EuroAI: Europe’s Path to AI Sovereignty

This memorandum proposes EuroAI, a comprehensive European AI sovereignty initiative structured as a Joint Undertaking modeled on ESA and EuroHPC. The program achieves three moonshot objectives by 2031-2035:

1. World’s #1 LLM trained entirely on European silicon (2031)

2. Complete European AI silicon autonomy via €43B semiconductor stack (2030)

3. Universal digitization of 132M legal deposit volumes creating world’s largest multilingual training corpus (2035)

Total investment: €58B over 10 years (€5.8B annually, 0.034% of EU budget)

Strategic rationale: Current trajectory leads to permanent dependence on US compute (Nvidia), US models (OpenAI/Anthropic), and US platforms. EuroAI creates complete vertical integration from training data through silicon to deployed models, exploiting Europe’s structural advantages in multilingual data, distributed research excellence, and patient capital.

Legal basis: Council Regulation establishing Joint Undertaking under Article 187 TFEU, requiring qualified majority (not unanimity)digital-strategy.europa+2​

Governing Board:

• 50% European Commission voting weight

• 50% member states, weighted by financial contribution

• Executive Director with 6-year term, operational autonomygov+1​

Budget: €58B (2026-2035)

• €20B EU budget contribution (InvestEU, Digital Europe, NextGenEU residuals)

• €18B member state direct contributions

• €20B via EIB leveraged lending (using €190B spare capacity at 290% authorized leverage)findevlab+1​

Geo-return coefficient: 0.90 (member states receive contracts worth 90% of contributions over 5-year rolling periods)ibanet​

Why Germany cannot block: QMV in Council, Commission controls 50% of board votes, and German industry demands participation to access €14B German allocationeunews+2​

• Function: Governance, procurement coordination, member state liaison

• Staff: 200 (program managers, legal, coordination)

• Model: ESA Paris HQ—political nerve center, not research labwikipedia+1​

• Function: Lead all LLM training runs (parity, world #1, sovereign)

• Infrastructure: BSC expansion to 150,000 European AI accelerators by 2031

• Staff: 500 researchers (model architects, training engineers)

• Portugal’s role: Co-host via Spain-Portugal consortium, permanent Portuguese researcher positionsdigital-strategy.europa+2​

• Function: Develop inference-time reasoning systems, Process Reward Models

• Why: Test-time compute scaling enables smaller models to outperform 14x larger competitorsopenreview+1​

• Staff: 300 researchers (optimization theory, reinforcement learning)

• Portugal’s participation: Mathematics departments contribute PRM development via ELLIS networkfct+1​

• Function: Design European training/inference accelerators

• Infrastructure: Co-located with TSMC Dresden fab, imec chiplet center

• Staff: 400 engineers (chip architecture, verification, physical design)

• Portugal’s contribution: Critical Software satellite office for validation services (€25M geo-return contracts)techsoda.substack+2​

• Function: Curate training data for 24 EU languages, lead Portuguese/Lusophone modeling

• Strategic importance: Portugal’s flagship contribution, indispensable for multilingual supremacy

• Staff: 200 specialists (computational linguists, cultural experts)

• Infrastructure: 5,000 GPUs for data processing, 100 petabytes storageairevolution.poltextlab+1​

• Function: Pre-deployment safety testing, bias evaluation, AI Act compliance verification

• Staff: 250 specialists (AI safety researchers, red teams, policy experts)

• Why: Independent testing equivalent to pharmaceutical clinical trialsdigital-strategy.europa+1​

• Function: Pilot deployments in healthcare, climate, manufacturing, public services, education

• Infrastructure: 15,000 GPUs for production inference

• Portugal’s participation: SNS healthcare pilots, Champalimaud radiology AI showcaseellis+1​

Total new construction: €2.5B (€350M average per center)wikipedia​

1. Sines, Portugal: 50,000 GPUs, Iberian co-hosting, operational 2027

2. Dresden, Germany: 60,000 GPUs, co-located with chip fab, operational 2027

3. Nordic Consortium: 40,000 GPUs, renewable-powered, operational 2028

Total: 150,000 GPU-equivalent capacity by 2031cfg+2​

Objective: 100B parameter model matching GPT-4 benchmarks, proving European technical viability

2026: €8B EuroAI Joint Undertaking approved. 60,000 H100 procurement. Portugal contributes €320Mlinkedin+1​

2027: EuroLLM-Parity-100B trained on 60,000 H100s. MMLU 86%, multilingual MMLU 82% average across 24 EU languages. Open-source Apache 2.0 releasemistral+1​

Cost: €320M compute + €50M data curationcudocompute​

Objective: Achieve LMSYS #1 ranking (1500+ ELO) trained entirely on European-designed accelerators

2026-2027: IPCEI-AST funds €15B European AI accelerator design consortium (Graphcore + STMicro + Infineon + universities)smart+1​

2028-2029: Prototype chips (7nm, 2000 TOPS INT8, 80GB HBM3) fabbed and tested on 10B parameter modelsidtechex+1​

2030: Volume production—50,000 European accelerators manufactured annually across Dresden + Catania + Barcelona fabs. Unit cost: €15k vs Nvidia’s €30keetimes+1​

Q1-Q2 2031: EuroLLM-Sovereign-200B training begins on 150,000 European accelerators. 30 trillion tokens, 6-month durationreddit+1​

Q4 2031: Model achieves world #1 via superior test-time compute reasoning + multilingual training depth. LMSYS ELO 1500+, MMLU 95%+, multilingual average 98%lmarena+1​

Strategic rationale: By 2031, US labs hit scaling diminishing returns; European advantage from algorithmic sophistication + unified infrastructure + multilingual corpusepoch+2​

Objective: Digitize 100% of legal deposit holdings (132M volumes), creating world’s largest multilingual training corpus

Scale:

• Germany: 25M volumes

• France: 18M volumes

• Spain: 12M volumes

• Italy: 15M volumes

• Poland: 10M volumes

• Portugal: 4M volumes

• Others: 48M volumeswikipedia+1​

Budget: €5B over 10 years

• €4.62B digitization (€35/volume average)

• €200M infrastructure (7 high-throughput scanning centers)

• €180M coordination/quality controldash.harvard+1​

Infrastructure: 7 regional hubs processing 50,000 books/day collective capacity by 2027cccc.ncte+1​

Legal framework: Article 3 TDM exception for cultural heritage preservation + statutory compensation model (€160M fund, €1.21/book to authors via CMOs)informationlabs+2​

Output: 20 trillion tokens from 80M digitized books by 2031, creating 5x larger book corpus than any US competitorairevolution.poltextlab+2​

Portugal’s role: Host Lusophone digitization hub in Lisbon, coordinate with Brazil/Angola/Mozambique. €400M investment creates 200 jobs, positions Portugal as global center for Portuguese language digital preservationbooks.openedition+1​

EIB operates at 210% leverage vs 290% authorized limit = €190B spare capacityfindevlab​

Mechanism:

• EIB deploys €20B toward AI semiconductor + infrastructure using existing capacity

• 20-year loans at 4-5% to European chipmaker consortia

• Maintains AAA rating, requires zero fiscal outlay from member stateseib+1​

Why legal: Council already approved 290% leverage ratio, no treaty change neededfindevlab​

Important Projects of Common European Interest allow up to 100% funding gap coverage for “first industrial deployment”competition-policy.europa+1​

Structure: Convert fragmented Chips Act subsidies into unified IPCEI-AST for European AI accelerator fabsbayern-innovativ+2​

Why Germany participates: Siemens, Infineon, BMW win €10B+ contracts via geo-return, making opposition politically toxic domesticallybruegel+1​

Legal basis: Article 127 TFEU secondary mandate to support EU economic policies “without prejudice to price stability”europarl.europa​

Mechanism: ECB purchases bonds issued by IPCEI semiconductor consortium on secondary markets at 1-2% below commercial rateseuroparl.europa​

Justification: Supply chain resilience in critical technologies prevents inflationary shocks from geopolitical disruptions (cite 2021 chip shortage)europarl.europa​

Why legal: Corporate debt (not sovereign), secondary market purchases, ECB independence prevents member state vetocogitatiopress+1​

Capital structure:

• €10B from EIB excess reserves (no member state call)

• €10B member state guarantees (contingent liabilities, not cash—doesn’t hit fiscal deficits)

• €5B Commission from NextGenEU/MFF flexibilityleap.luiss+2​

Leverage: €25B first-loss capital raises €300B in bond markets with implicit EIB guarantee (AAA-rated)findevlab​

Deployment: Exclusively to EU-headquartered semiconductor companies for fab construction, equipment, design ecosystemleap.luiss​

Portugal’s participation: Via Spain consortium share, no direct capital contribution requireddigital-strategy.europa+1​

• European AI accelerator architecture competitive with H100 class by 2030

• Consortium: Graphcore (neural processor IP) + STMicro (fab access) + Infineon (power management) + universities

• Target: 2000 TOPS INT8, 1000 TFLOPS BF16, 80GB HBM3, 500W TDPsmart+1​

Three leading-edge fabs producing 500,000 AI accelerators annually by 2030:

1. Dresden (TSMC partnership, German state aid)

2. Catania/Grenoble (STMicro, Italian/French co-funding)

3. Barcelona consortium (new Spanish-Portuguese-French facility)idtechex+2​

Process nodes: 7nm initially (2029), scaling to 5nm by 2031, targeting 3nm by 2033euronews+1​

• Lithography: ASML (Dutch)—already European monopolyifri​

• Advanced packaging: €3.2B Italian Silicon Box facility + chiplet centerseenewseurope+1​

• Materials: Soitec (French), European equipment partnershipssoitec​

Target: 80% of AI accelerator BOM value manufactured in Europe by 2030idtechex​

80M digitized volumes by 2031 = 20 trillion tokens, representing:linkedin+1​

• Language diversity: 24 EU languages with balanced representation vs English-dominated web

• Quality: Curated, professionally edited text vs noisy web scraping

• Cultural depth: 500+ years European intellectual heritage vs shallow internet content

• Legal access: TDM exception + statutory compensation creates corpus unavailable to US competitorsdash.harvard+1​

Three-part approach:

1. Split opposition coalition:francescatabor+2​

   • Authors: Neutralized via statutory compensation (€1.21/book, €160M total fund)

   • CMOs: Co-opted as administrators (€16-24M revenue over 10 years)

   • Publishers: Isolated, forced to negotiate or appear anti-cultural preservation

2. Cultural heritage framing:

   • Primary mission is preservation/public access, AI training is efficiency byproduct

   • Invokes Article 167 TFEU cultural policy mandate + UNESCO heritage conventionseur-lex.europa+1​

   • 60-70% of corpus is public domain (pre-1955)—start there, demonstrate successwikipedia​

3. Transparency advantage:

   • Maintain public database of all training materials (AI Act Article 53 compliance)

   • Revenue from EuroLLM flows to compensation fund with full attribution

   • Positions EuroAI as “good actor” vs opaque US labs training on pirated Books3 corpussaa-authors+1​

Pilot program (2026): Digitize 1M public domain volumes, train 10B parameter model, prove feasibility before in-copyright negotiationsdoria+1​

Opt-out mechanism: Publishers can exclude specific titles (not catalogs) but forfeit compensation. Behavioral economics ensures 75%+ participation for out-of-print backlistnytimes​

Why it collapsed (2019-2021):tandfonline​

• No concrete deliverable—vague “federated cloud” concept

• US hyperscalers infiltrated governance

• No industrial return mechanism—German companies saw no benefit

• Regulatory approach without infrastructure investment

1. Concrete moonshots: World #1 LLM is measurable, visible, politically compelling (Apollo Program model)research-and-innovation.europa+1​

2. Geo-return guarantees: Every member state sees direct industrial benefits proportional to contribution. Germany gets €14B semiconductor contracts, France gets €9B algorithm development, Portugal gets €400M infrastructure + cultural sovereigntyibanet​

3. Institutional proven model: ESA achieved European space competitiveness using identical structure—Joint Undertaking with geo-return has 50-year track recordwikipedia+1​

4. Multiple funding paths: Doesn’t depend on Eurobonds. Uses EIB leverage, IPCEI state aid exemptions, ECB corporate purchases—each legally defensible separately, unstoppable combinedcompetition-policy.europa+2​

5. Strategic patience: 2031 timeline allows proper execution vs rushed failures. Conservative estimates with risk buffersseifeur+1​

6. Asymmetric advantage exploitation:

   • Multilingual data (US can’t replicate 24-language legal deposit corpus)

   • Test-time compute (algorithmic sophistication over brute capital)

   • Patient capital (no VC exit pressure forcing premature commercialization)openreview+2​

€320M contribution (2% of EuroAI budget) generates:

Infrastructure:

• 50,000 GPU Sines datacenter (co-owned with Spain)

• €400M construction investment, 200 permanent jobs

• Revenue-generating by 2029: €1.2B annual returns from inference servicescfg+1​

Research leadership:

• Lisbon Multilingual AI Center (Portugal’s flagship, €100M facility)

• Permanent positions for Portuguese researchers at BSC Barcelona

• Priority access to frontier compute for Portuguese universitiesfct+2​

Cultural sovereignty:

• 4M Portuguese depósito legal volumes digitized (€140M investment)

• Portugal becomes global hub for Lusophone digital heritage (260M speakers)

• Portuguese literature preserved in world’s best AI training corpusbooks.openedition+1​

Industrial capacity:

• Critical Software €25M validation contracts (chip verification)

• Efacec cooling systems, Portuguese construction firms for datacenter

• 500+ jobs in AI supply chain by 2031ibanet+1​

• Southern European leadership: Portugal-Spain consortium counterbalances Franco-German axis

• Lusophone gateway: Brazil, Angola, Mozambique coordinate through Lisbon, not Paris or Berlin

• Academic prestige: Portuguese universities co-author papers on world’s #1 LLM, reversing brain drainfct+1​

Q1:

• Commission proposes EuroAI Joint Undertaking regulation

• EIB Board approves €50B Strategic Tech-EU lending facility

• Portugal-Spain consortium signs Sines datacenter agreement

Q2:

• European Parliament approves (QMV in Council, Commission backing ensures passage)

• GPU procurement launched (80,000 H100s, 6-9 month delivery)

• CENL framework agreement for library digitization

Q3:

• IPCEI-AST approved with €43B state commitments

• First digitization hub operational (1M book pilot begins)

• Lisbon Multilingual Center construction starts

Q4:

• EuroAI Governing Board inaugural meeting

• Executive Director appointed

• Member states ratify contribution agreements

Q1:

• Sines datacenter operational (35,000 GPUs)

• Pre-training data preparation complete (20T tokens)

Q2-Q4:

• EuroLLM-Parity-100B training run (4 months)

• Lisbon Multilingual Center operational

• 10M books digitized EU-wide

Q4:

• Public launch achieving GPT-4 parity

• Open-source release, inference-as-a-service at €2/M tokens

• Political validation: “Europe Achieves AI Frontier Status”

• European AI accelerator prototypes tested

• 50M books digitized (European Book Corpus operational)

• First 50,000 European chips manufactured

• EuroLLM captures 20% European market share

Q1-Q2: EuroLLM-Sovereign-200B training on 150,000 European accelerators

Q4:

• Launch achieving LMSYS #1 ranking (1500+ ELO)

• Trained entirely on European silicon

• 80M book corpus (5x larger than US competitors)

• Complete vertical sovereignty: data → silicon → model → deployment

• Remaining 52M books digitized (132M total complete)

• European chip production scales to 500,000 units/year

• EuroLLM continuous improvement via expanding corpus

• European AI ecosystem achieves 35% global market share

GPU supply chain delays:

• Diversify procurement (Nvidia + AMD + Intel)

• Cloud burst option for training if owned hardware delayed

• Staged deployment allows learning from early phases

Silicon development failures:

• Target 80% H100 performance initially, not 100%—easier to achieve

• Focus inference optimization (Europe’s actual need) not training performance

• Multiple design teams in parallel (redundancy)

Training run failures:

• Conservative timeline with 18-month buffers

• Checkpoint architecture allows restart without full loss

• Smaller validation runs before committing full resources

German opposition:

• Geo-return ensures German industry demands participation (€14B contracts)

• QMV prevents single-state veto

• Industrial lobbying (Siemens, BMW, SAP) overwhelms fiscal hawk objections

Copyright litigation:

• Strong TDM legal foundation (Article 3 exception)

• Statutory compensation removes author opposition

• Public domain pilot (1M books) demonstrates before in-copyright expansion

• Cultural heritage framing makes litigation politically toxic for publishers

Member state coordination failures:

• Joint Undertaking structure proven over 50 years (ESA, EuroHPC)

• Geo-return eliminates free-rider problems

• Commission’s 50% voting weight ensures cohesion

Cost overruns:

• Conservative €58B budget includes 20-30% contingency

• EIB leverage provides flexible additional capacity

• Revenue from inference services (2029+) offsets later-stage costs

Market adoption:

• Open-source release ensures usage regardless of commercial uptake

• Public sector procurement mandates create guaranteed demand

• Multilingual superiority creates defensible market niche

• ✓ 100B parameter model trained

• ✓ MMLU ≥85%, matches GPT-4 performance

• ✓ 10M books digitized across 24 languages

• ✓ 60,000 GPUs operational in European datacenters

• ✓ 50,000 daily active European users of EuroLLM API

• ✓ LMSYS Arena ELO ≥1500 (world #1 ranking)

• ✓ 150,000 European AI accelerators in production

• ✓ 80M digitized volumes in training corpus

• ✓ 500M daily queries served via European infrastructure

• ✓ 20% European AI market share (€40B annual economic impact)

• ✓ Zero dependency on non-EU silicon for training/inference

• ✓ 132M legal deposit volumes digitized (100% coverage)

• ✓ 500,000 European chips manufactured annually

• ✓ 35% global AI market share for European companies

• ✓ €200B cumulative European AI industry value created

• ✓ Complete vertical sovereignty: data → compute → models → applications

1. Draft EuroAI Joint Undertaking regulation for January 2026 Council submission

2. Initiate EIB lending facility negotiations (target December Board approval)

3. Convene CENL digitization framework discussions

4. Prepare IPCEI-AST notification to Competition Directorate

1. Portugal: Sign Spain-Portugal Sines consortium MOU, brief PM on national flagship status

2. Germany: Engage Siemens/Infineon on semiconductor consortium participation

3. France: BSC expansion funding commitment for training center

4. All: Designate national EuroAI coordinators, prepare Governing Board representatives

1. Expand mandate to include AI (regulatory amendment)

2. Accelerate AI Factories deployment timeline

3. Coordinate with EuroAI on infrastructure planning

1. Inventory digitization-ready collections

2. Assess infrastructure for high-throughput scanning

3. Prepare rights management frameworks

4. Designate liaison officers to Lisbon hub coordination

President Kennedy’s 1961 moon speech didn’t promise incremental progress—it committed to achieving impossible leadership within decade. EuroAI follows identical logic:

1961: “We choose to go to the moon”
2026: “We choose to build the world’s best AI”

Apollo’s components:

• Mission control (Governing Board)

• Distributed contractors (7 research centers)

• Geographic distribution (50 US states → 27 member states)

• Visible moonshot (lunar landing → world #1 LLM)

• Strategic patience (8-year timeline → 10-year timeline)

• Infrastructure permanence (NASA persists → EuroAI becomes permanent)

Why Apollo succeeded: Converted political fragmentation into asset via contractor distribution. Every Congressional district benefited, ensuring funding survived administrations.

Why EuroAI will succeed: Converts European fragmentation into asset via geo-return. Every member state benefits proportionally, ensuring funding survives electoral cycles and national governments.

The question isn’t whether Europe can achieve AI sovereignty—the technical path is clear. The question is whether European political leadership possesses the vision and courage to mobilize at civilizational scale.

Apollo landed humans on moon in 1969, eight years after commitment.
EuroAI will land Europe at #1 in AI by 2031, five years after commitment.

The difference: Europe has superior research talent, multilingual cultural depth, patient capital, and legal frameworks enabling training corpus access US competitors cannot match. What’s lacking is purely political will to coordinate existing advantages.

This memorandum provides the blueprint. Implementation requires only decision.

1. https://digital-strategy.ec.europa.eu/en/policies/high-performance-computing-joint-undertaking

2. https://www.gov.pl/attachment/9c03f03b-4c3f-466d-8dcd-c49d6732531e

3. https://en.wikipedia.org/wiki/European_High-Performance_Computing_Joint_Undertaking

4. https://findevlab.org/financing-europe-a-balance-sheet-analysis-of-the-european-investment-bank/

5. https://www.ibanet.org/Change-for-European-space-institutions-EU-Space-Agency

6. https://www.eunews.it/en/2025/03/10/germanys-no-to-defense-eurobonds-proposes-funding-joint-projects-with-eu-budget/

7. https://en.wikipedia.org/wiki/CERN

8. https://www.esa.int/Enabling_Support/Space_Engineering_Technology/European_Space_Agency_ESA

9. https://digital-strategy.ec.europa.eu/en/news/commission-receives-seven-proposals-ai-factories-which-will-boost-ai-innovation-eu

10. https://www.barcelona.cat/internationalwelcome/en/news/the-bsc-will-host-one-of-seven-european-ai-factories-to-drive-its-development-in-the-business-ecosystem-1464124

11. https://openreview.net/forum?id=4FWAwZtd2n

12. https://iclr.cc/virtual/2025/oral/31924

13. https://www.fct.pt/en/internacional/espaco-europeu-de-investigacao/eurohpc-ju/

https://ellis.eu

1. https://smart.gov.pl/wp-content/uploads/2025/06/Central-Narrative-for-the-Candidate-IPCEI-AST.pdf

2. https://airevolution.poltextlab.com/eurollm-9b-europes-new-multilingual-ai-model/

3. https://digital-strategy.ec.europa.eu/en/policies/ai-office

4. https://en.wikipedia.org/wiki/European_Artificial_Intelligence_Office

5. https://publicai.co/airbus-for-ai.pdf

6. https://cfg.eu/special-compute-zones-submission/

7. https://www.power-and-beyond.com/construction-of-tsmcs-dresden-chip-factory-scheduled-to-begin-in-2024-a-2e53f8ff68863f2c3e12b2645b2444ab/

8. https://www.linkedin.com/pulse/how-much-cost-train-gpt-4-mark-babayev-5yngf

9. https://mistral.ai/news/mistral-medium-3

10. https://www.cudocompute.com/blog/what-is-the-cost-of-training-large-language-models

11. https://competition-policy.ec.europa.eu/state-aid/ipcei_en

12. https://www.idtechex.com/en/research-article/strategic-silicon-geopolitics-is-redirecting-semiconductor-investment/33412

13. https://www.eetimes.eu/europe-semiconductor-plan-caught-between-vision-and-reality/

14. https://www.reddit.com/r/singularity/comments/1bi8rme/jensen_huang_just_gave_us_some_numbers_for_the/

15. https://seifeur.com/gpt-4-training-time/

16. https://lmarena.ai/leaderboard

17. https://epoch.ai/gradient-updates/why-gpt5-used-less-training-compute-than-gpt45-but-gpt6-probably-wont

18. https://www.tobyord.com/writing/inference-scaling-and-the-log-x-chart

19. https://en.wikipedia.org/wiki/Legal_deposit

20. https://dash.harvard.edu/bitstreams/7312037c-52e6-6bd4-e053-0100007fdf3b/download

21. https://digitization.archive.org/pricing/

22. https://cccc.ncte.org/cccc/committees/ip/2005developments/google/

23. https://stacks.wellcomecollection.org/digitisation-strategy-2020-2025-c965dd77624f

24. https://informationlabs.org/ai-training-and-eu-copyright-is-it-legal-a-deep-dive-into-the-tdm-exception/

25. https://www.francescatabor.com/articles/2025/7/12/european-parliament-demands-unwaivable-right-to-fair-pay-for-creators-in-generative-ai-training

26. https://www.saa-authors.eu/the-art-of-ai-authorisation-remuneration-and-transparency

27. https://books.openedition.org/cidehus/1760?lang=en

28. https://www.eib.org/attachments/lucalli/20240364_eib_group_operational_plan_2025_en.pdf

29. https://vdivde-it.de/en/node/2283

30. https://www.bayern-innovativ.de/en/emagazine/digitization/detail/ipcei-ast-new-funding-measure-for-europes-semiconductor-industry/

31. https://www.bruegel.org/policy-brief/joint-public-procurement-tool-european-union-industrial-policy

32. https://www.europarl.europa.eu/cmsdata/207502/Whelan_FINAL%20online.pdf

33. https://www.cogitatiopress.com/politicsandgovernance/article/download/9811/4196

34. https://leap.luiss.it/wp-content/uploads/2025/02/WP7.24-How-to-institutionalise-European-industrial-policy-leap.pdf

35. https://www.eca.europa.eu/en/publications/SR-2025-18

36. https://www.euronews.com/next/2024/08/20/first-european-high-performance-chips-to-be-made-in-dresden

37. http://www.ifri.org/en/memos/groundbreaking-chip-sovereignty-europes-strategic-push-semiconductor-race

38. https://www.eenewseurope.com/en/imec-to-set-up-e40m-chiplet-development-centre-in-germany/

39. https://www.soitec.com/home/technology/collaborative-funded-projects

40. https://euperspectives.eu/2025/10/eu-lawmakers-debate-new-copyright-rules-for-gen-ai/

41. https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX%3A52005DC0465

42. https://www.heritageresearch-hub.eu/funding/data-space-for-cultural-heritage-deployment/

43. https://www.doria.fi/bitstream/handle/10024/190644/KK_digitointiohjelma%202025%20en.pdf

44. https://www.nytimes.com/2010/11/18/business/global/18book.html

45. https://www.tandfonline.com/doi/full/10.1080/01402382.2025.2491962

46. https://research-and-innovation.ec.europa.eu/document/download/3fb7597a-1680-4dff-86a3-1b82b02ad4b0_en?filename=mission_oriented_r_and_i_policies_case_study_report_apollo_project-us.pdf&prefLang=it

47. https://digital-strategy.ec.europa.eu/en/policies/ai-factories