Algorithmic Industrial Policy Systems and Selective Market Distortion

1. Introduction

Algorithmic industrial policy systems refer to governmental or quasi-governmental systems in which algorithms, automated scoring models, data platforms, AI systems, or computational decision tools are used to determine which firms, sectors, technologies, regions, projects, or products receive preferential regulatory or economic treatment.

Examples include algorithms used for:

• allocation of industrial subsidies;
• tax incentives and credits;
• government procurement;
• research and development grants;
• concessional financing;
• industrial licences and permits;
• access to public infrastructure;
• export incentives;
• technology-development programmes;
• energy and carbon allowances;
• strategic-sector investment;
• government-backed venture financing.

The competition-law problem arises when an apparently neutral algorithm selectively changes competitive conditions. A system may favour particular undertakings or technologies through its eligibility criteria, training data, scoring methodology, risk thresholds, procurement rankings, or automated allocation rules.

EU State-aid law expressly recognises that selective governmental advantages can distort competition. Article 107(1) TFEU generally concerns measures that confer an advantage selectively on particular undertakings or production and are capable of distorting competition and affecting trade.

The important point is that algorithmic administration does not eliminate legal responsibility merely because the discriminatory or preferential outcome is produced computationally.

2. Meaning of Algorithmic Industrial Policy

An algorithmic industrial policy system can be represented as:

Government policy objective → data collection → algorithmic eligibility criteria → scoring/ranking → allocation of benefits → market response → feedback data → algorithmic adjustment

For example:

Government wants to accelerate domestic battery production → AI system scores applicants according to investment size, domestic sourcing, technological capability and projected employment → highest-scoring firms receive grants and concessional electricity → supported firms expand production → competitors lose investment opportunities → new market data is fed into the system.

The algorithm therefore becomes more than an administrative instrument. It can become a competitive-selection mechanism.

3. Why Selectivity Matters

The central competition-law concern is selective advantage.

A government measure may appear generally available but operate selectively in practice.

For example:

EU guidance recognises that a measure apparently applicable to undertakings generally can nevertheless be selective in its actual operation.

4. Algorithmic Selectivity

Algorithmic selectivity can arise at several levels.

A. Input selectivity

The government chooses which data are used.

For example, an industrial-support algorithm may give substantial weight to:

• existing production capacity;
• historical tax contributions;
• previous government contracts;
• domestic ownership;
• employment numbers.

Large incumbent firms may consequently receive higher scores than innovative entrants.

B. Criterion selectivity

The algorithm may reward particular characteristics.

Example:

Subsidies are available to all battery manufacturers, but eligibility requires production above a particular scale.

The rule may therefore favour established manufacturers.

C. Weighting selectivity

Even neutral criteria can generate preferential outcomes if particular variables receive disproportionate weight.

D. Threshold selectivity

A small difference around an algorithmic threshold may determine whether a company receives millions in government support.

E. Feedback selectivity

This is particularly important.

If subsidised firms subsequently generate better performance data, the algorithm may interpret their performance as evidence that they deserve more future support.

This creates:

initial advantage → improved performance → higher algorithmic score → additional support → further advantage

The result can be a self-reinforcing industrial-policy loop.

5. Selective Market Distortion

The economic distortion may occur through several channels.

5.1 Entry distortion

A subsidised incumbent can make market entry more difficult for new firms.

5.2 Investment distortion

Investors may redirect capital toward firms receiving algorithmically determined government support.

5.3 Innovation distortion

The system may favour established technological pathways rather than competing technologies.

5.4 Output distortion

Supported firms may produce more than would otherwise be commercially justified.

5.5 Price distortion

Government support can allow beneficiaries to price below competitors without necessarily possessing greater productive efficiency.

5.6 Geographic distortion

Algorithms can concentrate industrial investment in selected regions.

5.7 Vertical distortion

Preferential support at one level of the supply chain can disadvantage firms operating at another level.

The European Commission's R&D&I State-aid framework expressly recognises risks involving distorted entry and exit, dynamic investment incentives, market power and location decisions.

6. Algorithmic Industrial Policy and Competition Law

The legal analysis should distinguish between legitimate industrial policy and competition-distorting implementation.

Industrial policy itself is not automatically unlawful.

Governments may legitimately pursue:

• national infrastructure;
• energy security;
• technological development;
• environmental objectives;
• employment;
• strategic resilience;
• regional development;
• R&D.

The competition question is instead:

Does the governmental intervention confer a selective advantage that materially changes competitive conditions, and is that intervention legally justified?

Under EU State-aid law, the principal analytical elements include:

1. State intervention or State resources;
2. economic advantage;
3. selectivity;
4. distortion or potential distortion of competition;
5. effect on trade.

7. Six Important Case Laws

1. Commission v Italy — Case 173/73

Principle

The Court of Justice established an important foundation for identifying State intervention that affects competitive conditions.

The case concerned financial advantages granted through public mechanisms. The Court adopted a substantive approach rather than allowing the legal form of the governmental intervention to determine whether State-aid rules were engaged.

Relevance to algorithmic industrial policy

An algorithm cannot immunise a governmental benefit merely because the benefit is distributed through:

• automated scoring;
• digital platforms;
• AI recommendations; or
• computational allocation.

The legal analysis remains focused on the economic substance of the intervention.

2. Adria-Wien Pipeline GmbH v Finanzlandesdirektion für Kärnten — Case C-143/99

Principle

The Court examined the selectivity of an Austrian environmental tax exemption.

A measure may be selective where it differentiates between undertakings that are otherwise comparable in light of the objective of the relevant system.

Relevance

This is highly relevant to algorithmic industrial-policy systems.

Suppose an AI-based industrial tax system provides benefits only to firms satisfying particular computationally determined criteria.

The analysis cannot stop at:

"The algorithm applies the same formula to everyone."

The more important question is:

Does the formula place otherwise comparable undertakings in different competitive positions without sufficient justification?

3. Paint Graphos — Joined Cases C-78/08 to C-80/08

Principle

The Court considered whether a tax treatment associated with cooperative societies constituted selective State aid.

The case illustrates the importance of identifying the reference framework and determining whether a measure constitutes a derogation from that framework.

Algorithmic relevance

An algorithm can hide selectivity within a supposedly general regulatory framework.

For example:

General industrial tax system → algorithmic exception for "strategic enterprises" → preferential rate.

The legal analysis must identify:

1. the normal taxation framework;
2. the beneficiaries;
3. the difference in treatment;
4. the justification for the difference.

4. Commission v Netherlands — C-279/08 P

Principle

The case concerned an environmental scheme involving emission-related advantages and the concept of selectivity.

The Court examined whether the structure of the measure conferred an advantage on particular undertakings.

Algorithmic relevance

Modern industrial-policy algorithms may distribute:

• carbon allowances;
• green-transition subsidies;
• renewable-energy incentives;
• emissions credits.

A system described as an environmental programme may nevertheless create selective advantages.

The environmental objective does not automatically eliminate competition-law scrutiny.

5. Ryanair v Commission — C-441/21 P

Principle

The Court's 2024 judgment concerned Spain's recapitalisation scheme for strategically important undertakings during the COVID-19 period.

The Court reaffirmed that selectivity requires examination of whether an economic advantage specifically benefits certain undertakings and places them in a more favourable position than comparable undertakings. It also recognised that Treaty derogations can permit compatible State aid where the relevant conditions are satisfied.

Algorithmic relevance

This is particularly significant for strategic industrial policy.

An algorithm may classify companies as:

• strategically important;
• systemically important;
• economically essential;
• nationally significant.

That classification can create a selective benefit.

The legal question then becomes whether the differentiation is supported by an applicable legal justification and whether the intervention satisfies the relevant necessity and proportionality requirements.

6. Eventech Ltd v Parking Adjudicator — C-518/13

Principle

The case concerned preferential access to London's bus lanes for licensed taxis compared with private-hire vehicles.

The Court examined whether the preferential regulatory treatment could constitute an economic advantage capable of distorting competition.

Algorithmic relevance

This demonstrates that selective advantage does not necessarily have to take the form of a direct cash subsidy.

An algorithmic industrial policy can provide preferential:

• infrastructure access;
• licensing;
• procurement;
• regulatory treatment;
• network access;
• priority allocation.

Thus, non-financial algorithmic preferences can also affect competition.

8. Additional Important Case: Deutsche Post — C-399/08 P

The Deutsche Post litigation illustrates the importance of examining whether public compensation provides an economic advantage exceeding what is necessary to compensate for public-service obligations.

Algorithmic relevance

An automated public-service compensation system could systematically overcompensate selected firms if its model:

• overestimates costs;
• underestimates revenues;
• uses asymmetric datasets;
• gives incumbent firms preferential benchmarks.

Therefore, algorithmic calculation does not itself establish neutrality.

9. The Algorithmic Feedback-Loop Problem

One of the most important issues is path dependence.

Consider:

Stage 1

Government algorithm identifies Firm A as a strategic enterprise.

Stage 2

Firm A receives subsidised financing.

Stage 3

Firm A expands production.

Stage 4

The algorithm observes Firm A's increased production.

Stage 5

The algorithm interprets increased production as evidence of competitiveness.

Stage 6

Firm A receives additional support.

This creates:

Government selection → subsidy → growth → improved algorithmic score → additional subsidy

The initial policy choice therefore becomes increasingly difficult to reverse.

This problem has particular economic significance because empirical research on Chinese shipbuilding industrial policy found that large-scale policy support substantially increased domestic investment, entry and market share but also produced fragmentation and idle capacity.

10. Algorithmic Industrial Policy in China

China provides an important example of the broader industrial-policy problem because industrial policy has historically involved substantial government support for strategic sectors.

Potential competition concerns include:

• subsidies;
• preferential financing;
• land allocation;
• energy pricing;
• government procurement;
• tax incentives;
• local-government industrial funds;
• technology programmes;
• state-owned enterprise advantages;
• strategic-sector designation.

The competition analysis becomes especially complex when the algorithm or administrative system uses industrial-policy priorities to determine beneficiaries.

For example:

"Strategic emerging industry" classification → preferential credit → increased capacity → lower prices → competitors exit → increased concentration.

The resulting market structure may not reflect purely competitive selection.

11. Algorithmic Distortion of the Competitive Process

A useful distinction is between consumer-price distortion and competitive-process distortion.

Consumer-price distortion

Government intervention causes prices to become artificially low or high.

Competitive-process distortion

Government intervention changes:

• who enters;
• who expands;
• who obtains financing;
• who receives infrastructure;
• who obtains government contracts;
• which technology becomes dominant.

The second form can be more difficult to detect because the immediate consumer price may appear beneficial.

12. Competition Concerns Created by Algorithmic Industrial Policy

A. Incumbency bias

Historical data may systematically favour established firms.

B. Data advantage

Beneficiary firms may generate data that subsequently improves their algorithmic ranking.

C. Lack of transparency

Competitors may not understand why one undertaking receives support.

D. Error propagation

An incorrect classification can be reproduced through subsequent algorithmic decisions.

E. Strategic gaming

Companies may modify their behaviour to maximise algorithmic scores rather than economic efficiency.

F. Regulatory capture

Companies may attempt to influence:

• data definitions;
• eligibility variables;
• weighting;
• thresholds;
• scoring methodologies.

G. Reduced technological diversity

Government algorithms may favour one technological pathway.

H. Exit suppression

Continued support can prevent inefficient companies from exiting.

The European Commission specifically recognises that poorly targeted R&D&I support can prevent inefficient undertakings from exiting and can deter competitors from entering or investing.

13. Algorithmic Industrial Policy and State-Owned Enterprises

The problem becomes more complicated where the beneficiary is an SOE.

An algorithm might rank an SOE highly because of:

• government ownership;
• employment contribution;
• strategic importance;
• national-security classification;
• infrastructure ownership.

The same enterprise may therefore simultaneously be:

1. market participant;
2. beneficiary of government support;
3. infrastructure provider; and
4. data provider to the industrial-policy algorithm.

This creates a potential institutional feedback loop.

14. Public Procurement as an Algorithmic Selection Mechanism

Government procurement algorithms can significantly influence market structure.

Suppose procurement software automatically scores suppliers according to:

• previous government contracts;
• production scale;
• domestic content;
• financial stability;
• delivery history.

Incumbents may repeatedly receive high scores.

That creates:

past government contract → better procurement score → future contract → larger scale → better future score.

Thus procurement algorithms can create algorithmic incumbency advantages.

15. Foreign Subsidies and Algorithmic Industrial Policy

Modern competition policy increasingly examines subsidies originating outside the domestic jurisdiction.

The EU Foreign Subsidies Regulation is particularly relevant because its framework examines whether foreign subsidies improve an undertaking's competitive position and whether they actually or potentially negatively affect competition in the internal market. The Commission's 2026 guidelines identify possible distortions involving acquisitions, operating decisions, investment and different levels of the value chain.

The same analytical logic can apply to algorithmically allocated industrial support.

16. Legal Tests for Reviewing an Algorithmic Industrial Policy

A structured competition-law review can ask:

Step 1 — Identify the intervention

What government measure is being administered?

Step 2 — Identify the algorithm

What model, scoring system or automated rule determines allocation?

Step 3 — Identify beneficiaries

Which undertakings actually receive the advantage?

Step 4 — Establish the reference framework

What is the ordinary regulatory or economic system?

Step 5 — Test selectivity

Does the algorithm differentiate between comparable undertakings?

Step 6 — Identify the advantage

Is the beneficiary receiving:

• money;
• tax relief;
• credit;
• infrastructure;
• procurement;
• licensing;
• preferential access?

Step 7 — Analyse competitive effects

Does the intervention affect:

• entry;
• exit;
• prices;
• investment;
• innovation;
• market concentration?

Step 8 — Examine justification

Is the differentiation connected to a legitimate public-policy objective?

Step 9 — Examine proportionality

Is the intervention appropriately limited to achieving that objective?

Step 10 — Examine cumulative effects

Multiple individually modest algorithmic advantages may collectively create substantial market distortion.

17. Remedies

Potential remedies depend upon the legal regime involved but may include:

1. redesigning eligibility criteria;
2. eliminating discriminatory variables;
3. independent algorithmic auditing;
4. transparency regarding material decision criteria;
5. periodic reassessment of beneficiaries;
6. sunset clauses;
7. competitive-neutrality requirements;
8. open procurement procedures;
9. interoperability requirements;
10. separation of governmental data from beneficiary firms;
11. recovery of incompatible aid where legally required;
12. monitoring cumulative subsidy effects.

The objective is not necessarily to eliminate industrial policy, but to prevent industrial policy from silently becoming a mechanism for permanent competitive preference.

18. Key Doctrinal Principle

The central principle can be expressed as:

Algorithmic neutrality is not the same as competitive neutrality.

An algorithm may apply the same mathematical formula to every applicant and still produce a selective competitive advantage.

Therefore, legal scrutiny should examine the structure, inputs, reference framework, beneficiaries and market effects of the system, rather than merely asking whether the algorithm is formally uniform.

19. Exam-Oriented Conclusion

Algorithmic industrial policy represents a new intersection between State intervention, competition law, industrial strategy and automated decision-making. Algorithms can improve administrative efficiency, identify strategic industries and allocate scarce public resources. At the same time, they can embed selective preferences within apparently objective computational systems.

The principal legal concern is not simply that governments use algorithms. It is that algorithmic criteria may determine which undertakings receive advantages that alter competitive conditions.

The jurisprudence concerning State aid and selective advantages—including Commission v Italy, Adria-Wien Pipeline, Paint Graphos, Commission v Netherlands, Ryanair v Commission and Eventech—provides important principles for examining such measures. The cases demonstrate that legal analysis focuses on the economic substance, selectivity, comparability, competitive advantage and justification of governmental intervention.

Accordingly, future competition-law scrutiny of algorithmic industrial policy should examine not only whether an algorithm is accurate or technologically neutral, but also whether it creates path-dependent advantages, incumbent protection, entry barriers, investment distortions or persistent selective market structures.

Core formula:

Algorithmic policy criteria → selective allocation → economic advantage → altered competitive incentives → market distortion → possible competition-law consequences.