Additional Objectives

Pillar 1: Securing access to frontier AI

O1.1

Secure ongoing access to frontier AI systems

O1.1

Secure ongoing access to frontier AI systems

Why it matters

Europe’s prosperity and security will depend on access to frontier AI models. There have already been instances of access provided by foreign companies being suddenly withdrawn by foreign governments. Europe needs to safeguard its frontier capabilities. The risk of cutoff can be reduced through ‘compute for access’ deals that tie the provision of compute inputs to access guarantees, and by reducing the perceived risk of foreign providers hosting frontier models in Europe. Funding a European fast-follower can reduce the downside if access is restricted.

Why it matters

Recommendations at the Union level

01

Require model portability and multi-vendor terms in public AI procurement

Very high

01

Require model portability and multi-vendor terms in public AI procurement

Very high

02

Fund a collective fast-follower model and European-only high-security hosting

Very high

02

Fund a collective fast-follower model and European-only high-security hosting

Very high

03

Certify Europe as a secure importer and coordinate distillation enforcement

Very high

03

Certify Europe as a secure importer and coordinate distillation enforcement

Very high

01 Require model portability and multi-vendor terms in public AI procurement

01

Require model portability and multi-vendor terms in public AI procurement

Action

Include model portability as a condition in public procurement contracts to prevent European institutions and governments from becoming locked in to a single provider. This should at least apply for sensitive public services, such as those covered by Article 29 of the proposed Cloud and AI Development Act (CADA). At the EU level, model portability could be inserted into the proposed CADA, while the Act’s multi-vendor requirements could be strengthened (bearing in mind the limited scope of the Act, which does not cover AI systems and underlying models themselves). The public procurement directives can be similarly revised to prevent lock-in. Finally, the Commission can support the development of switching tools, building on the switching requirements in the Data Act and allocating funding through the Digital Europe Programme.

Action

Implementation

Potential instruments:

The proposed CADA, specifically inserting portability requirements and making multi-vendor provisions stronger. While other legal instruments, such as the Data Act, already introduce portability requirements, frontier AI procurement should be subject to stricter requirements, meaning that model providers must meet a higher standard to be considered for EU AI procurement.

EU public procurement rules can also be used to prevent lock-in, specifically through a revision of the proposed Public Procurement Act.

The Data Act explicitly addresses model switching, in particular through the switching rules outlined in Chapter VI.

Funding to support the creation of switching tools can be allocated through the Digital Europe Programme. The Programme includes a provision allocating funding for the deployment of critical technologies, which could be used for developing switching tools.

Potential first steps (next 12 months):

Q4 2026: As the Council and Parliament reach their final positions on CADA, aligned Member States and MEPs can submit amendments on portability and multi-vendor provisions in the Council working party and the Parliament’s committee stage.

Q1 2027: The Commission should include portability and multi-vendor clauses in its own contracts with AI providers, setting itself as the model for national procurement.

Q2 2027: Portability terms should be strengthened while the procurement directives are being negotiated. Funding should be allocated under the Digital Europe Programme to support the development of switching tools.

Success indicators: Portability requirements and stronger multi-vendor provisions in the trilogue text of CADA; portability clauses added to new EU AI contracts.

Implementation

Potential instruments:

EU public procurement rules can also be used to prevent lock-in, specifically through a revision of the proposed Public Procurement Act.

The Data Act explicitly addresses model switching, in particular through the switching rules outlined in Chapter VI.

Potential first steps (next 12 months):

Q1 2027: The Commission should include portability and multi-vendor clauses in its own contracts with AI providers, setting itself as the model for national procurement.

Success indicators: Portability requirements and stronger multi-vendor provisions in the trilogue text of CADA; portability clauses added to new EU AI contracts.

Considerations

These proposals can be mirrored at the national level to similarly prevent lock-in (see national recommendations below).

Considerations

These proposals can be mirrored at the national level to similarly prevent lock-in (see national recommendations below).

02 Fund a collective fast-follower model and European-only high-security hosting

02

Fund a collective fast-follower model and European-only high-security hosting

Action

Allocate European funding to a collective European fast-follower. The aim is to prevent funding from being diffused between different Member State candidates and ensure that Europe has the best chance of maintaining a credible fallback option for certain applications. To ensure European sovereignty for critical security purposes, develop a set of European-only hosting sites that are EU-owned and operated. Services hosted at these sites should be certified at the proposed CADA assurance Level 4.

Action

Implementation

Potential instruments:

Existing funding vehicles can support the development of gigafactories, for example the EuroHPC Joint Undertaking, which has an open call for AI gigafactories, and InvestAI, which includes a fund for AI gigafactories.

The proposed CADA for providing certification of services hosted in European-only sites. Level 4 will require EU ownership and control as well as a “high” cybersecurity certificate.

The Cybersecurity Act’s European Cybersecurity Certification Scheme for Cloud Services (EUCS) provides standards for high-security hosting.

Potential first steps (next 12 months):

Q4 2026: Establish a consortium to support the development of a European fast-follower, with co-funding committed by Member States.

Q1 2027: The Commission and EuroHPC Governing Board amend the selection criteria for AI gigafactories, ensuring that at least one facility is designated for European-only high-security hosting.

Q2 2027: Develop a certification for high-security, European-only hosting based on CADA Level 4 and the EUCS scheme.

Success indicators: A low capability lag of the European fast-follower compared to the commercial frontier; the opening of a certified European-only hosting site.

Implementation

Potential instruments:

The proposed CADA for providing certification of services hosted in European-only sites. Level 4 will require EU ownership and control as well as a “high” cybersecurity certificate.

The Cybersecurity Act’s European Cybersecurity Certification Scheme for Cloud Services (EUCS) provides standards for high-security hosting.

Potential first steps (next 12 months):

Q4 2026: Establish a consortium to support the development of a European fast-follower, with co-funding committed by Member States.

Q1 2027: The Commission and EuroHPC Governing Board amend the selection criteria for AI gigafactories, ensuring that at least one facility is designated for European-only high-security hosting.

Q2 2027: Develop a certification for high-security, European-only hosting based on CADA Level 4 and the EUCS scheme.

Success indicators: A low capability lag of the European fast-follower compared to the commercial frontier; the opening of a certified European-only hosting site.

Considerations

While a fast-follower would reduce dependence by providing a domestic option in domains where frontier performance is unnecessary, it would likely not in itself provide an adequate fallback for a world with transformative AI.

If frontier developers do not release their best models publicly, emulation is more difficult, meaning that the strategic value of having a fast-follower may decrease as compounding advantages widen the distance to the frontier.

Considerations

03 Certify Europe as a secure importer and coordinate distillation enforcement

03

Certify Europe as a secure importer and coordinate distillation enforcement

Action

Establish an EU-wide certification for secure hosting of frontier models to reduce the perceived risk for foreign companies of hosting their models in Europe. Data centre and model weight security requirements should be benchmarked to the security standards set by frontier AI companies and the US government. National enforcement to prevent model distillation and theft should be harmonised to ensure the highest standards apply wherever frontier models are hosted. Whenever possible, shared instruments and standards should be developed with US counterparts.

Action

Implementation

Potential instruments:

The CADA assurance framework (Levels 3–4) may be able to provide the scaffolding for a certification system for secure data centres. The CADA assurance levels should be more closely aligned with the hardware controls needed to credibly guarantee data centre security; the CADA framework may need to be amended to include these hardware-related requirements.

Cybersecurity Act certification (which is currently stalled) to provide the technical security scheme.

Certification can be benchmarked to counterpart measures in the US, for example on adversarial distillation and model weight security.

Common instruments can be developed with US partners through closed forums, for example the (currently dormant) Trade and Technology Council (TTC).

Potential first steps (next 12 months):

Q4 2026: Benchmarking of CADA Level 4 and EUCS-high requirements, possibly using RAND’s security-level benchmarks as the model (see Considerations below). Publish for international partner visibility.

Q1 2027: Pilot certification of a small number of sites. Form a EU–US technical working group on secure hosting and distillation enforcement, potentially inside a rebooted TTC.

Q2 2027: Establish a coordination protocol for national enforcement, to secure against industrial-scale distillation from EU bases. Ensure consistency between Member States and US measures.

Success indicators: number of sites certified to US standards; frontier model weights actually hosted on EU soil under the scheme; US recognition of the certification in hosting approvals; documented national enforcement actions against distillation that are consistent across Member States and coordinated with US action.

Implementation

Potential instruments:

Cybersecurity Act certification (which is currently stalled) to provide the technical security scheme.

Certification can be benchmarked to counterpart measures in the US, for example on adversarial distillation and model weight security.

Common instruments can be developed with US partners through closed forums, for example the (currently dormant) Trade and Technology Council (TTC).

Potential first steps (next 12 months):

Q1 2027: Pilot certification of a small number of sites. Form a EU–US technical working group on secure hosting and distillation enforcement, potentially inside a rebooted TTC.

Q2 2027: Establish a coordination protocol for national enforcement, to secure against industrial-scale distillation from EU bases. Ensure consistency between Member States and US measures.

Considerations

RAND’s security-level benchmarks can inform certification categories. Security Level 5 (SL5), for instance, refers to systems designed to resist top-tier state attackers, up to the highest standard needed to defend against attackers from the most cyber-capable institutions.

The primary audience for the certification is foreign governments and developers, so security measures must be visible to these actors.

Considerations

The primary audience for the certification is foreign governments and developers, so security measures must be visible to these actors.

Recommendations at the national level

01

Tie compute buildout to contractual access guarantees

Very high

01

Tie compute buildout to contractual access guarantees

Very high

02

Implement secure-importer standards nationally

Very high

02

Implement secure-importer standards nationally

Very high

01 Tie compute buildout to contractual access guarantees

01

Tie compute buildout to contractual access guarantees

Action

Member States should treat the inputs needed for frontier AI data centres, such as sites, energy, or connections to the grid, as strategic assets. They should be made available to frontier model providers only when attached to contractual access guarantees that include parity of access with foreign customers, continuity of service commitments, and European access to any security programmes with limited access (like Project Glasswing). These access guarantee provisions should apply to any hosting agreements above 100 MW, with Member States notifying the Commission of all such arrangements.

Action

Member State mode

Kind-specific, depending on the particular assets of a given state or region.

Member State mode

Kind-specific, depending on the particular assets of a given state or region.

Implementation

Potential instruments:

National investment-screening to identify relevant hosting arrangements.

Member State laws on permitting, energy, and grid connections, depending on national legal provisions and assets relevant to the deal.

Potential first steps (next 12 months):

Q4 2026: Develop a playbook for Member States, covering what guarantees should include and how they should be structured. This should cover access parity, continuity of access, and participation in limited access programmes.

Q1 2027: Member States should designate a negotiating unit spanning, inter alia, energy and digital ministries, depending on the configuration of assets and governing authority within a given state. Member States should notify the Commission whenever new deals are struck.

Success indicators: Share of newly contracted hosted capacity (MW) covered by access-guarantee clauses; notifications filed with the Commission; access parity being extended to European governments or companies during a restricted model release.

Implementation

Potential instruments:

National investment-screening to identify relevant hosting arrangements.

Member State laws on permitting, energy, and grid connections, depending on national legal provisions and assets relevant to the deal.

Potential first steps (next 12 months):

Considerations

The leverage from such compute buildout assets will decay as the US closes gaps in energy, capital, and talent.

National governments should mirror EU-level portability and multi-vendor procurement terms in public procurement and in the hosting deals themselves. As for procurement, Member States should avoid infrastructure deals that bind national systems to a single model provider.

Considerations

The leverage from such compute buildout assets will decay as the US closes gaps in energy, capital, and talent.

02 Implement secure-importer standards nationally

02

Implement secure-importer standards nationally

Action

Frontier AI data centres in Member States should be security certified up to the level demanded by foreign AI providers and their governments. This should include enforcing know-your-customer obligations as well as developing national enforcement competence aimed at preventing model distillation and theft.

Action

Member State mode

Universal for hosting states, applying to any Member State hosting or seeking to host frontier AI compute.

Member State mode

Universal for hosting states, applying to any Member State hosting or seeking to host frontier AI compute.

Implementation

Potential instruments:

Frontier AI data centres and cloud operators should be designated as critical entities, subject to heightened security requirements.

Know-your-customer obligations should apply to compute providers.

Distillation enforcement competence should be assigned to a named national authority, coupled with Europe-wide enforcement harmonisation.

Potential first steps (next 12 months):

Q4 2026: Audit existing and planned frontier-relevant sites.

Q1 2027: Adopt national KYC guidance for compute providers, aligned with the emerging Union scheme (Union-level recommendation 3). Table national legislation (or issue administrative direction) establishing distillation-enforcement competence and assigning it to a named national authority, with a Union coordination layer to ensure consistency (Union-level recommendation 3).

Q2 2027: Hold an exchange with US counterparts to discuss common enforcement requirements, for example for preventing distillation.

Success indicators: Number of national sites meeting SL4/SL5-class benchmarks; KYC coverage of national compute sales (share of capacity sold under verified-customer terms); documented enforcement actions against distillation taken in step with Union and US action.

Implementation

Potential instruments:

Frontier AI data centres and cloud operators should be designated as critical entities, subject to heightened security requirements.

Know-your-customer obligations should apply to compute providers.

Distillation enforcement competence should be assigned to a named national authority, coupled with Europe-wide enforcement harmonisation.

Potential first steps (next 12 months):

Q4 2026: Audit existing and planned frontier-relevant sites.

Q2 2027: Hold an exchange with US counterparts to discuss common enforcement requirements, for example for preventing distillation.

Considerations

This is a relatively cheap lever that protects against the US imposing security-motivated restrictions by reducing the perceived security risk of allowing capable models to run abroad.

Enforcement asymmetries between Member States will be exploited by evaders and may be used as a basis for denying access. Member States therefore need to work together through harmonised rules.

Considerations

This is a relatively cheap lever that protects against the US imposing security-motivated restrictions by reducing the perceived security risk of allowing capable models to run abroad.

Enforcement asymmetries between Member States will be exploited by evaders and may be used as a basis for denying access. Member States therefore need to work together through harmonised rules.

O1.2

Protect European assets

O1.2

Protect European assets

Why it matters

Europe does not currently systematically review how much of its own capital and technology is flowing into foreign AI and semiconductor industries as part of outbound investments. Member States do not have the legal power to compel a firm to disclose what it is building overseas. Moreover, some AI-assets of strategic importance are not fully covered by Europe’s inbound screening architecture, and Europe lacks a buyer of last resort that can inject capital when an asset is at imminent risk of being acquired.

Why it matters

Recommendations at the Union level

01

Adopt an outbound-investment regulation

High

01

Adopt an outbound-investment regulation

High

02

Make the outbound-investment review annual and comparable

High

02

Make the outbound-investment review annual and comparable

High

01 Adopt an outbound-investment regulation

01

Adopt an outbound-investment regulation

Action

Currently, the European Union does not have a regulation governing foreign investments of European firms. As of today, there exists only a Commission Recommendation asking Member States to review outbound investments. The EU should introduce a new regulation on the review of outbound investments related to critical technologies, modelled after Regulation (EU) 2019/452. The following principles should apply to outbound reviews:

Action

Each Member State has autonomy to choose which investments to screen. If a Member State does review an investment, it should meet common minimum standards and reviews should be shared in a common information-sharing system.

A Member State should notify the Commission and other Member States when they choose to review a transaction. Other Member States may comment, and the Commission may issue an opinion.

If a Member State chooses not to screen a transaction in its jurisdiction, another Member State may still comment and flag it for an opinion from the Commission. Competent Authorities should be established and equipped with the legal right to request details on firms’ investments.

To minimise administrative overhead, requests should be limited in scope and any information shared with the aforementioned authority must remain confidential; The Commission should establish a classified system and database for exchanging case information.

A Member State should notify the Commission and other Member States when they choose to review a transaction. Other Member States may comment, and the Commission may issue an opinion.

Implementation

Potential instruments: A new regulation on the review of outbound-investments modelled after the legal basis established in Regulation (EU) 2019/452.

Potential first steps (next 12 months):

Late 2026: The Commission mandates the Expert Group on Outbound Investment to outline the regulation’s scope, drawing on the comparative assessment published under Union-level recommendation 2.

Early 2027: The Commission opens a public consultation and starts work on an impact assessment, drawing on the comparative assessment published under Union-level recommendation 2.

Mid-2027: The Commission presents the proposal.

Late 2027: The Council working party begins examination and the European Parliament appoints a rapporteur.

Success indicators: Commission proposal presented by mid-2027; number of Member States with an outbound review regime under the regulation; number of transactions flagged through the cooperation mechanism per year; number of comments and opinions issued per year; number of staff assigned to outbound review per Member State.

Implementation

Potential instruments: A new regulation on the review of outbound-investments modelled after the legal basis established in Regulation (EU) 2019/452.

Potential first steps (next 12 months):

Late 2026: The Commission mandates the Expert Group on Outbound Investment to outline the regulation’s scope, drawing on the comparative assessment published under Union-level recommendation 2.

Early 2027: The Commission opens a public consultation and starts work on an impact assessment, drawing on the comparative assessment published under Union-level recommendation 2.

Mid-2027: The Commission presents the proposal.

Late 2027: The Council working party begins examination and the European Parliament appoints a rapporteur.

Considerations

A Union-level framework that requires (rather than facilitates and coordinates) outbound-investment screening is likely not politically feasible at this point. When the European Commission tabled its proposal for what would become Regulation (EU) 2019/452, 11 Member States already operated a national FDI screening mechanism. By contrast, Member States do not systematically review and assess outbound investment for security purposes as of today. Since this proposed regulation would require all Member States to build an institution that none currently has, it is unlikely to be readily adopted. However, as the experience with the 2019 FDI screening regulation shows, a voluntary outbound-investment regime modelled on that regulation could prepare the ground for a future regulation mandating outbound screening within a specific scope.

A voluntary regime would still reach cross-border cases. Under Article 7 of Regulation (EU) 2019/452, Member States may comment on an investment that another Member State is not screening, the Commission may issue an opinion, and the Member State concerned must give due consideration to both. A Member State that chooses not to review is therefore not outside the system. In cases where a company is incorporated in one Member State while the relevant technology sits in a subsidiary in another, the second Member State can raise the case even if the first has no regime of its own.

Harmonising reviews and risk assessments will help make transaction definitions and risk categories comparable among Member States. It will also build contact points across the Union by putting officials in touch with each other and enabling them to create secure lines of communication.

This regulation would not cover research arrangements. Imec, for example, may host American companies in its laboratories under collaboration agreements, which transfer technology and know-how without any investment taking place. Such arrangements fall outside any outbound-investment instrument. Closing this gap will require a separate regulatory intervention.

Considerations

02 Make the outbound-investment review annual and comparable

02

Make the outbound-investment review annual and comparable

Action

Recommendation (EU) 2025/63 asked Member States to take stock of their outbound investments in AI, semiconductors, and quantum, and to assess these investments for their economic security risks. However, this was a one-off exercise. The Commission should issue a follow-up Recommendation asking Member States to review and risk-assess their outbound investments yearly during the same time period and using the same format. The Commission should aggregate the results in a yearly report that covers trends in deals and risks that arise across countries, and how Member States differ in their handling of these risks. This report can be modelled on the Commission’s recurrent report on the screening of foreign direct investment into the Union. The report should also be transparent about where the underlying data is too thin to support a conclusion. Moreover, the Commission should leverage the results of the initial stocktake under Recommendation (EU) 2025/63 to agree on what transactions to review and how to score risks.

Action

Implementation

Potential Instruments: Recommendation (EU) 2025/63 and a successor recommendation establishing the annual cycle; the Commission's own analytical capacity for the comparative assessment.

Potential first steps (next 12 months):

Late 2026: The Commission publishes a first comparative assessment of the national review and risk assessment reports.

Mid-2027: The Commission adopts a successor recommendation fixing the annual reporting date, with the first annual cycle covering 2027.

Success indicators: Number of Member States submitting complete reports; publication of an annual Commission comparative assessment.

Implementation

Potential Instruments: Recommendation (EU) 2025/63 and a successor recommendation establishing the annual cycle; the Commission's own analytical capacity for the comparative assessment.

Potential first steps (next 12 months):

Late 2026: The Commission publishes a first comparative assessment of the national review and risk assessment reports.

Mid-2027: The Commission adopts a successor recommendation fixing the annual reporting date, with the first annual cycle covering 2027.

Success indicators: Number of Member States submitting complete reports; publication of an annual Commission comparative assessment.

Considerations

According to Recommendation (EU) 2025/63, the review and risk assessment should inform subsequent policy choices. However, the EU should not wait for their outcome before deciding whether EU-level legislation is needed. Member States currently lack a legal instrument to request outbound-investment information from firms, so the exercise cannot be expected to generate a complete evidence base. If only a few identified risks are found, this could reflect the limits of voluntary information-gathering rather than the absence of risk, and it should not be treated as a reason to close the file. The legal instrument that would give Member States such a right-to-ask is proposed in Union-level recommendation 1.

In its 2024 White Paper on outbound-investment controls, the Commission proposed to validate and compare the Member State reports in a comprehensive report of its own. Recommendation (EU) 2025/63 no longer mentions this, leaving it unclear whether it will happen. We urge the Commission to produce this report.

An annual outbound-investment review would be valuable not only for what it finds, but also for what it builds: A recurring cycle of reviews and risk assessments will instil a culture of monitoring and prudence in national administrations. It will also raise awareness among firms that they might have to disclose their outbound investments.

This recommendation is also meant to prepare the ground for the outbound-investment regulation proposed in Union-level recommendation 1. If the Union creates a template and a risk methodology for the purposes of this review, this can later be codified, so that the regulation is based on an instrument that administrations already use.

Considerations

Recommendations at the national level

01

Implement national outbound-investment controls

High

01

Implement national outbound-investment controls

High

02

Extend the scope of national inbound FDI screening

High

02

Extend the scope of national inbound FDI screening

High

03

Name a national buyer of last resort for strategic assets

High

03

Name a national buyer of last resort for strategic assets

High

01 Implement national outbound-investment controls

01

Implement national outbound-investment controls

Action

Member States with strategic AI-stack assets should introduce national outbound-investment controls. Which transactions are in scope can be determined based on Recommendation (EU) 2025/63, namely advanced semiconductors, artificial intelligence, and quantum technologies. The broader sectors highlighted in national-level recommendation 2 below should also be included. To minimise red tape, legislation should differentiate between two sets of transactions – a small set of transactions that are prohibited per se, as well as a broader set of transactions that require companies to notify relevant authorities after the investment is completed. Which transactions fall into which category should be determined in secondary legislation, so that they can be updated as technology evolves. To prevent regimes from overlapping, transfers which already require a licence under the dual use Regulation (Regulation (EU) 2021/821) should be excluded. Moreover, the authority responsible for inbound screening should be put in charge of implementation, so that the controls can draw on their expertise and established processes, rather than creating a parallel body. Adding one to three additional full-time staff members per Member State would be sufficient to cover these tasks.

Action

Member State mode

Kind-specific – only Member States hosting strategic AI-stack assets (e.g. the Netherlands, Germany, France, Belgium, Austria).

Member State mode

Kind-specific – only Member States hosting strategic AI-stack assets (e.g. the Netherlands, Germany, France, Belgium, Austria).

Implementation

Potential instruments: National primary legislation, either based on a dedicated outbound-investment control legislation or an extension of an existing inbound-investment screening or foreign trade instrument. Secondary legislation for the prohibited and notifiable lists, so that they can be updated without reopening the statute.

Potential first steps (next 12 months):

September–December 2026: The responsible ministry assesses whether the existing screening or foreign trade statute may cover outbound controls.

Early 2027: A draft bill is prepared and industry is consulted on the notification burden.

Mid-2027: The bill is introduced in parliament and a budget line for one to three additional full-time staff is agreed.

Success indicators: Number of Member States hosting strategic AI-stack assets with an outbound regime in force; number of transactions notified per year.

Implementation

Potential first steps (next 12 months):

September–December 2026: The responsible ministry assesses whether the existing screening or foreign trade statute may cover outbound controls.

Early 2027: A draft bill is prepared and industry is consulted on the notification burden.

Mid-2027: The bill is introduced in parliament and a budget line for one to three additional full-time staff is agreed.

Success indicators: Number of Member States hosting strategic AI-stack assets with an outbound regime in force; number of transactions notified per year.

Considerations

Although national action is recommended first, the Union should still implement its own outbound-investment controls. National regimes can be in force within two years and can later be folded into the Union framework proposed in Union-level recommendation 1, which is designed to accommodate divergent national regimes.

There is a real risk of creating a patchwork of national legislations. To avoid this, the Member States moving first should align on scope and definitions in the Expert Group from the start.

As highlighted above in Union recommendation 1, the regime will also require governments to have a statutory right to ask companies established in the Member State about their outbound investments, coupled with a duty to keep confidentiality and protect commercially sensitive information. This right should be limited to the small set of technologies in the regulation to prevent undue bureaucracy.

The United States has operated an outbound-investment regime since January 2025 under 31 CFR Part 850, which distinguishes a narrow set of prohibited transactions from a broader set requiring notification within 30 days of closing, with no approval pathway at all. This demonstrates that a notification-based regime is manageable and imposes low costs on investors.

Considerations

There is a real risk of creating a patchwork of national legislations. To avoid this, the Member States moving first should align on scope and definitions in the Expert Group from the start.

02 Extend the scope of national inbound FDI screening

02

Extend the scope of national inbound FDI screening

Action

The revised foreign screening Regulation (EU) 2026/1386 requires Member States to have a specified national screening mechanism in place by January 2028.²⁴ As per Article 4(16), Member States may apply their screening mechanism to foreign investments beyond the minimum scope that is determined in Article 4(15) and Annex I. Member States that hold strategic AI assets should use this option and bring these strategic assets within the scope of their screening mechanism. In particular, the target list should be extended to investments in companies that: (i) develop specialised AI systems for narrow but security-relevant purposes, such as biological and chemical design, offensive cyber capabilities, and the control of critical industrial processes; (ii) service or maintain semiconductor manufacturing equipment, such as ASML’s lithography machines; (iii) produce equipment on which critical digital infrastructure depends, including radio access networks and 6G equipment, submarine fibre-optic cable systems together with the vessels that install and repair them, and high-voltage grid equipment including large power transformers; (iv) hold strategic training datasets. Separately, Member States should apply Article 4(15)(d) as it stands to designate existing AI data centre operators as critical digital infrastructure.

Action

Member State Mode

Kind-specific – only Member States hosting strategic AI-stack assets (e.g. the Netherlands, Germany, France, Belgium, Austria). Universal for AI data centres.

Member State Mode

Kind-specific – only Member States hosting strategic AI-stack assets (e.g. the Netherlands, Germany, France, Belgium, Austria). Universal for AI data centres.

Implementation

Potential instruments: The national screening instrument currently in drafting ahead of the January 2028 deadline; the designation of existing data centre operators as critical digital infrastructure, requiring no primary legislation.

Potential first steps (next 12 months):

September 2026–September 2027: Passing the bills that extend the mandatory scope to categories (i) to (iv).

March 2027: Article 4(15)(d) designation of existing AI data centre operators adopted.

Success indicators: Number of the four asset categories within mandatory national scope, per Member State; whether AI data centre operators are designated as critical infrastructure.

Implementation

Potential first steps (next 12 months):

September 2026–September 2027: Passing the bills that extend the mandatory scope to categories (i) to (iv).

March 2027: Article 4(15)(d) designation of existing AI data centre operators adopted.

Success indicators: Number of the four asset categories within mandatory national scope, per Member State; whether AI data centre operators are designated as critical infrastructure.

Considerations

There is no Union-level legislative counterpart to this recommendation. However, the European Commission may consider issuing a non-binding recommendation for specific Member States to expand their minimum scope. Further, the Commission can give additional guidance on screening methodologies for AI-relevant sectors.

Greenfield investments are not within the scope of the revised inbound FDI screening regulation. This exclusion should stay as it is. Bringing greenfield projects into the mechanism would impose notification duties on many transactions that raise no economic security concern, and the EU has an interest in remaining an attractive destination for such investments.

Considerations

03 Name a national buyer of last resort for strategic assets

03

Name a national buyer of last resort for strategic assets

Action

FDI inbound screening mechanisms may block the foreign acquisition of a strategic AI asset, but that does not solve the owner’s potential capital constraints. To make sure that the relevant asset remains under European control, the Member States hosting these assets should establish an anti-capture facility to quickly mobilise the needed public and private capital in such situations. Rather than building an entirely new facility, Member States should designate an existing public bank or investment vehicle (such as Bpifrance in France, SFPIM in Belgium, or Invest-NL in the Netherlands). Authorities should make sure that the money is available on demand, and headline figures together with per-case ceilings should be published in advance. In addition, a confidential list of qualifying assets or companies should be maintained and reviewed regularly. This way, an asset’s eligibility is determined before a bid arrives, rather than having to be negotiated under time pressure. There should also be a channel of information exchange between the screening authorities and the capital facility, allowing the screening authority to alert the fund in critical situations.

Action

Member State Mode

Kind-specific: only Member States holding AI stack companies

Member State Mode

Kind-specific: only Member States holding AI stack companies

Implementation

Potential instruments: A ministerial instruction to an existing public bank or fund.

Potential first steps (next 12 months):

November 2026: Draft the confidential list of qualifying assets.

February 2027: Name the fund and issue the mandate, including the per-case ceiling.

Mid-2027: Legislate the information gateway alongside the national measures to be notified under Article 3(2) of Regulation (EU) 2026/1386 ahead of the January 2028 deadline.

Success indicators: Member States with a named fund and decision-maker; callable funding committed per country, in euros; days from trigger to final decision.

Implementation

Potential instruments: A ministerial instruction to an existing public bank or fund.

Potential first steps (next 12 months):

November 2026: Draft the confidential list of qualifying assets.

February 2027: Name the fund and issue the mandate, including the per-case ceiling.

Mid-2027: Legislate the information gateway alongside the national measures to be notified under Article 3(2) of Regulation (EU) 2026/1386 ahead of the January 2028 deadline.

Success indicators: Member States with a named fund and decision-maker; callable funding committed per country, in euros; days from trigger to final decision.

Considerations

This fund can be activated on short notice when a foreign buyer offers to buy an EU AI stack company and this company needs additional financing, either because it risks bankruptcy or because it needs funding to scale up. The aim is to prevent foreign control of strategic assets on specific occasions. Care should be taken to prevent firms from strategically pretending to have a potential foreign acquirer in order to unlock this fund. Responding to existing offers will also help to preclude any issues under state aid rules.

Give the public bank or fund a standing mandate to acquire assets on the national list, and let its own investment committee take the decision under its normal rules so as to enable swift decision-making.

The main reason to place this fund at Member State rather than Union level is that existing EU funding has mandates that do not fit. The European Tech Champions Initiative invests in other funds, not companies. The EIC (Fund) invests between €0.5 million and €30 million for minority stakes. The Scaleup Europe Fund is externally managed by EQT, which takes all investment decisions on commercial terms.

Not every Member State can do this. A country with a small screening unit and no public investment vehicle will not execute a large purchase in six weeks.

Considerations

Not every Member State can do this. A country with a small screening unit and no public investment vehicle will not execute a large purchase in six weeks.

Footnotes

24.

As part of the Industrial Accelerator Act (COM(2026)100) the Commission proposed an investment control regime in 2026 that is parallel to the revised FDI screening regulation (Regulation (EU) 2026/1386). However, unlike the FDI screening regulation, it is not designed to protect European assets but to leverage Europe’s market access to extract valuable capabilities in sectors where Europe is lagging behind, such as battery technologies and electric vehicles.

24.

Pillar 2: Building economic strength

O2.1

Make Europe the best place to build and scale high-growth companies

O2.1

Make Europe the best place to build and scale high-growth companies

Why it matters

Transformative AI may allow young innovative firms to reach global scale at unprecedented speed, concentrating fiscal value and strategic capability where those firms are located. The known competitiveness issues in the European Union therefore need to be pursued at a higher level of ambition, and most importantly, with a higher sense of urgency. It is therefore crucial that Member States and Union-level institutions collaborate towards the common goal of competitiveness to avoid further fragmentation of important input markets (first and foremost capital and labour). The recommendations below are mutually reinforcing. Implementing only selected elements will substantially reduce their effectiveness, or could even generate unintended outcomes. It is therefore crucial that the proposed reforms are pursued as a coherent package.

Why it matters

Recommendations at the Union level

01

Unlock institutional capital for European growth companies

Very high

01

Unlock institutional capital for European growth companies

Very high

02

Equip European science for the AI age

Very high

02

Equip European science for the AI age

Very high

03

Attract the world’s best AI talent

Very high

03

Attract the world’s best AI talent

Very high

04

Create European lead markets for defence innovation

High

04

Create European lead markets for defence innovation

High

05

Establish a Single European Growth Market

High

05

Establish a Single European Growth Market

High

01 Unlock institutional capital for European growth companies

01

Unlock institutional capital for European growth companies

Action

The Union should mobilise institutional capital for European growth firms. To achieve this, the Commission should launch a Tibi/WIN-style initiative at the Union level to secure investment commitments into European growth funds from large investors. The Union should make it easier for insurers and pension funds to invest in long-term equity while maintaining reasonable prudential safeguards. The Commission should widen General Block Exemption Regulation (GBER) provisions so Member States can fund viable innovation projects that cannot secure private financing because of market failures. Finally, the Market Integration and Supervision Package should become a core deliverable of the Savings and Investments Union.

Action

Implementation workstreams

Launch the European Institutional Investors Pact for Innovation: The Commission, together with the EIB, should launch the planned European Institutional Investors Pact for Innovation (hereafter ‘the Pact’) to secure voluntary commitments from investors to invest in European growth funds. This Pact should set common eligibility criteria for growth funds. Annual reports of commitments raised should distinguish between announced versus legally committed and deployed capital, using common definitions. The additionality of these commitments (whether the capital would have been mobilised in the absence of the Commission’s intervention) should be assessed independently. The EU-level commitments should work alongside national schemes without the same investment being counted twice in both the Union's and a Member State's totals. Investment decisions and due diligence should rest with participating investors.

Unlock institutional investment in long-term equity: The Commission and Member States should swiftly implement the Solvency II reform to make it easier for insurance companies to invest in equities and other long-term assets. To ensure this reform works as intended across the EU, the European Insurance and Occupational Pensions Authority (EIOPA) and national supervisors must apply the revised long-term equity rules consistently, so equivalent investments face the same capital charge. Pension funds should not face unnecessary restrictions on their investment choices. Parliament and Council should therefore swiftly adopt the reform of the Directive on Institutions for Occupational Retirement Provision (IORP II). Nonetheless, funds should preserve the prudent-person principle.

Financing for first commercial deployment: The Commission should create a streamlined route in the ongoing GBER review for Member States to support financing of first-of-a-kind (FOAK) and first-commercial-deployment projects, where there are demonstrated market failures. This support should be targeted and remain limited to the minimum necessary. It should absorb risks for private investors to crowd in private capital. The Commission and Member States should maintain tax-payer safeguards proportionate to the size of the aid. Safeguards should be designed to minimise administrative burden for beneficiary firms.

Decisions and first steps in the next 12 months:

By the end of 2026: Commission and EIB launch the European Institutional Investors Pact for Innovation with a first group of institutional investors. Participating investors should report capital committed and deployed separately.

By the end of 2026: The Commission incorporates a defined FOAK and first-commercial-deployment route into the final GBER.

Ahead of 30 January 2027: The Commission should adopt the remaining Solvency II technical standards and ensure that EIOPA and national supervisors are ready to apply the revised long-term-equity treatment consistently. From application, EIOPA should monitor divergent supervisory practices. Where national law or administrative practice conflicts with Union law, the Commission should take appropriate measures.

During the IORP II negotiations: The European Parliament should adopt its position promptly to enable an agreement that preserves a risk-based approach to long-term investment.

Implementation workstreams

Decisions and first steps in the next 12 months:

By the end of 2026: The Commission incorporates a defined FOAK and first-commercial-deployment route into the final GBER.

During the IORP II negotiations: The European Parliament should adopt its position promptly to enable an agreement that preserves a risk-based approach to long-term investment.

Measures of progress

Implementation indicators: Number of participating investors in the Pact, and their assets under management; number of national schemes recognised under the Pact; adoption of the FOAK state-aid route.

Performance indicators:

For the Pact: Announced and deployed capital; independently estimated additional capital; risk-adjusted returns by vintage and asset class.

For institutional capital: Annual net purchases and holdings of EU equities by EU pension funds and insurers (€ and % of assets); net new financing raised by EU firms from EU institutional investors.

For FOAK Projects: Share of FOAK projects reaching financial close on schedule; share entering commercial operation on schedule; follow-on deployment without equivalent FOAK support.

Longer-term outcomes for evaluation: Depth of EU late-stage financing; number and value of European growth rounds; size and performance of EU-based growth funds; share of European-founded firms able to scale substantially in Europe.

Measures of progress

Performance indicators:

For the Pact: Announced and deployed capital; independently estimated additional capital; risk-adjusted returns by vintage and asset class.

For FOAK Projects: Share of FOAK projects reaching financial close on schedule; share entering commercial operation on schedule; follow-on deployment without equivalent FOAK support.

Considerations

Pact reporting should show how mobilised capital is distributed across investors, funds, sectors, and Member States. It should assess whether the Pact is truly deepening Union capital markets and broadening access to growth capital.

Public co-investment of FOAK projects should avoid picking national winners. Therefore, it should be based on competitive selection (if multiple alternatives exist), proportionate aid, and a demonstrated financing gap.

Crowding out of private investment through GBER reforms should be evaluated and monitored. Where possible this should be done using ex-post comparison with similar projects closely below funding thresholds for better causal inference.

Considerations

02 Equip European science for the AI age

02

Equip European science for the AI age

Action

The proposed Horizon Europe programme for 2028–2034 should address three market failures that currently limit the productivity of European research and innovation and hold back breakthrough science. First, it should support the development and deployment of scientific public goods (e.g. advanced research tools) and advanced research infrastructure (e.g. self-driving laboratories). Second, it should create organisational capabilities that cannot easily be provided through conventional project grants or temporary research consortia. Third, it should strengthen incentives to produce reliable, reproducible, and complete scientific knowledge to prepare the European research sector for AI-based science.

Action

Implementation workstreams

Scientific tools and infrastructure as public goods: The Horizon Europe 2028–2034 programme should include a dedicated funding line for both the development and deployment of trusted, interoperable, and reusable research tools and infrastructure. This includes but is not limited to scientific software, AI models, curated datasets, advanced instruments, self-driving laboratories, and other experimental automation tools. Funding should also support technical teams that are required to maintain, improve, and support this infrastructure. Funding should be open to universities, research organisations, nonprofits, and companies, and reward reliability, scientific uptake, community adoption, and downstream research impact. Where fragmented demand prevents promising scientific tools from being developed or deployed, the Commission should implement advance market commitments or comparable demand-pull instruments tied to ambitious, measurable performance requirements for these tools.

New organisational capabilities: The Horizon Europe 2028–2034 programme should establish new organisational capabilities fit for addressing ambitious scientific and technological goals. It should complement the ERC’s support for individual researchers with a mechanism focused on organisational excellence, to address scientific problems that require large (interdisciplinary) teams. It should be open to new experimental organisational forms like Focused Research Organisations (FROs): purpose-built, time-limited teams combining scientists, engineers, and operators to overcome defined scientific bottlenecks (e.g. building a new instrument, biological resource, or open experimental infrastructure). The mechanism should provide high-performing research units and novel organisational models with large, flexible grants, subject to rigorous mid-term review and renewable for up to ten years. Selection should be international and excellence-based, with no geographic allocation or presumption of permanent funding. In the ongoing negotiations, Parliament and Council should ensure that the proposed ARPA-like function of the EIC will have legally protected operational autonomy (building on the German SPRIND experience), allowing expert-led control over programme design, portfolio management, and funding decisions, while maintaining transparent objectives, robust conflict-of-interest safeguards, and independent external evaluation.

Reliable science for the AI era: AI is already dramatically increasing the volume and speed of scientific output. This makes it even more important to improve the reliability, transparency, and evaluability of research. The Horizon Europe 2028–2034 programme should therefore both aim to make individual scientific results more reliable, and invest in the capacity to systematically evaluate how the science system works. This requires infrastructure that enables both humans and AI systems to assess, verify, and build on scientific knowledge. Horizon Europe should therefore support both the verification of scientific findings through replication, validation, and shared benchmarking, and the creation of comprehensive machine-readable scientific records that capture protocols, workflows, code, datasets, intermediate outputs, and negative or null results. On top of this, the Commission should establish a Metascience and Policy Experimentation Unit to monitor and evaluate the productivity of European science, with the authority to test and evaluate alternative ways of funding research. These include, but are not limited to: alternative selection procedures, like partial lotteries among proposals deemed equally fundable after expert review; faster and more flexible grants; milestone funding; and new organisational models. Successful approaches should have a defined route into mainstream programmes.

Decisions and first steps in the next 12 months:

Autumn 2026: Secure the legal basis and budget. In the Horizon Europe and MFF negotiations, Parliament and Council should secure funding for shared scientific tools, new research organisations, research verification, and metascience experiments. They should also grant the EIC full operation autonomy to design ARPA-style programmes and manage funding portfolios. The Commission should support these priorities in the negotiations and determine which activities belong under Horizon Europe and which belong under the European Competitiveness Fund.

Winter 2026/27 to spring 2027: Define how each proposal will work. The Commission should determine how each proposal will be funded and delivered. It should develop rules for the funding of long-term infrastructure funding, FRO-style grants, and the development of new scientific tools. These rules should define who can apply, what exactly will be funded, how proposals will be selected, and how performance will be assessed. To develop infrastructure for replication and machine-readable scientific records, the Commission should involve open science experts and technical teams with experience building widely used scientific databases and software. The Commission should define the mandate of the Metascience unit, and ensure that it has sufficient budget and legal freedom to run randomised experiments. Research funders and scientific organisations should help refine all of these arrangements and identify suitable first projects.

Spring to autumn 2027: Prepare the first funding rounds. The Commission and Member States should incorporate the proposals into Horizon Europe’s Strategic Plan and the 2028–2029 Work Programmes. They should allocate budgets and assign responsibility for delivery. Finally, the Commission should appoint the leadership of the Metascience Unit and experienced programme managers to turn the EIC into a true ARPA-like organisation.

From 2028: Fund the first projects and expand what works. The Commission should open the first calls for proposals for scientific tools and infrastructure and select an initial portfolio of projects. The EIC’s ARPA-like function should launch its first programmes. Approaches that perform well should receive larger budgets or be incorporated into later Work Programmes.

Implementation workstreams

Decisions and first steps in the next 12 months:

Measures of progress

Implementation indicators: Number of dedicated funding instruments and attached budget for all the different measures; share of supported research infrastructure receiving long-term funding.

Performance indicators: Number of institutions using the funded tools; cost and time required to produce and validate scientific results, by field; reliability and technical performance (e.g. precision) of tools; share of funding experiments that lead to changes in funding practice.

Longer-term outcomes: Scientific output quality e.g., the number of discoveries deemed important by the field community; user rates of European research tools and infrastructure.

Measures of progress

Implementation indicators: Number of dedicated funding instruments and attached budget for all the different measures; share of supported research infrastructure receiving long-term funding.

Longer-term outcomes: Scientific output quality e.g., the number of discoveries deemed important by the field community; user rates of European research tools and infrastructure.

Considerations

Long-term support for new institutions and infrastructure should be periodically reviewed and subject to competition, to keep funding contestable.

New instruments should be coordinated with existing programmes to avoid duplication and fragmentation.

When using demand-pull tools, reward adoption and scientific use; do not simply reward the creation of tools, datasets, or infrastructure.

Balance strategic investments and investigator-driven research to preserve scientific diversity and independence.

Make sure that reproducibility and reporting requirements improve scientific quality without unreasonable administrative burdens.

Experiment with different policies: Scale successful approaches and discontinue those that do not deliver results.

Considerations

Long-term support for new institutions and infrastructure should be periodically reviewed and subject to competition, to keep funding contestable.

New instruments should be coordinated with existing programmes to avoid duplication and fragmentation.

When using demand-pull tools, reward adoption and scientific use; do not simply reward the creation of tools, datasets, or infrastructure.

Balance strategic investments and investigator-driven research to preserve scientific diversity and independence.

Make sure that reproducibility and reporting requirements improve scientific quality without unreasonable administrative burdens.

Experiment with different policies: Scale successful approaches and discontinue those that do not deliver results.

03 Attract the world’s best AI talent

03

Attract the world’s best AI talent

Action

The Union should make it dramatically easier for international founders to hire top (AI) talent from across the world. To support recruitment and retention, young innovative firms should be able to offer competitive and portable equity compensation across the Single Market. The Union should also create a Founder Gateway: a single digital front end that founders can use to navigate the administrative procedures involved in setting up a business. This should include simpler, better-coordinated immigration procedures for the founders and talent that start-ups and growth firms need, while admission decisions should still rest with national authorities.

Action

Implementation workstreams

Competitive and portable equity compensation. Parliament and the Council should keep a workable employee stock option scheme in the EU Inc. Regulation; tax on stock options should be due only when employees obtain liquidity. The Commission should collaborate with Member States to make sure that the tax is still deferred when employees relocate, and to prevent double taxation. These arrangements should protect legitimate national tax bases and prevent employees from shifting employment-related income to a lower-tax jurisdiction shortly before liquidity. If binding common rules are needed, the Commission should propose the appropriate legal instrument. Otherwise, Member States should coordinate on national rules.

End-to-end Founder Gateway: The Commission should create a Founder Gateway within the Single Digital Gateway. Through this, founders should be able to access EU Inc. services, the European Business Wallet, and national administrative procedures by following a coherent user journey, built on the once-only principle. Following the UK Government Digital Service model, the Gateway should be built to consumer-grade usability standards under empowered Commission product leadership, with permanent capacity for multidisciplinary design, engineering, and service-delivery. Development should be iterative and driven by continuous testing with founders.

Bringing innovation talent to Europe: Next to facilitating the formation of businesses, the Founder Gateway should also streamline the administrative layer of immigration processes for founders, researchers, and highly skilled employees needed by start-ups and growth firms. The Gateway should provide access to relevant EU and national migration routes (including the Blue Card and Single Permit frameworks), and first and foremost to the dedicated founder pathways to be established by Member States (see the accompanying recommendations at the national level). It should support digital applications, status tracking, and coordinated processing of linked family applications. The Commission should extend the EU Digital Identity Wallet and Once-Only Technical System to allow verified personal and business information to be reused across national procedures. Individual admission decisions should nevertheless remain with the national authorities

Decisions and first steps in the next 12 months

Sep–Nov 2026: Establish governance and define scope: The Commission should designate a product lead for the Founder Gateway: This lead will be responsible for its delivery roadmap and budget, and empowered to coordinate work across Commission services. The product lead should then establish a joint implementation group with willing Member States and the national authorities responsible for the initial founder journeys. They should then publish a roadmap for a minimum viable product (MVP) identifying the initial journeys and participating Member States, and defining milestones, responsibilities, funding, and performance metrics. In the EU Inc. negotiations, Parliament and the Council should make sure that employees are taxed only when they sell their shares or otherwise receive liquidity. The Commission, together with Member State finance ministries and tax authorities, should start to develop rules to preserve tax deferral after relocation and prevent double taxation, ensuring the employment-related tax base is allocated according to where it was earned.

Dec 2026–Feb 2027: Resolve legal and policy design issues: The Commission should review where gaps in EU migration rules make it more difficult for founders, researchers, and highly skilled workers to migrate to Europe, and propose changes or additional common permit categories to address these. The Gateway product lead and participating Member States should agree on common Gateway service standards, covering digital applications, case tracking, accessibility in English, and the reuse of verified information. The Commission and participating Member States should identify the legal basis and data-sharing arrangements required for the initial Gateway journeys. They should prepare any necessary EU or national legislative changes in parallel with product development.

Mar–May 2027: Build the first software release: The Gateway product lead should establish a permanent product team, with dedicated design, engineering, and service-operations capabilities. This product team should develop the core Gateway service layer, incorporating the European Business Wallet. Together with delivery teams from participating Member States, it should connect the national systems needed for selected MVP journeys. The Commission product team should develop common user access, status-tracking, and secure data-sharing functions for the MVP. Once a MVP is developed, founders and growth firms should be asked to iteratively test complete journeys, and the results should be published.

Jun–Aug 2027: Pilot and prepare rollout: The Commission and willing Member States should launch the Gateway’s pilot. Collectively, the pilot should cover at least one founder journey in each of three areas: company formation, talent admission, and secure business-data sharing. Individual Member States may initially connect different journeys according to where progress can be made most quickly. Participating Member States should test that relevant credentials can be reused across borders. The product team should publish performance data for each pilot journey. Following the pilots, the product team should prepare the next phase of rollout: including additional journeys and Member States and preparing for any necessary legislative changes. Product development and legal changes should continue to proceed as parallel workstreams.

Implementation workstreams

Decisions and first steps in the next 12 months

Measures of progress

Implementation indicators: Number of participating Member States connected to the Gateway; number of financial institutions and public authorities accepting verified business data through the Gateway.

Performance indicators: Median time required to establish a company and complete linked administrative procedures through the Gateway; end-to-end task completion rates; median processing time for complete talent visa applications, including accompanying family members; share of applications using previously verified data; number of duplicate information requests; number and share of cross-border employee-equity cases benefiting from recognised portability arrangements.

Longer-term outcomes for evaluation: Share of startups and high-growth firms using employee equity compensation; administrative costs and time-to-market for firms expanding across the Single Market.

Measures of progress

Considerations

The Gateway should integrate with national systems and existing EU infrastructure; it should not itself create a new administrative layer.

Secure Member State commitment for the Gateway early before investing heavily in technology that depends on national participation.

Make end-to-end completion of key founder journeys the main objective of the Gateway; redirection to external portals should not be treated as success.

Resolve legal uncertainty upfront on data-sharing, digital credentials, migration procedures, and cross-border equity compensation.

Considerations

The Gateway should integrate with national systems and existing EU infrastructure; it should not itself create a new administrative layer.

Secure Member State commitment for the Gateway early before investing heavily in technology that depends on national participation.

Make end-to-end completion of key founder journeys the main objective of the Gateway; redirection to external portals should not be treated as success.

Resolve legal uncertainty upfront on data-sharing, digital credentials, migration procedures, and cross-border equity compensation.

04 Create European lead markets for defence innovation

04

Create European lead markets for defence innovation

Action

The Union should create a European common market for defence innovation. To achieve this, it should create integrated procurement pathways to allow common procurement for innovative products for willing Member States. It should further reduce regulatory fragmentation and lower cross-border entry barriers by creating common standards, interoperable interfaces, and shared operational testing, or by facilitating mutual recognition of test results for defence technology.

Action

Implementation workstreams

Market-building through rapid defence innovation. The Commission should launch its programme for agile and rapid defence innovation (AGILE) on schedule, and meet its four-month time-to-grant target. It should use mission-driven challenges and staged technical milestones where appropriate, and provide fast-track access to testing and certification. Each challenge should make it clear how the product can go from successful demonstration to operational testing and potential procurement. The Commission should scale those AGILE mechanisms that are evaluated to be effective in the pilot, under the next Multiannual Financial Framework. It should connect these to successor instruments for common procurement and industrial ramp-up.

Shared testing and operational validation. Europe should reduce inefficient duplication in national testing facilities and procedures to lower costs and make it easier for defence suppliers to enter markets across the Union. The Commission, together with participating Member States, should therefore establish a European network of testing facilities around existing activities of the Hub for EU Defence Innovation (HEDI), EU Defence Innovation Scheme (EUDIS), and AGILE. The network should use common evaluation standards, and national procurement authorities should recognise test results from anywhere across the network. EU-funded projects should have fast access to operational testing, and the next Multiannual Financial Framework (MFF) should fund gaps in shared testing capacity.

Support joint defence procurement. The EU should foster joint procurement by Member States with shared capability needs (see national recommendation 6). The Commission should prioritise European Defence Industry Programme (EDIP) support for joint procurement that replaces separate national orders, based on common functional requirements and binding budget commitments.

A single market for defence products. Parliament and the Council should remove unnecessary legal barriers to cross-border procurement and transfers, and adopt the Defence Readiness Omnibus proposals. The Commission should enforce the resulting rules and monitor whether national practices continue to favour domestic suppliers.

Decisions and first steps in the next 12 months:

Autumn 2026: Parliament and the Council should formally adopt AGILE and the Defence Readiness Omnibus. The Commission should then finalise AGILE’s 2027 work programme: It should identify the first innovation contests and define funding stages and performance measures.

By January 2027: The Commission should launch the first AGILE calls.

By March 2027: The Commission and European Defence Agency (EDA) together with interested Member States should identify testing facilities that could join the European network. Member States should identify rules that prevent EU companies from using testing facilities in their country. They should also identify rules that prevent authorities from accepting test results from the network.

By June 2027: The Commission and EDA should set the rules for access to the testing network and work with participating Member States to introduce common evaluation methods. Member States begin removing the barriers identified in March.

By September 2027: Member States begin connecting the first testing facilities to the network.

Implementation workstreams

Decisions and first steps in the next 12 months:

By January 2027: The Commission should launch the first AGILE calls.

By September 2027: Member States begin connecting the first testing facilities to the network.

Measures of progress

Implementation indicators: Number of testing facilities formally connected to the network.

Performance indicators: Share of AGILE grants signed within four months; share of completed tests accepted by procurement authorities in more than one country; median time from grant signature to first operational test and from successful test to procurement decision; share of procurement contracts awarded to first-time defence suppliers.

Longer-term outcomes for evaluation: Share and volume of cross-border defence contracts across all defence contracts in the Union.

Measures of progress

Implementation indicators: Number of testing facilities formally connected to the network.

Longer-term outcomes for evaluation: Share and volume of cross-border defence contracts across all defence contracts in the Union.

Considerations

Creating a single market for defence with more competition will ultimately lead to lower procurement costs. Larger and longer-term orders should still be awarded competitively and remain open to new entrants.

The goal of common standards is to promote interoperability without locking procurement into one supplier or technology. Where technological uncertainty remains high, programmes should preserve competing approaches until operational testing provides enough evidence to choose.

Considerations

05 Establish a Single European Growth Market

05

Establish a Single European Growth Market

Action

Europe’s public markets are fragmented across national venues. Concentrating more growth-company listings in one market could bring together the investors, analysts and advisers needed for successful IPOs. This would create a self-reinforcing ecosystem that attracts further listings and investment. While such concentration may occur naturally through competition between various European exchanges, the process is slowed by regulatory fragmentation and language barriers. A designated, but non-binding ‘focal venue’ could build market depth more quickly by coordinating business and investor expectations. The Commission should therefore coordinate an open process to designate one exchange as Europe’s non-binding focal market for growth-company IPOs. To reduce regulatory barriers to deeper public capital markets, Parliament and Council should swiftly agree on the Market Integration and Supervision Package.

Action

Implementation workstreams

Selecting the focal market. The Commission should invite European exchanges to present plans for attracting growth-company listings. It should select the strongest candidate using published criteria on liquidity, ease of access for international investors and the strength of its existing IPO ecosystem. The European Securities and Markets Authority (ESMA), interested Member States, issuers, and institutional investors should advise the Commission, but the final decision should remain with the Commission. The designation should last for a fixed period and be reviewed against published performance measures (see below).

Enabling cross-border regulatory framework. Parliament and Council should swiftly agree on the Market Integration and Supervision Package while preserving the Commission proposal’s core market-integration measures. The final rules should provide a workable PEMO framework that allows a single operator to operate trading venues across Member States under one ESMA authorisation, without additional host-country requirements. They should also harmonise trading venue rules and facilitate efficient cross-border access to clearing and settlement services.

Decisions and first steps in the next 12 months

Sep–Nov 2026: Test market appetite. The Commission should seek political backing from Member States and identify an initial group willing to support the focal market. In parallel, it should test interest among (capital) market participants. In this process it should focus on identifying barriers to participation and commitments needed to make the focal market viable and inform its market design.

Nov 2026–Jan 2027: Set the framework. If there is sufficient political and market support, the Commission and participating Member States should agree on the designation process and how they will promote the selected market to issuers and investors.

Jan–Mar 2027: Invite and assess proposals. The Commission should publish the selection criteria and invite applications from European exchanges. Applicants should show that they already have a strong position in European technology listings and explain how the designation would help them reach a globally competitive scale. An independent expert panel should assess the applications. To asses the credibility of each application, the panel should seek confidential structured feedback from market participants. Panel members and respondents should disclose relevant commercial interests, and the Commission should publish an anonymised summary of the findings.

Mar–Jun 2027: Select and launch the focal ecosystem. The Commission should complete the assessment and designate a focal market if a candidate meets the required standard. No ecosystem should be designated as a focal market unless it has a credible route to critical mass. Participating Member States should connect relevant IPO-readiness and investor-development initiatives to the focal ecosystem.

Implementation workstreams

Decisions and first steps in the next 12 months

Measures of progress

Implementation indicators: Share of national IPO-readiness programmes aligned with the focal market.

Performance indicators: Share of eligible European growth-company listings choosing the focal market, and volume raised; median IPO costs (% of proceeds); secondary-market turnover (% of market capitalisation); capital provided by international institutional investors (% of focal market IPO proceeds).

Long-term outcomes: Share of European growth companies listing outside the EU; number of cross-border listings on the focal market and € raised; volume of follow-on equity raised by focal-market companies.

Measures of progress

Implementation indicators: Share of national IPO-readiness programmes aligned with the focal market.

Considerations

Select an ecosystem only if it has a credible path to critical mass. If no candidate meets the standard, the Commission should make no designation. Still, the designation should not restrict where companies may list or give the selected exchange exclusive legal rights.

Member States should support the common process and avoid incentives or rules designed primarily to steer companies towards national exchanges.

Selection must be based on objective performance criteria, and rule out geographic rotation or political allocation. The ultimate goal is agglomeration.

Considerations

Member States should support the common process and avoid incentives or rules designed primarily to steer companies towards national exchanges.

Selection must be based on objective performance criteria, and rule out geographic rotation or political allocation. The ultimate goal is agglomeration.

Recommendations at the national level

01

Mobilise national capital for European growth

Very high

01

Mobilise national capital for European growth

Very high

02

Boost national research and innovation productivity

Very high

02

Boost national research and innovation productivity

Very high

03

Make Europe the best place to attract and reward top talent

Very high

03

Make Europe the best place to attract and reward top talent

Very high

04

Make it easier for innovative companies to take risks and recover from failure

High

04

Make it easier for innovative companies to take risks and recover from failure

High

05

Enable companies to scale across the Single Market

High

05

Enable companies to scale across the Single Market

High

06

Use procurement to build European lead markets for defence

High

06

Use procurement to build European lead markets for defence

High

01 Mobilise national capital for European growth

01

Mobilise national capital for European growth

Action

In order to facilitate European growth firms’ access to capital and lower funding costs, Member States must build a bigger pool of domestic capital. Member States should implement Tibi- or WIN-style schemes to mobilise additional institutional investment and help growth funds reach efficient scale. National regulators should remove unjustified barriers that prevent pension funds from investing in well-governed venture and growth funds. Where appropriate, Member States should offer targeted tax incentives that encourage additional angel investment in early-stage firms. Member States should support a focal European growth market that attracts specialist investors and builds liquidity over time, making larger European listings and exits more viable. Finally, in line with recommendations at the Union level, Member States should swiftly agree on the Market Integration and Supervision Package while preserving the Commission proposal’s core market-integration measures.

Action

Member State mode

Collective with kind-specific and coalition workstreams. All Member States should help deepen European capital markets and avoid new national measures that fragment them. Tibi/WIN-style schemes are largely useful in countries with large domestic investors and established growth funds. Angel investment incentives are most relevant to Member States with a strong pipeline of potential angel investors.

Member State mode

Implementation workstreams

Tibi-style investor mobilisation – Kind-specific. Member States with large institutional investor sectors should establish Tibi/WIN-style schemes that mobilise additional investment into accredited venture and growth funds. Participating investors should maintain autonomy over their investment decisions. Member States should coordinate reporting with the European Institutional Investors Pact for Innovation (see Union recommendations) and equivalent national schemes so that each commitment is counted only once.

First steps in the next 12 months: The government should appoint a lead ministry and an implementing coordinator, such as the treasury or national promotional bank. The lead ministry should convene investors to define the scheme’s scope and develop transparent criteria for the eligibility of venture and growth funds. The implementing body should then publish these criteria and invite venture and growth funds to apply for accreditation. Accreditation should only confirm that a fund is eligible for investment counted under the scheme. Participating investors should then commit investment volumes into this portfolio of accredited venture/growth funds, while the specific allocation should remain their responsibility. Reporting on the progress of such a scheme should differentiate between capital committed and deployed. The convenor is responsible for coordinating the reporting with the European Pact and equivalent national schemes to prevent double-counting.

Success indicators: Capital committed, and actually deployed; independently estimated additional investment; cross-border share of investment.

Unlock institutional capital – Collective. Insurers should be able to invest more easily in long-term equity. Member States should therefore swiftly transpose the Solvency II reforms. As part of the implementation of these reforms, national supervisory bodies should coordinate with EIOPA to make sure the revised long-term equity rules are interpreted and applied consistently across the Union. To allow pension funds to invest more easily in long-term equity, Member States should transpose the IORP II reforms without adding unjustified national restrictions, once they are adopted on the Union level. Member States should also enable smaller pension funds to invest in long-term assets through diversified pooled funds.

First steps in the next 12 months: National ministries should swiftly translate the Solvency II package into national law. Following these reforms, EIOPA and national supervisors agree on common guidance on the eligibility tests for long-term equity. The national supervisory decisions in the years after should be regularly reviewed. Once IORP II is adopted, the responsible ministry drafts the national implementing law without adding new investment restrictions. The responsible national ministries should consult smaller pension funds to identify legal barriers to pooled investment and swiftly remove them.

Success indicators: Insurers’ long-term equity holdings (€ and % of assets); volume and share of assets of pension funds invested in European long-term equity; volume and share of assets of smaller pension funds invested through diversified pooled funds; number of diverging applications of eligibility for long-term equity (under Solvency II) across the Union.

Targeted tax incentives for angel investment – Kind-specific. To mobilise additional angel investment, Member States with weak early-stage equity markets should introduce EIS/SEIS-style tax relief for individuals who invest in young firms raising new equity. The relief should address a documented funding gap and comply with EU State-aid rules. It should be limited to investors who are not connected to the company. Investors should continue to bear genuine commercial risk. To further strengthen the Single Market, investments in eligible EU-companies should not be excluded from relief unless there is a good objective justification for it. The tax scheme should be reviewed after a fixed period and only continued if it generates additional investment at a reasonable fiscal cost.

First steps in the next 12 months: The ministry of finance should estimate how large the early-stage financing gap is and how much additional investment the tax incentive would generate. It should only proceed if a cost-benefit analysis suggests the tax scheme would be the best option. In this case, it should set clear eligibility and anti-avoidance rules, and establish the scheme in law.

Success indicators: Additional private investment mobilised by the tax scheme; fiscal cost per additional euro invested; investment raised after firms leave the scheme; growth of supported firms relative to comparable firms (%).

Alignment with a European focal growth market – Coalition. Member States should support the European focal market for growth company listings proposed in Union recommendation 5. To speed up the growth of such a listing ecosystem, Member States should remove unnecessary barriers to cross-border listings and refrain from adding policies that steer companies towards separate national markets. If Union-wide agreement cannot be reached quickly, a coalition of willing Member States should establish a common framework and keep it open to others.

First steps in the next 12 months: Participating governments should commit politically to the common focal market. The ministry of finance should review national policies and subsidies that could stand in the way of companies that decide to list in the common focal market. National promotional banks and enterprise agencies should help growth companies prepare to list on the common market.

Success indicators: Number of participating markets and their share of EU market capitalisation; number of companies choosing to do their primary listing in the focal market, and their market capitalisation.

Implementation workstreams

Success indicators: Capital committed, and actually deployed; independently estimated additional investment; cross-border share of investment.

Considerations

Mobilisation schemes should not limit investor freedom by prescribing geographic quotas or weakening fiduciary and prudential standards.

Common EU recognition of national mobilisation schemes should avoid duplicate certification and double-counting, without requiring identical national investment policies.

Pension reform should protect retirement incomes, not force investment into inefficient assets.

Angel investment incentives should be targeted, capped, and temporary, and withdrawn if they mainly subsidise investments that would have happened anyway.

Member States should not use public financial support to divert liquidity towards national exchanges.

Considerations

Mobilisation schemes should not limit investor freedom by prescribing geographic quotas or weakening fiduciary and prudential standards.

Common EU recognition of national mobilisation schemes should avoid duplicate certification and double-counting, without requiring identical national investment policies.

Pension reform should protect retirement incomes, not force investment into inefficient assets.

Angel investment incentives should be targeted, capped, and temporary, and withdrawn if they mainly subsidise investments that would have happened anyway.

Member States should not use public financial support to divert liquidity towards national exchanges.

02 Boost national research and innovation productivity

02

Boost national research and innovation productivity

Action

To foster economic growth driven by scientific discovery, Member States need to strengthen the institutions that support frontier research and its commercialisation. Member States should strengthen their capabilities to do high-risk/high-reward research. For this, they should establish or expand ARPA-style organisations, and national research funders should experiment with new funding models. Research institutions should seize the opportunities opened by AI-driven science. Member States should build or get access to metascience capabilities to regularly evaluate the productivity of their research and innovation system. To translate these discoveries into successful companies, public research institutions should adopt transparent and founder-friendly spin-out terms.

Action

Member State mode

Collective with kind-specific delivery. All Member States should build metascience capacity and improve their spin-out terms. Member States with research institution systems large enough to sustain autonomous portfolio-based agencies should establish or expand ARPA-style programmes. Smaller Member States should form joint or cross-border arrangements.

Member State mode

Implementation workstreams

ARPA-style funding structures – Kind-specific. Member States should establish or expand ARPA-style funding structures, drawing on the SPRIND model. These should have legally guaranteed operational independence, expert control over funding decisions, and explicit tolerance of project failure. Member States should remove rules that impede rapid funding, active portfolio management, and new research organisations.

First steps in the next 12 months: The responsible ministry should prepare the legal and governance framework for the ARPA-style funding body. It is crucial that this agency has operational independence (especially for programme managers) from political interest. Where needed, parliaments should support any legislation needed to protect this operational independence. The government should then recruit a leadership team with exceptional scientific, technological, and entrepreneurial judgement. This team should start operations by selecting priority areas. Qualified programme managers should be hired for each domain.

Success indicators: Time from application to funding decision; share of projects redirected or stopped following technical reviews; share of projects meeting independently verified technical milestones; follow-up funding secured by completed projects within three years (amount and share of completed projects receiving any follow-up funding).

Experimenting with funding mechanisms – Universal for autonomous public research funders. Conventional grant programmes are not necessarily the best fit for the future of AI-integrated research. Autonomous public research funders should therefore devote a defined share of their budgets to test and evaluate alternative funding mechanisms in order to identify which funding approaches deliver better research outcomes. Experiments could include rapid small grants awarded through short applications and accelerated review, milestone-based funding, and lotteries among proposals that meet a certain quality threshold. Funding bodies should make use of a limited number of ‘golden tickets’ allowing expert reviewers to support promising high-risk proposals that would otherwise not be funded.

First steps in the next 12 months: The research funder should allocate an experimentation budget, identify appropriate comparison groups, and pre-register evaluation plans. It should then launch controlled trials of alternative funding mechanisms. Experimental results should be published where feasible.

Success indicators: Share of the funder’s budget allocated to funding experiments (%); application-to-award time and administrative cost per award (days and €); quality of research outcomes relative to comparable conventional grants, based on predefined and potentially field-specific measures.

Independent metascience capabilities – Collective. To be able to monitor the productivity of its science system, each Member State should ensure access to independent metascience capability, nationally or through formal cross-border arrangements. This metascience organisation should systematically monitor and evaluate research-system performance. It should be able to suggest experiments for funding methods, and recommend improvements to how research funding is organised and allocated. National bodies should share methods and evidence through the European Research Area and the proposed European metascience function (Union recommendation 2).

First steps in the next 12 months: In countries with a significant research system, the responsible ministry should designate an existing body or establish a metascience unit with statutory independence and secured multiannual funding. The unit should appoint a scientific leadership team, with expertise in policy evaluation and the evaluation of incentive schemes. This team should set an initial research agenda to evaluate the productivity of the existing research system. If administrative research data exists, the unit should get access to it. Otherwise, it should make sure such a dataset (including funding decisions and funding programmes by actors in the science system) is built. Public research funders should provide the necessary data and cooperate in evaluations and controlled funding experiments. National units should agree on common evidence and data standards through the European Research Area. Countries with smaller research systems should collaborate to build joint metascience units.

Success indicators: Share of funding programmes changed following evaluation; amount of funding reallocated towards better-performing approaches; and uptake of the unit’s recommendations by public research funders.

Prepare research systems for AI-enabled science – Collective. Member States should prepare their research systems to use AI, automation, and robotics to accelerate scientific discovery. They should financially support new research tools and shared infrastructure that integrate AI with experimentation, including automated biofoundries and self-driving laboratories. They should facilitate the sharing of practical lessons on how to organise research around these tools between research institutions and funders. Where the needs for expertise or investment exceed national capacity, Member States should build joint facilities. The metascience body should assess how AI-enabled research affects scientific productivity, reproducibility, and integrity, and identify which organisational and funding models work best.

First steps in the next 12 months: Responsible ministries and research funders should identify existing AI research tools and facilities, determine what is missing, and fund the first shared tools and pilot facilities. They should determine which facilities require national or cross-border scale and designate institutions to lead their development. Research funders should create a practitioner network to exchange practical experience on how to re-organise science around these AI tools to reap the biggest productivity benefits. The metascience body should prepare baseline measures and evaluation plans.

Success indicators: Number of AI-enabled tools and facilities launched; facility utilisation rate; reduction in experiment time and cost; share of fully reproducible AI-enabled research papers, per field; share of AI-enabled research outputs independently reused or validated.

Founder-compatible spin-out terms – Universal. Publicly funded universities and research organisations should make it easier for researchers to turn into founders by adopting attractive spin-out terms. This includes speedy decision timelines, so researchers do not waste years waiting for an intellectual property (IP) agreement. Terms should reflect the institution’s contribution, preserve founders’ incentives, and allow for follow-on investment, while complying with State-aid rules. Institutional policies should follow the EU Code of Practice for knowledge valorisation. Once the EU Blueprint for IP licensing and spin-off creation is finalised, universities should follow its guidance to develop standard terms.

First steps in the next 12 months: Research institutions should review recent spin-out agreements and consult founders and investors on what would be needed to make agreements attractive. They should then publish standard term sheets covering institutional equity, royalties, IP access, and founder obligations. These term sheets should set decision deadlines and approval responsibilities for standard cases. Any departure from the published terms should require written justification.

Success indicators: Median time to agree on spin-out terms; share of use of standard (rather than bespoke) terms; external equity raised by spin-outs within 12 months (volume, and share of spin-outs raising any external equity)

Implementation workstreams

Considerations

Operational independence of ARPA-style organisations should be protected in law and practice, within clear missions, budgets, and conflict-of-interest rules.

Project failure should be tolerated across the portfolio of ARPA-style organisations, but not where it reflects weak governance, research misconduct, or undisclosed conflicts. The suggested performance indicators (e.g. share of projects reaching milestones) have no universal optimal levels and should therefore not be treated as targets. Rather, they should monitor whether the organisation is making use of its independence to pursue high-risk, high-reward projects. In this case, one would expect, for example, a positive but small share of projects reaching milestones.

Member States should test different funding approaches, not try to converge on a single model.

AI-enabled research should be subject to safeguards. Areas for safeguards include reproducibility, human accountability, research integrity, including disclosure of AI use and prevention of fabricated results, cybersecurity, and dual-use risks.

Institutional equity shares in spin-outs should normally be low, but may be higher where the institution has provided substantial IP, infrastructure, or development funding.

Considerations

Operational independence of ARPA-style organisations should be protected in law and practice, within clear missions, budgets, and conflict-of-interest rules.

Member States should test different funding approaches, not try to converge on a single model.

Institutional equity shares in spin-outs should normally be low, but may be higher where the institution has provided substantial IP, infrastructure, or development funding.

03 Make Europe the best place to attract and reward top talent

03

Make Europe the best place to attract and reward top talent

Action

Member States should make it easier for innovative firms to attract and retain talent. To help younger firms compete, even when cash is scarce, they should allow employee-equity schemes to be taxed competitively. Member States should also provide fast and predictable migration admission routes for founders, researchers, and highly skilled workers, making it easier for innovative firms to recruit talent from across the world. More broadly, national migration rules and administrative procedures should minimise unnecessary burdens on growing firms. Public services should be founder-friendly, make greater use of digital and once-only processes (including through the Founder Gateway), and support mobility across the Union.

Action

Member State mode

Universal. All Member States should make employee-equity taxation and talent admission procedures internationally competitive (though national tax rates, migration volumes, and institutional arrangements may differ).

Member State mode

Implementation workstreams

Competitive employee-equity taxation – Universal. Member States should allow tax on EU Inc. employee stock options to be deferred until employees sell their shares or otherwise obtain liquidity. National rules should make this tax deferral predictable, and ensure competitive tax rates. Most importantly, the new tax scheme needs to be workable in practice. Member States should apply similarly attractive rules to employee stock options issued under national law. Income tax should be deferred until the shares are sold, while any subsequent increase in value should be taxed as a capital gain. When employees relocate within the EU, relocation itself should not trigger taxation, and Member States should avoid the double taxation of employee equity. At the same time, each country should keep the right to tax the share of equity gains earned within this country. Cross-border and anti-avoidance rules should prevent employment income from being shifted to lower-tax jurisdictions or reclassified as capital gains.

First steps in the next 12 months: The finance ministry and tax authority should review employee-equity taxation and identify the changes needed to defer tax until employees sell their shares or otherwise obtain liquidity. They should then introduce the necessary legislation and publish administrative guidance. The finance ministry and tax authority should work with the Commission and other Member States to find arrangements that preserve the deferral of taxation and prevent double taxation when an employee moves across borders within the Union. To maintain fairness and prevent employees from avoiding tax by relocating, they should agree on how to divide the taxable value from equity options between the countries where the employee worked.

Success indicators: Employee equity taxed before employees receive liquidity (% of employee-equity cases); share of firms offering employee equity; number of cases of early or double taxation after relocation within the EU.

Fast and specialised talent admission – Universal. Europe needs an easier migration route for top (AI) talent (researchers, founders, and highly skilled workers). Member States should therefore create dedicated migration units for innovation talent that can assess qualifications competently and decide quickly, ideally within 30 days. To enable fast case-processing and make applying a smooth experience for applicants, the entire process (including the verification of qualifications) should be fully digital, following the once-only principle, and applicants should be able to track their case online. This digital workstream must be compatible with the Founder Gateway (Union recommendation 3) and should be available in English. Talent seeking to migrate with their family should be able to submit one application for the whole family. Non-EU nationals with a credible plan to build a business in Europe should have access to a dedicated founder visa. To make sure that these visas do not create loopholes in the migration system, they should be valid for a limited time only (e.g. six months), and then renewed upon evidence of actual ongoing company formation. Changing career plans should not entail substantial additional paperwork. Instead, a simple change-of-status procedure should allow these permit holders to move between research, employment, and entrepreneurship without applying for residence again.

First steps in the next 12 months: The ministry of the interior, working with ministries of economy and research where relevant, should designate the innovation talent unit and review existing migration routes for founders, researchers, and highly skilled workers. The unit should be sufficiently staffed to be able to process applications within 30 days, with a target processing time of 14 days. In case no fully digital application system exists, the innovation talent unit should start to pursue the product development of such a platform, and make sure that it will be compatible with the Founder Gateway on the EU level (see also next recommendation). Where no suitable route for founders exists, the interior ministry should propose a new legal pathway.

Success indicators: Share of complete applications decided within 14 and 30 days; median processing time for family applications; number of founder visas issued; share of founder visas leading to company formation within 12 months.

National service integration for the Founder Gateway – Universal. Member States should make it possible for founders to complete key business procedures through the Founder Gateway (Union level recommendation 3). To this end, they should simplify and digitalise the national procedures used by founders. Progressively, these national procedures should be connected to the Gateway through common EU standards and interfaces. So founders do not have to enter the same information multiple times, authorities should reuse verified information already held by another public authority. They should accept forms and supporting documents in English, and be able to provide case support in English as well. Where necessary, national rules should be changed to enable this. Member States should start this process with the founder journeys that can be simplified and connected most quickly to the Gateway. Over time, all the main bureaucratic procedures for founders should be covered through the Gateway.

First steps in the next 12 months: Each government should appoint its digital agency (or an equivalent technical delivery team) to lead national integration with the Founder Gateway. This team should map the main founder journeys and select at least one that can be simplified and connected quickly. To build new digital processes (or at least the integration into the gateway), teams of designers, software engineers, and service operators should collaborate. They should be mindful of the once-only principle and do systematic user-testing with founders before connecting it to the Gateway. In parallel, legal teams together with responsible ministries should resolve regulatory barriers. By month nine, early-adopter Member States should have connected at least one tested process. With the help of the Commission’s coordination, the connected early journeys should collectively cover company formation, talent admission, and secure business-data sharing, so that the broader Gateway integration on the European level can also be tested. Other Member States should simplify at least one founder procedure and adopt a roadmap for connecting their founder services to the Gateway.

Success indicators: End-to-end completion rate and median completion time for journeys connected to the Gateway; number of repeated data requests; user satisfaction scores.

Implementation workstreams

Success indicators: End-to-end completion rate and median completion time for journeys connected to the Gateway; number of repeated data requests; user satisfaction scores.

Considerations

Fast admission routes should cover the entire process, including qualification recognition, security checks, and applications for accompanying family members.

Founder visas should follow a staged pathway, since “plans to found” upon entry are difficult to verify. An initial permit should be granted on the basis of a concise business proposal and potentially an interview assessing the applicant’s experience and commitment. Renewal should require evidence of real business formation activity (e.g. product development, customer interviews, meetings with potential investors, spending tied to building the business). As the process of company formation differs a lot across sectors and business models, authorities should assess the evidence as a whole against published criteria. Commercial success should not be required at the first renewal.

Considerations

Fast admission routes should cover the entire process, including qualification recognition, security checks, and applications for accompanying family members.

04 Make it easier for innovative companies to take risks and recover from failure

04

Make it easier for innovative companies to take risks and recover from failure

Action

As part of broader flexicurity reforms (as recommended in Objective 3.2), Member States should reduce the cost of business failure and restructuring while preserving worker security. They should implement Union insolvency reforms effectively and ensure that viable firms can restructure quickly, while allowing the capital and workers of unviable firms to move to more productive activities. They should also support the simplified liquidation procedures proposed for EU Inc.

Action

Member State mode

Universal with nationally differentiated delivery. All Member States should reduce excessive and unpredictable restructuring costs, while adapting reforms to their national labour market institutions, collective bargaining systems, and social protection arrangements.

Member State mode

The specific implementation details for this recommendation can be found in Objective 3.2.

05 Enable companies to scale across the Single Market

05

Enable companies to scale across the Single Market

Action

In order to help companies scale across Europe, Member States need to lower unnecessary fixed costs of entering new markets. This would simultaneously make it easier for firms, especially smaller ones, to expand across the EU, and strengthen competition. Firms should not have to repeat the same procedures across different jurisdictions. Instead, authorities should reuse evidence and recognise decisions from other Member States wherever possible. Member States must also apply Single Market rules across the Union without unjustified national additions, and provide a rapid route for challenging cross-border market-access decisions.

Action

Member State mode

Universal. Even though the barriers that need to be removed will differ across Member States, all Member States should be disciplined in applying these changes. The goal should be a single Union market.

Member State mode

Implementation workstreams

Prevent gold-plating and inconsistent implementation – Universal. Member States should avoid adding requirements beyond Union law unless they can show that these additional barriers to the creation of a Single Market are justified by clear public benefits. The burden of proof should rest with the authority proposing the additional requirement. Member States should also implement Single Market legislation fully and on time. National provisions that exceed Union requirements should be periodically reassessed and revised where they are no longer justified.

First steps in the next 12 months: Each ministry and regulator should review how EU law in its area of responsibility is translated into national practice. They should use evidence from the Commission’s Terrible Ten work, Single Market Enforcement Taskforce (SMET), and SOLVIT to identify national laws and enforcement practices that create unnecessary barriers to companies from other Member States who want to enter local markets. Ministries and regulators should revise those standards going beyond EU law without good justification. They should clearly communicate the changes to everyone who applies these new rules. To prevent unjustified new barriers, ministries and regulators should justify that wider EU benefits outweigh the additional entry costs when introducing additional rules beyond EU law. They should train responsible officials to apply EU Single Market rules and proportionality tests consistently.

Success indicators: Compliance costs caused by national requirements that go beyond EU law (split into differences with and without justification); annual savings from removing unjustified requirements; share of new national laws that go beyond EU law with a published cost-benefit assessment.

Expand mutual recognition and regulatory reliance – Universal. Companies entering another EU market should not have to resubmit documents already accepted by a public authority. National authorities should recognise evidence and decisions from other Member States unless EU law or a legitimate public-interest concern requires a separate assessment.

First steps in the next 12 months: Sector ministries should ask domestic companies where they faced duplicate requirements when trying to enter other Member State markets. National Single Market coordinators should use these cases to identify recurring barriers and work through SMET and the Commission to remove them.

Success indicators: Share of cross-border applications completed without resubmitting documents; costs of duplicate evidence requirements; median time to resolve market-access cases.

Provide rapid review for cross-border market-access barriers – Universal. Member States should require authorities to give written reasons for restricting market access and set deadlines for reviewing disputed decisions. Cases that remain unresolved should be referred to SOLVIT. The Commission and national authorities should track recurring cross-border market-access barriers and work through SMET to remove them.

First steps in the next 12 months: Member States should introduce deadlines for the review of dispute cases and ensure that national SOLVIT centres have the staff and authority needed to handle urgent cases.

Success indicators: Median time to resolve cross-border market-access disputes; number of dispute cases resolved without litigation.

Implementation workstreams

Success indicators: Share of cross-border applications completed without resubmitting documents; costs of duplicate evidence requirements; median time to resolve market-access cases.

Success indicators: Median time to resolve cross-border market-access disputes; number of dispute cases resolved without litigation.

Considerations

Make sure that removed requirements are not reintroduced or maintained in other forms, e.g., in administrative practice. Member States must therefore clearly communicate simplified rules to everyone affected.

Assess firms’ compliance costs in practice, not just the number of requirements set out in law.

Considerations

Assess firms’ compliance costs in practice, not just the number of requirements set out in law.

06 Use procurement to build European lead markets for defence

06

Use procurement to build European lead markets for defence

Action

To create a larger demand side, Member States should coordinate to create a single European market for defence capabilities. Member States should use defence budgets to create a credible route from technical validation to procurement at scale for innovative firms. Defence procurement processes should define the capabilities needed and award contracts based on performance in paid competitive trials. Procurement authorities should provide a clear, budget-backed route for successful prototypes to production at scale. Production contracts should be awarded based on operational performance of the technology alongside its cost and interoperability with existing systems.

Action

Member State mode

Coalition with universal national reforms. All Member States with significant defence procurement should improve access for innovative firms. Those with shared capability needs should agree on common requirements and pool their procurement budgets.

Member State mode

Implementation workstreams

Multiannual demand and open innovation competitions – Universal for procuring Member States. Procurement authorities should publish multiannual capability needs and run open innovation contests to acquire these capabilities. These contests should focus on technical performance and include paid trials of several promising solutions (if multiple exist), to preserve competition during development. Successful providers should be admitted to multi-supplier framework contracts and compete for follow-on orders. For small and fast-changing technologies, Member States should also allow authorised military units to directly place limited operational orders. Eligibility rules must be proportionate and not unnecessarily rule out startups. Where relevant, contests should link to AGILE and EUDIS activities. The application process should make sure that companies can submit a first-stage proposal in under 2 hours (5 pages maximum), supported by demonstrations of existing capabilities (e.g. video evidence). Applicants should be notified within 6–8 weeks whether they have been selected for the next stage. Small and innovative providers should be able to participate in procurement as prime contractors, consortium partners, or suppliers. Contracts should protect contractors’ existing IP while allowing governments to integrate and maintain the technology, and to procure future upgrades competitively.

First steps in the next 12 months: Responsible ministries in close cooperation with the military should identify priority capability needs and allocate funding for innovation competitions and paid trials. Procuring authorities should then launch the first competitions. The ministries responsible for procurement law should identify and remove any legal or regulatory barriers to this approach.

Success indicators: Share and value of suitable procurement run via innovation contests; share of applicants that are startups of first-time defence suppliers; median time from application to selection.

Prototype-to-production contracts – Universal for procuring Member States. Procurement contracts for innovative technology should include funded options to move successful prototypes into production. Where relevant this could be linked to industrial scale-up support under EDIP. The decision to move successful pilots to production should rest on the following criteria: performance in operational tests, lifetime costs, interoperability and compatibility with existing systems, and the supplier’s ability to scale production.

First steps in the next 12 months: Ministries of defence should bookmark a sufficient budget for successful prototypes to enter production. Procuring authorities should define performance and affordability thresholds before trials and include production options into initial contracts.

Success indicators: Share of prototypes entering production; median time from successful trial to production order.

Establish a European Defence Mechanism (EDM) – Coalition. European countries should establish a permanent mechanism for joint defence procurement, to make cooperation credible and create the scale for competitive European defence markets. This Mechanism should remain open to European countries outside the Union and operate on behalf of its members. Members must commit capital subscriptions to the institution, and jointly define capability needs and common functional requirements. Selected capabilities should then be procured jointly through the Mechanism. Before the Mechanism launches a tender, participating members should agree on functional requirements, expected order volumes, delivery schedules, as well as national budget commitments.

First steps in the next 12 months: At the Foreign Affairs Council meeting on 28 September 2026, interested defence ministers should test support for the Mechanism and identify an initial coalition. The participating governments should then appoint a joint task force of defence and finance ministry officials to prepare its mandate, governance, capital structure, and initial procurement areas. Within 12 months, they should establish a founding framework and identify the first capabilities to be procured jointly. The Mechanism should remain open to countries wishing to join later.

Success indicators: Value and share of binding joint orders; number of common configurations replacing national variants; unit costs and delivery time compared to similar, unilateral procurement procedures.

Common standards and shared testing facilities – Coalition. Member States should use the EDA’s Defence Test and Evaluation Base to involve military users, share testing facilities, and avoid duplicating equivalent tests across countries. They should make use of EDSTAR to align technical standards.

First steps in the next 12 months: Interested defence ministries together with the EDA should designate facilities for joint testing. National security authorities should identify which certifications and clearances can be mutually recognised.

Success indicators: Number of avoided duplicate tests; qualification time and cost for defence suppliers.

Implementation workstreams

Success indicators: Share of prototypes entering production; median time from successful trial to production order.

Success indicators: Number of avoided duplicate tests; qualification time and cost for defence suppliers.

Considerations

Requirements must reflect real capability needs, and not pre-determine specific technical solutions.

Joint purchasing should increase States’ buying power while increasing supplier competition by creating a better integrated European defence market.

IP terms should preserve innovation incentives and, at the same time, protect governments from supplier lock-in and supply disruption.

Considerations

Requirements must reflect real capability needs, and not pre-determine specific technical solutions.

Joint purchasing should increase States’ buying power while increasing supplier competition by creating a better integrated European defence market.

IP terms should preserve innovation incentives and, at the same time, protect governments from supplier lock-in and supply disruption.

O2.2

Develop indispensable assets across the AI value chain

O2.2

Develop indispensable assets across the AI value chain

Why it matters

Countries that own no part of the AI value chain risk being economically sidelined under transformative AI. Europe currently holds strong positions in only a few layers, and their importance could wane over time. Europe should hence spread its bets widely and combine areas of existing strength (e.g. industrial AI) with contestable niches (e.g. inference chips) and selected alternative-paradigm moonshots.

Why it matters

Recommendations at the Union level

01

Co-fund the world’s first maximum-security AI data centre

Very high

01

Co-fund the world’s first maximum-security AI data centre

Very high

02

Make Europe’s data spaces ready for sectoral AI

High

02

Make Europe’s data spaces ready for sectoral AI

High

01 Co-fund the world’s first maximum-security AI data centre

01

Co-fund the world’s first maximum-security AI data centre

Action

Establish a dedicated Union funding line that co-finances, together with one or a small group of host Member States, the world’s first AI data centre designed to meet RAND Security Level 5 or equivalently the SL5 Standard for AI Security by 2028.

Action

Implementation

Potential instruments: A dedicated maximum-security tier or follow-up call under the EuroHPC Joint Undertaking’s AI Gigafactories initiative (Council Regulation (EU) 2026/150), drawing on the InvestAI initiative and matched Member State funding.

Potential first steps (next 12 months):

October 2026: Commission and Member State security authorities, drawing on external expertise, begin drafting the AI facility specification.

November 2026: A Member State informally offers to finance and host the facility, potentially together with other States.

January–March 2027: Commission selects a host Member State (or coalition), which selects a site and hires a world-class team of data centre security specialists.

April 2027: Union funding decision taken; facility specification published; Member State team has an operational core and begins its work.

Success indicators: A signed EU–host Member State co-financing agreement, in line with applicable state aid rules; independently assessed maximum-security data centre operational by 2028.

Implementation

Potential first steps (next 12 months):

October 2026: Commission and Member State security authorities, drawing on external expertise, begin drafting the AI facility specification.

November 2026: A Member State informally offers to finance and host the facility, potentially together with other States.

January–March 2027: Commission selects a host Member State (or coalition), which selects a site and hires a world-class team of data centre security specialists.

April 2027: Union funding decision taken; facility specification published; Member State team has an operational core and begins its work.

Success indicators: A signed EU–host Member State co-financing agreement, in line with applicable state aid rules; independently assessed maximum-security data centre operational by 2028.

Considerations

A maximum-security data centre would be robust against top-priority attacks from the most capable, state-level actors, who could try to steal or sabotage AI models running on insufficiently secure infrastructure. Today, no AI data centre in the world meets such a security standard.

European maximum-security data centres would have three main use cases: (i) storing and serving frontier model weights, for example to enable European access to security-sensitive model weights under compute-for-access arrangements, (ii) highly sensitive public-sector inference workloads, for example in defence, and (iii) evaluations and safety research on potentially dangerous models.

Researchers estimate that a small, proof-of-concept, RAND SL5 facility for AI inference could be built within 14 months if it were a national priority, relying entirely on known technologies, and would cost $37 million to $50 million ($277 million to $345 million for an enterprise-scale version).

The security specifications for the relevant facility should be kept technically separate from sovereignty specifications, which address a different problem.

Considerations

The security specifications for the relevant facility should be kept technically separate from sovereignty specifications, which address a different problem.

02 Make Europe's data spaces ready for sectoral AI

02

Make Europe's data spaces ready for sectoral AI

Action

Clarify legal status and provide technical infrastructure so that companies can easily share their data for the purposes of developing domain-specific AI models and applications in sectors such as manufacturing, healthcare, or automotive. This should build on the Common European Data Spaces and Data Labs and include as specific measures: (i) ensure that the Commission guidance on data pooling committed under the Data Union Strategy addresses multi-party data pooling between competitors for AI development, explaining how it can be compatible with Article 101 TFEU, building on Chapter 6 of the 2023 Horizontal Guidelines, which addresses data pooling in general terms but not the specific case of pooling AI training data between competitors, (ii) publish model contractual terms for multi-party data sharing for AI development, building on the Data Act model contractual terms published by the Commission in draft form in November 2025, (iii) direct at least €50 million of EU research funding to privacy-enhancing technologies (PETs) to be deployed in the common European data spaces and the AI Factories’ data labs.

Action

Implementation

Potential instruments: Commission soft law for the competition guidance mentioned in measure (i) above; a Commission Recommendation extending the Data Act Article 41 model contractual terms; existing PETs topics in the Horizon Europe Work Programme 2026–27 and, for additional funding, the first work programme of its successor.

First steps (next 12 months):

January 2027: Commission publishes model contracts and the guidance on data pooling, which includes AI-specific examples.

June 2027: Commission earmarks at least €50 million for PETs across current and upcoming research work programmes.

September 2027: At least three Common European Data Spaces have operational data pools for multi-party AI training operating under the model terms.

Success indicators: Number of multi-party data pools for AI training operating under the published model terms; amount of EU funding committed to PETs; number and diversity of companies contributing proprietary data; number of AI models or applications developed using the data spaces.

Implementation

First steps (next 12 months):

January 2027: Commission publishes model contracts and the guidance on data pooling, which includes AI-specific examples.

June 2027: Commission earmarks at least €50 million for PETs across current and upcoming research work programmes.

September 2027: At least three Common European Data Spaces have operational data pools for multi-party AI training operating under the model terms.

Considerations

Starting in 2020, the EU has built Common European Data Spaces across 14 sectors, which aim to facilitate data sharing through governance frameworks and technical infrastructure. However, as the 2025 Data Union Strategy recognises, the challenge is to move beyond pilot projects and make data spaces useful for AI training at scale. The primary bottleneck is companies being reluctant to contribute commercially valuable data to multi-party data spaces: They need legal certainty that pooling data with competitors is compatible with competition law, model contracts governing access rights and how the value created from pooled data is distributed, and mature technical tools that let models learn from sensitive data without exposing it to untrusted parties. The measures above target these demand-side barriers by lowering entry costs for data holders and using sectoral model training as a concrete use case around which to organise.

The Commission has already committed in its Data Union Strategy to issue guidance on data pooling. The guidance should use concrete examples to address, among other things, (i) which data competitors can pool for AI training and when PETs are needed to shield commercially sensitive data, (ii) how contribution requirements and access fees can prevent free-riding, and (iii) how companies can use the Informal Guidance Notice to obtain timely comfort for cross-border AI data pooling initiatives.

Data pools should be structured so that most of the value stays with European companies, while still complying with Article 101 TFEU. Under the 2023 Horizontal Guidelines, access rules need to be fair, objective, transparent, and non-discriminatory. Therefore, simply excluding non-EU developers raises competition concerns. Instead, access could be tied to reciprocal contributions to the pool or subject to fair access fees.

Considerations

Recommendations at the national level

01

Direct ARPA-style vehicles toward strategic AI bets

Very high

01

Direct ARPA-style vehicles toward strategic AI bets

Very high

02

Host and co-finance the first maximum-security AI data centre

Very high

02

Host and co-finance the first maximum-security AI data centre

Very high

03

Anchor demand for European inference chips

High

03

Anchor demand for European inference chips

High

01 Direct ARPA-style vehicles toward strategic AI bets

01

Direct ARPA-style vehicles toward strategic AI bets

Action

Resource national ARPA-style agencies to fund a strategic AI portfolio, focusing on some or all of the following categories: (1) sectoral AI champions in areas of existing strength, especially where valuable proprietary data can be leveraged (e.g. manufacturing, healthcare, automotive), (2) specialised AI hardware, such as AI inference chips, and the corresponding orchestration layer, (3) security-relevant hardware infrastructure, such as maximum-security data centres or hardware-enabled mechanisms for chip verification, (4) selected high-risk/high-return approaches in AI research (e.g. world models, safe-by-design AI, worker-friendly AI).

Action

Member State mode

Universal – every Member State should do this, using an existing ARPA-style vehicle (e.g. Germany’s SPRIND) or a newly created one (as recommended in Objective 2.1). Some Member States, especially smaller ones, may want to orient their portfolio around strategic bets that are most promising in light of their economic position, resources, and areas of existing strength, e.g. semiconductor R&D in Belgium, AI-relevant connectivity in Finland, AI applications in healthcare in Denmark, public sector AI applications in Estonia, or photonics in Lithuania.

Member State mode

Implementation

Potential instruments: Administrative direction plus a dedicated budget line, in accordance with State-aid rules.

Potential first steps (next 12 months):

October 2026: Member States direct their ARPA-style vehicles to define a strategic AI portfolio.

December 2026: Categories have been defined; Member States commit the budget through reallocation or a new funding line.

January 2027: First calls published in one or two priority categories per country.

April 2027: First funding contracts signed.

Success indicators: € committed to the strategic AI portfolio, total value of funding contracts signed, follow-on private capital raised by funded companies within 24 months of the ARPA award.

Implementation

Potential instruments: Administrative direction plus a dedicated budget line, in accordance with State-aid rules.

Potential first steps (next 12 months):

October 2026: Member States direct their ARPA-style vehicles to define a strategic AI portfolio.

December 2026: Categories have been defined; Member States commit the budget through reallocation or a new funding line.

January 2027: First calls published in one or two priority categories per country.

April 2027: First funding contracts signed.

Success indicators: € committed to the strategic AI portfolio, total value of funding contracts signed, follow-on private capital raised by funded companies within 24 months of the ARPA award.

Considerations

High-risk/high-return bets in AI research (the fourth category above) should be seen as complements rather than substitutes to securing access to AI models at the current frontier. These models have been the main driver of AI progress for years, with no indication of a slowdown in capability growth, and access to them will be crucial for protecting against imminent security risks (e.g. cyberattacks).

Member States, especially smaller ones, may want to coordinate informally to avoid duplicating efforts.

For specialised AI hardware, national funding should complement (and where appropriate, co-finance) the Union-level instruments under Pillar I of the EU’s Chips Act, which funds research infrastructure and early-stage support for chip startups and scale-ups. However, the Commission's own evaluation found that its startup fund was exhausted within two years and that Europe lacks the late-stage capital to scale these companies. National ARPA-style funding can help to fill this gap alongside private capital (in line with the recommendations for better private capital mobilisation in Objective 2.1).

Considerations

Member States, especially smaller ones, may want to coordinate informally to avoid duplicating efforts.

For these bets to be successful, the Union and Member States will need to implement the broader competitiveness measures outlined in Objective 2.1. The recommendations above will only bear fruit in an economy that makes it easy to found and scale high-growth companies, for example, by giving companies access to growth capital and enabling them to hire and retain the world’s best talent. Creating these conditions is the single most powerful contribution that the Union and its Member States can make to Europe’s position in the AI value chain.

02 Host and co-finance the first maximum-security AI data centre

02

Host and co-finance the first maximum-security AI data centre

Action

Build the world’s first maximum-security AI data centre by 2028 (see Union-level recommendation 1), either alone or within a Member State coalition. This requires (i) a site with secured access to energy (e.g. 10+ MW with the possibility to scale) and exceptional protection against natural or adversarial threats, (ii) a national co-financing share for building the facility and attracting the world-class talent to make it highly secure, and (iii) cooperation with national intelligence and security agencies (e.g. personnel vetting, counter-intelligence, physical protection).

Action

Member State mode

Coalition or kind-specific – either one Member State or a nimble coalition willing to pool funding, talent, suitable sites, and security/intelligence capability.

Member State mode

Coalition or kind-specific – either one Member State or a nimble coalition willing to pool funding, talent, suitable sites, and security/intelligence capability.

Implementation

Potential instruments: Application to a to-be-established maximum-security tier of the AI Gigafactories initiative (see Union-level recommendation 1 above); national budget commitment; mandate to national intelligence and security agencies; procurement commitments for high-sensitivity workloads.

Potential first steps (next 12 months):

November 2026: Member State government informally tells the Commission that they are willing to finance and host the facility, potentially together with coalition members.

January 2027: A host Member State is selected by the Commission; potentially together with coalition members, it commits a dedicated share of its national budget to the EU co-financing line and begins hiring a world-class team of data centre security specialists.

April 2027: Union funding decision taken; facility specification published; Member State team has an operational core and begins its work.

July 2027: Member States(s) select a suitable site that meets the relevant security requirements; intelligence and security agencies begin to prepare it.

Success indicators: A signed EU–host Member State co-financing agreement, in line with applicable State-aid rules; independently assessed maximum-security data centre operational by 2028.

Implementation

Potential first steps (next 12 months):

November 2026: Member State government informally tells the Commission that they are willing to finance and host the facility, potentially together with coalition members.

April 2027: Union funding decision taken; facility specification published; Member State team has an operational core and begins its work.

July 2027: Member States(s) select a suitable site that meets the relevant security requirements; intelligence and security agencies begin to prepare it.

Success indicators: A signed EU–host Member State co-financing agreement, in line with applicable State-aid rules; independently assessed maximum-security data centre operational by 2028.

Considerations

European maximum-security data centres would have three main use cases: (i) storing and serving frontier model weights, for example, to enable European access to security-sensitive model weights under compute-for-access arrangements, (ii) highly sensitive public-sector inference workloads, for example in defence, and (iii) evaluations and safety research on potentially dangerous models.

The security specifications for the relevant facility should be kept technically separate from sovereignty specifications, which address a different problem.

Considerations

The security specifications for the relevant facility should be kept technically separate from sovereignty specifications, which address a different problem.

03 Anchor demand for European inference chips

03

Anchor demand for European inference chips

Action

Create conditional purchase commitments for inference chips designed in the EU from publicly supported AI data centres, such as the EU’s AI Gigafactories, once those chips meet pre-specified criteria for use in AI data centres. Member States should target aggregate commitments of 10% of the public funding for the AI Gigafactories (i.e., €1 billion) by 2029.

Action

Member State mode

Kind-specific – Member States planning to build significant amounts of public AI compute, such as the host countries of the AI Gigafactories.

Member State mode

Kind-specific – Member States planning to build significant amounts of public AI compute, such as the host countries of the AI Gigafactories.

Implementation

Potential instruments: Use existing joint and innovation-procurement tools and, once adopted, the demand-side measures in Chips Act 2.0 to coordinate requirements and purchases across Member States.

Potential first steps (next 12 months):

December 2026: Interested Member States form a purchasing coalition and agree an indicative collective budget for conditional purchase commitments.

March 2027: The coalition consults European chip companies and agrees common requirements, including performance, security, and energy efficiency.

September 2027: Participating purchasing bodies publish coordinated conditional purchase commitments for European inference chips.

Success indicators: € value of published advance purchase commitments; share of public inference procurement accounted for by European suppliers of inference chips.

Implementation

Potential first steps (next 12 months):

December 2026: Interested Member States form a purchasing coalition and agree an indicative collective budget for conditional purchase commitments.

March 2027: The coalition consults European chip companies and agrees common requirements, including performance, security, and energy efficiency.

September 2027: Participating purchasing bodies publish coordinated conditional purchase commitments for European inference chips.

Success indicators: € value of published advance purchase commitments; share of public inference procurement accounted for by European suppliers of inference chips.

Considerations

The proposed Chips Act 2.0 would provide useful tools for coordinating semiconductor procurement, but contains no quantified commitment to purchase European AI inference chips. The €1 billion target would be an additional commitment by participating Member States.

This would create an early market for European suppliers without slowing the buildout of AI compute. Until suitable European inference chips are available, buyers should procure on the open market. This is because the highest priority for Europe is to host AI compute in proportion to its global GDP share as fast as possible (see Immediate Objective 3). Member States should procure domestic inference chips only insofar as this is consistent with that goal.

National funding and procurement commitments should be designed in accordance with applicable EU State-aid and public-procurement rules.

Considerations

National funding and procurement commitments should be designed in accordance with applicable EU State-aid and public-procurement rules.

Pillar 3: Ensuring safety and security

O3.1

Ensure prioritised and targeted use of EU rules to address risks from highly capable AI

O3.1

Ensure prioritised and targeted use of EU rules to address risks from highly capable AI

Obligations on providers of general-purpose AI models are already applicable and enforceable, and the GPAI Code of Practice is already being relied upon. For this reason, there are no detailed recommendations for this objective.

O3.2

Prepare for labour market impacts

O3.2

Prepare for labour market impacts

Why it matters

Transformative AI would likely displace human labour at unprecedented levels. Managing labour market transitions well will be necessary to sustain people’s livelihoods, while unmitigated job losses threaten social stability.

Why it matters

Recommendations at the Union level

01

Build a monitoring stack for AI labour market impacts

Very high

01

Build a monitoring stack for AI labour market impacts

Very high

02

Fund and steer conditional flexicurity adjustment

Very high

02

Fund and steer conditional flexicurity adjustment

Very high

01 Build a monitoring stack for AI labour market impacts

01

Build a monitoring stack for AI labour market impacts

Action

Prepare for AI labour-market impacts in the EU by September 2027 with these measures:

Action

Prepare for AI labour-market impacts in the EU by September 2027 with these measures:

(i) introduce an AI-displacement module in the EU Labour Force Survey (LFS), with the module in the field no later than the 2029 survey year.

(ii) create a quarterly Cedefop bulletin tracking hiring patterns in vacancies in AI-exposed occupations. This should leverage Cedefop's Skills-OVATE (Skills Online Vacancy Analysis Tool for Europe) online job advertisement analytics platform.

(iii) adjust JRC’s occupational AI exposure index to the four-digit level of the International Standard Classification of Occupations (ISCO-08) in collaboration with the International Labour Organization (ILO). Ensure the mapping to the principal international exposure measures (Anthropic's Economic Index, the OpenAI/University of Pennsylvania exposure measure, and the ILO's global index of occupational exposure).

(iv) negotiate with AI developers to access usage data.

(v) improve occupational coding for quarterly occupation-level firm-side statistics on wages and employment.

(i) introduce an AI-displacement module in the EU Labour Force Survey (LFS), with the module in the field no later than the 2029 survey year.

(iv) negotiate with AI developers to access usage data.

(v) improve occupational coding for quarterly occupation-level firm-side statistics on wages and employment.

Implementation

Potential instruments:

LFS module: The Commission should implement delegated acts following Article 3(7) and under Regulation (EU) 2019/1700 (European social statistics framework), amending Commission Delegated Regulation (EU) 2025/1999 (multiannual rolling planning of survey topics), supported by an implementing act for technical specifications. Article 4 allows for a change in the schedule, which the Commission should prioritise. Alternatively, it could serve as the already scheduled 2030 topic.

LMB module: Article 8 of Regulation (EU) 2025/941 permits a feasibility study. This study should aim to make Member States’ administrative payroll sources comparable, particularly in terms of coverage and occupational coding. Subject to coverage of at least 10 Member States with more than 50% coverage, the Commission should, by an Article 4(3) delegated act, receive high-frequency data on earnings, labour costs, vacancies, and, if feasible, employment. The UK demonstrated how to provide high-frequency data with their monthly PAYE Real Time Information (RTI).

Usage data: The Commission should seek access to AI usage data through a Memorandum of Understanding (MoU) with leading AI companies (see considerations). In parallel, the Commission (Eurostat) should, in accordance with Article 17b of Regulation (EC) No 223/2009 (as amended by Regulation (EU) 2024/3018), request access to data from the appropriate private data holders. In cases where an agreement with a private data holder is not reached within the timelines set by the legislation, the Commission may use its powers to compel access to the relevant data in accordance with Article 17c of Regulation (EC) No 223/2009.

Potential first steps over the next 12 months include:

Q4 2026: Eurostat develops an AI-displacement ad hoc subject, and the ESS Committee needs to be consulted regarding planned statistical measures (Article 7 Reg 223/2009). The Commission adopts a Delegated Act amending Regulation (EU) 2025/1999 to change the content of the 2029 survey.

Q4 2026: Eurostat also coordinates with Member State NSIs to immediately add the AI displacement subject at the national level.

Spring 2027: JRC and Cedefop publish their annual exposure index based on the ISCO-08 4-digit level and quarterly exposure bulletin, respectively.

Success indicators: The EU adopts the delegated acts and JRC and Cedefop update their statistics. The EU receives usage data for analysis by 2027.

Implementation

Potential instruments:

Potential first steps over the next 12 months include:

Q4 2026: Eurostat also coordinates with Member State NSIs to immediately add the AI displacement subject at the national level.

Spring 2027: JRC and Cedefop publish their annual exposure index based on the ISCO-08 4-digit level and quarterly exposure bulletin, respectively.

Success indicators: The EU adopts the delegated acts and JRC and Cedefop update their statistics. The EU receives usage data for analysis by 2027.

Considerations

The UK AI Economics Institute’s joint statement with Anthropic, Google, OpenAI and Microsoft (who later became members of a working group in June 2026) is an early precedent for formal lab–government cooperation. However, this institution is not yet fully operational, and its data access remains unproven.

Considerations

If EU data-protection rules constrain the collection of API usage data, as may have been the case with Google’s ATLAS, regulators need to reduce legal uncertainty in analysing API data.

While Eurostat already surveys enterprise AI adoption, it has not yet collected data on AI usage.

The Commission should consider creating an EU AI Economics Institute to assess the economic impact of AI continuously. Either the JRC or the AI Office should host this research unit due to their research focus and network with AI companies, respectively.

While Eurostat already surveys enterprise AI adoption, it has not yet collected data on AI usage.

02 Fund and steer conditional flexicurity adjustment

02

Fund and steer conditional flexicurity adjustment

Action

The Commission should fund social programmes and incentivise flexibility reforms for Member States via the European Semester guidance.

Action

The Commission should fund social programmes and incentivise flexibility reforms for Member States via the European Semester guidance.

Commit funding from European Social Fund Plus (ESF+) to cushion workers from AI displacement by redeployment, employer-embedded retraining, and providing income bridges. Spending €2–2.3 billion on retraining for AI-displacement from the €71.2 billion social earmark for the 2028–2034 period would be an order of magnitude above the European Globalisation Adjustment Fund for Displaced Workers (EGF)’s €35 million/year. This could finance roughly 130,000–150,000 participants at around €15,000 per head.

Give Member States country-specific recommendations on high-earner flexibility now, with broader flexicurity reform explicitly conditional on the development of AI’s impact.

Complete, by September 2027, contingency design work (not a legislative proposal) for an EU-backed short-time work reinsurance for a fast-displacement scenario, modelled on the pandemic-era Support to mitigate Employment Risks in an Emergency (SURE).

Give Member States country-specific recommendations on high-earner flexibility now, with broader flexicurity reform explicitly conditional on the development of AI’s impact.

Implementation

Potential instruments:

European Semester: Deliver employment analysis and country-specific recommendations (CSRs) to Member States.

SURE could serve as precedent for a large-scale social programme by the EU (Council Regulation (EU) 2020/672 (Article 122 TFEU)). It was a ~€100 billion instrument that supported ~31.5 million people.

Potential first steps (next 12 months):

October 2026: The Commission Work Programme 2027 should carry a non-legislative initiative on AI labour-market retraining.

Q1–Q2 2027: The Commission should encourage Member States, during the negotiation of the National Regional Partnership Plans for 2028–34, to retain part of the 14% social earmark for retraining. Those should be paid out if monitoring indicators flag AI’s impact.

By September 2027: DGECFIN/EMPL prepare plans for a SURE-style displacement facility, including trigger design, financing route (Article 122 precedent), and interaction with national short-time work schemes.

November 2027: The Joint Employment Report draws on JRC and Cedefop series from Union level recommendation 1.

Success indicators: The Commission commits €2 billion of ESF+ to AI-displacement adjustment measures over 2028–34; AI-displacement section present in ≥1 full Semester cycle (Autumn Package through CSRs); funding for AI automation scenarios retained in ≥5 adopted NRPPs.

Implementation

Potential instruments:

European Semester: Deliver employment analysis and country-specific recommendations (CSRs) to Member States.

Potential first steps (next 12 months):

October 2026: The Commission Work Programme 2027 should carry a non-legislative initiative on AI labour-market retraining.

November 2027: The Joint Employment Report draws on JRC and Cedefop series from Union level recommendation 1.

Considerations

The reform should start with high earners, since the benefits are large and the political cost is low, while a broader flexicurity depends on monitoring results (see Union level recommendation 1).

Retraining measures require careful design to be effective: the average US retraining programme raises employment by roughly 1.7 percentage points and earnings by $800/year per person offered the programme. Employer-embedded sector programmes deliver the largest gains, but they resist rapid scaling. Programme funding should be tied to sector-programme design features, fast-growing sectors with talent gaps, and a built-in evaluation mandate.

Displacement adjustment should depend on aggregate monitoring indicators (see Union level recommendation 1) rather than on case-specific mobilisation or lengthy individual eligibility tests, to avoid failures of the EGF and the US Trade Adjustment Assistance mechanism.

The EU should explore joint borrowing measures to provide security in a fast-automation scenario.

Considerations

The reform should start with high earners, since the benefits are large and the political cost is low, while a broader flexicurity depends on monitoring results (see Union level recommendation 1).

The EU should explore joint borrowing measures to provide security in a fast-automation scenario.

Recommendations at the national level

01

Feed into the AI labour market monitoring stack

Very high

01

Feed into the AI labour market monitoring stack

Very high

02

Prepare high-earner-first flexicurity reforms

Very high

02

Prepare high-earner-first flexicurity reforms

Very high

01 Feed into the AI labour market monitoring stack

01

Feed into the AI labour market monitoring stack

Action

National statistical institutes (NSIs) add AI-displacement questions to national labour-force surveys for the 2027 or 2028 survey wave, ahead of the common Eurostat module (Union level recommendation 1).

Action

Each Member State can add national questions to its labour force survey under Regulation (EU) 2019/1700.

Implementation

Potential instruments: Professional independence of national statistical institutes (NSIs) in developing statistics is legally protected (Article 2, Reg. 223/2009), and Reg. 2019/1700 mandates only some harmonised questions. Consequently, Member States can add questions.

Potential first steps (next 12 months):

2027: NSIs add AI as a topic in their labour force surveys.

Success indicators: 10 Member States rollout surveys by 2027 successfully. Eurostat compares this data across countries.

Implementation

Potential first steps (next 12 months):

2027: NSIs add AI as a topic in their labour force surveys.

Success indicators: 10 Member States rollout surveys by 2027 successfully. Eurostat compares this data across countries.

02 Prepare high-earner-first flexicurity reforms

02

Prepare high-earner-first flexicurity reforms

Action

Legislate reduced employment protection for high earners, measured against a national salary threshold set by Member State governments.

Member States should lighten their dismissal rules for some professionals. They should pair these measures with stronger income support and active labour market spending, following for example the Danish model.

Deliver retraining and benefit payments to displaced workers in weeks.

Action

Legislate reduced employment protection for high earners, measured against a national salary threshold set by Member State governments.

Deliver retraining and benefit payments to displaced workers in weeks.

Member State mode

Employment and dismissal rules are mainly national competences. Member States therefore retain the principal responsibility for these reforms. A possible model is Germany’s proposed relaxation of dismissal rules.

Member State mode

Implementation

Potential instruments: Employment legislation such as the easing of employment protection for high earners should be balanced by active labour market policies to provide support for workers amid faster employment transitions. Portable, pre-funded severance arrangements on the Austrian model could make shorter notice periods more acceptable.

Potential first steps (next 12 months):

2026–27 legislative sessions: Pass flexibility measures for the most dynamic and high-earning professionals.

Q1–Q2 2027: Member States prepare flexibility reforms for future implementation.

Success indicators: 15 Member States enact the high-earner flexibility measures across Member States by late 2027; publication of contingency packages; shortening the length of time before payments arrive in cases of displacement.

Implementation

Potential first steps (next 12 months):

2026–27 legislative sessions: Pass flexibility measures for the most dynamic and high-earning professionals.

Q1–Q2 2027: Member States prepare flexibility reforms for future implementation.

Considerations

The measures should be informed by the failures of the EGF: Case-by-case mobilisation was too slow and small relative to the shocks it aimed to address.

This measure aims to reduce the costs of experimentation, where trying and failing fast is important. Workers employed by dynamic firms, for whom labour market mobility tends to be most important are likely to be less vulnerable to firings. Targeted flexibility measures also reduce political opposition.

Labour market flexibilisation for the broader workforce should be paired with income support and active labour market spending to deliver effective sector retraining programmes.

Considerations

The measures should be informed by the failures of the EGF: Case-by-case mobilisation was too slow and small relative to the shocks it aimed to address.

Labour market flexibilisation for the broader workforce should be paired with income support and active labour market spending to deliver effective sector retraining programmes.

Next chapter

What would be required

Part A: Strategic Priorities

Immediate Objectives

IO-1

Create a member state alliance for supply chain security

IO-2

Make Europe’s institutions ready to act in a transformative AI world

IO-3

Secure Europe's share of global AI compute

IO-4

Ensure resilience to AI crises

IO-5

Make Europe the global leader in AI assurance technology

Additional Objectives

Pillar 1

Securing access to frontier AI

O1.1

Secure ongoing access to frontier AI systems

O1.2

Protect European assets

Pillar 2

Building economic strength

O2.1

Make Europe the best place to build and scale high-growth companies

O2.2

Develop indispensable assets across the AI value chain

Pillar 3

Ensuring safety and security

O3.1

Ensure prioritised and targeted use of EU rules to address risks from highly-capable AI

O3.2

Prepare for labour market impacts

What would be required for a successful European frontier AI project?

Deep dives

Compute deep dive

Robotics deep dive

Part A: Strategic Priorities

Immediate Objectives

IO-1

Build a Member State Alliance for Supply Chain Security

IO-2

Make Europe’s institutions ready to act in a transformative AI world

IO-3

Secure Europe's share of global AI compute

IO-4

Ensure resilience to AI crises

IO-5

Make Europe the global leader in assurance technology

Additional Objectives

Pillar 1

Securing access to frontier AI

O1.1

Secure ongoing access to frontier AI systems

O1.2

Protect European assets

Pillar 2

Building economic strength

O2.1

Make Europe the best place to build and scale high-growth companies

O2.2

Develop indispensable assets across the AI value chain

Pillar 3

Ensuring safety and security

O3.1

Ensure prioritised and targeted use of EU rules to address risks from highly-capable AI

O3.2

Prepare for labour market impacts

What would be required for a successful European frontier AI project?

Deep dives

Compute deep dive

Robotics deep dive