germany (NFDI)
AdvancedGermany invested €650M in quantum technologies. Jülich Supercomputing Centre hosts analogue quantum simulators and one of the EuroHPC quantum computers. NFDI consortia are beginning to develop quantum workflows.
🇸🇪 Sweden (NAISS)
DevelopingSweden published a Quantum Agenda (2023) with nine priority areas. 2024 research bill designates quantum as a strategic research area with SEK 50–100M from 2027. Participates in LUMI-Q consortium alongside Finland.
🇫🇮
TiLa 2025 & DAHA 2030 — Quantum Computing
What the documents say · What is missing
What TiLa & DAHA address
TiLa — Quantum: absentTiLa 2025 contains no references to quantum computing. Published in January 2021, it predates Finland's national quantum strategy and the EuroHPC quantum mandate. The word "quantum" does not appear in TiLa's architecture document.
DAHA — Quantum: absent by designDAHA 2030 focuses on research data management rather than compute resources, so the absence of quantum computing as a data management challenge is partially by scope. However, quantum computing creates specific data challenges — quantum circuit representation, error-corrected qubit state data, and the management of quantum simulation outputs — that fall within DAHA's remit and are not addressed.
YTF 2025 acknowledgementThe 2025 YTF forum meeting explicitly listed quantum as one of three top priorities for any TiLa successor, alongside AI and geopolitical security. This is the first formal acknowledgement in the Finnish scientific computing governance framework that quantum requires architectural consideration — but it has not yet produced any documented architecture.
What is missing
Quantum as a service tier in TiLaA TiLa successor must define quantum computing as a distinct service category alongside classical HPC. This includes the LUMI-IQ experimental platform, access via LUMI-Q for VLQ (Czechia), and the future trajectory of hybrid classical-quantum-AI workflows. Without this, there is no national architectural home for Finnish quantum computing infrastructure.
Quantum data management in DAHAQuantum computing produces novel data types: quantum circuit files, QPU calibration data, error-correction logs, variational quantum eigensolver results, and quantum simulation outputs. None of these are covered in DAHA's data types or domain-specific guidance. As quantum research scales up in Finland, DAHA needs a quantum data management profile.
Post-quantum cryptography in DAHADAHA covers long-term preservation of research data to 2030 and beyond. Data encrypted today using RSA, ECC, or Diffie-Hellman key exchange will become vulnerable to cryptographically relevant quantum computers — a risk horizon that falls squarely within DAHA's 2030 timeframe. DAHA should include a post-quantum cryptography migration plan for long-term archives, aligned with NIST's 2024 PQC standards (CRYSTALS-Kyber, CRYSTALS-Dilithium, SPHINCS+).
Access governance for LUMI-QThe LUMI-Q consortium is a multi-country instrument involving Finland, Czech Republic, Sweden, Denmark, Poland, Norway, Belgium, and Netherlands. The governance of quantum access — who can access, for what purposes, under what export control regimes — is not addressed in any Finnish research infrastructure architecture document. This is particularly important given the growing international consensus on quantum export controls.
Connection to Finland's Quantum Strategy 2025–2035Finland published a national Quantum Technology Strategy 2025–2035 targeting €3B sector turnover and 10,000 jobs by 2035. This strategy was developed outside the TiLa/DAHA governance process and is not referenced in either document. A TiLa successor should explicitly map national quantum infrastructure against the strategy's objectives.
🤝
D — International Cooperation
Cross-border infrastructure sharing, data federation, bilateral agreements, and the open-science/security tension

International cooperation takes several forms: multilateral frameworks (EOSC, EuroHPC, RDA), bilateral agreements (NASA MOUs, AUKUS), and coordinated export controls (quantum technology restrictions, semiconductor controls). A notable tension has emerged between the open science norm and a growing security-oriented trend toward restricting technology transfer. LUMI, as pan-European shared infrastructure with users from all 11 consortium countries and all EuroHPC member states, sits directly at this intersection.

🇪🇺 EU (EOSC / EuroHPC)
LeadingEOSC is designed as a federated international cooperation mechanism. EuroHPC JU includes 32 European countries. A developing EOSC Federation model is being created for non-EU country participation. Horizon Europe funds international partnerships with LAC, Africa, and other regions.
🇩🇪 Germany (NFDI)
AdvancedNFDI has mandatory EOSC alignment, making it Europe's largest nationally-funded EOSC contributor. Domain consortia such as NFDI4Earth and NFDI4Chem participate in international disciplinary data platforms and RDA working groups.
🇬🇧 UK (UKRI)
AdvancedThe UK re-associated with Horizon Europe in 2024. UKRI maintains bilateral agreements with Australia, Canada, Japan, and the US. DARE UK's TRE model is increasingly referenced internationally as a blueprint for safe-access research infrastructure.
🇦🇺 Australia
AdvancedUnusually active bilateral cooperation: new UK-AU Space Bridge (October 2025), ESA Cooperative Agreement negotiations, NASA aeronautics partnerships, and JAXA collaborations. Australia co-hosted International Data Week 2025 (Brisbane). AUKUS intersects with research infrastructure through quantum and advanced computing export controls.
🇺🇸 USA (NITRD / NASA)
Leading (bilateral)NASA maintains the most extensive bilateral research cooperation portfolio of any agency surveyed. US technology export controls (quantum, semiconductors, AI chips) create structural tensions with open science norms, particularly for international HPC access.
🇨🇦 Canada
PartialThe National Security Guidelines for Research Partnerships (2021) introduced a security-screening framework for managing risks in international collaborations — one of the most developed such frameworks globally for academic research.
🇫🇮
TiLa 2025 & DAHA 2030 — International Cooperation
What the documents say · What is missing
What TiLa & DAHA address
TiLa — EOSC alignmentTiLa explicitly positions Finnish scientific computing within the EOSC landscape, describing the service architecture as Finland's contribution to the European Open Science Cloud. LUMI is referenced as the key European-scale resource in the Finnish portfolio, and TiLa describes services in terms of their EOSC interoperability.
TiLa — EuroHPC and LUMITiLa covers LUMI as part of the service architecture, noting Finland's role as host country within the LUMI consortium. The architecture acknowledges that LUMI serves a European user base and that national and international access allocation models differ.
DAHA — EOSC as reference frameworkDAHA explicitly aligns its data lifecycle model with the EOSC Architecture and Interoperability Framework, treating EOSC compliance as a design requirement for Finnish research data services. The DAHA capability model references EOSC Interoperability Framework (EIF) principles throughout.
DAHA — International open science normsDAHA references UNESCO's Recommendation on Open Science (2021) and the RDA FAIR principles as international frameworks that guide Finnish research data management. International alignment is treated as a normative rather than operational concern.
DAHA — Nordic cooperationDAHA mentions Nordic collaboration on research data infrastructure, including references to NordForsk and the context of coordinated Nordic RDM policy development, though without specifying concrete bilateral architecture agreements.
What is missing
Multi-country access governance for LUMILUMI is operated by CSC but used by researchers from 11 LUMI consortium countries plus EuroHPC member states. Neither TiLa nor DAHA defines governance arrangements for the international user community: how access decisions are made, how disputes are resolved, what obligations attach to non-Finnish users, and how Finnish national policy interacts with EuroHPC access rules.
NATO membership implicationsFinland joined NATO in April 2023. NATO membership creates new obligations around technology security, data sharing with allies, and restrictions on cooperation with adversarial states. The 2025 YTF flagged geopolitical context as a new architectural concern, but neither TiLa nor DAHA reflects any analysis of how NATO membership changes Finland's international cooperation posture for research infrastructure.
Quantum export controls and LUMI-QThe US, UK, Australia, Canada, and Netherlands have introduced coordinated export controls on quantum technologies (2024). Finland is a partner in LUMI-Q (eight-country quantum consortium). Neither TiLa nor DAHA addresses how export control regimes affect the governance of LUMI-Q access or the transfer of quantum computing results across national boundaries.
Research security screeningUnlike Canada's National Security Guidelines for Research Partnerships, Finnish research infrastructure architecture contains no framework for screening international collaborations or access requests that may pose security risks. Given LUMI's profile as Europe's most powerful publicly accessible supercomputer, this is a significant governance gap.
EU AI Act & technology sovereigntyThe EU AI Act and the January 2026 Cybersecurity Package introduce new technology sovereignty dimensions to cross-border infrastructure use. Neither TiLa nor DAHA addresses how Finnish research infrastructure will maintain EU technology sovereignty compliance when international users access LUMI, the AI Factory, or the LUMI-IQ quantum platform.
Cross-thematic summary

Across all four themes, Finland occupies a leading position in infrastructure — LUMI as top-10 global supercomputer, LUMI AI Factory, LUMI-IQ quantum platform, LUMI-Q consortium membership, CSC SD Services — but faces a consistent documentation gap.

TiLa 2025 was written before the AI Factory era, Finland's Quantum Technology Strategy, and the current geopolitical security environment. DAHA 2030 is more forward-looking but focuses on data management and does not address AI model lifecycle, post-quantum cryptography for preservation, or the international governance challenges that LUMI's pan-European role creates. The case for a TiLa 2030+ successor that explicitly addresses AI infrastructure tiers, quantum computing integration, cybersecurity design principles, post-quantum cryptography for research data, and international access governance is both architecturally urgent and strategically necessary given Finland's infrastructure leadership position in Europe.

7 of 7