Competition Law and Competition Concerns in Semantic Infrastructures

1. Introduction

“Semantic infrastructures” is not a traditional legal category in competition law. It is best understood as the digital infrastructure that determines how information is described, classified, connected, interpreted, ranked, and made understandable to machines and users.

It can include:

• search indexes and knowledge graphs;
• ontologies and taxonomies;
• metadata standards;
• product and business classification systems;
• recommendation and ranking systems;
• entity-resolution systems;
• semantic APIs;
• data schemas and interoperability protocols;
• AI systems that convert raw information into structured meaning.

For example, an online platform may decide that two products belong to the same category, that one webpage is more relevant than another, or that different references describe the same company. Those apparently technical decisions can determine whether businesses are discoverable, interoperable and commercially visible.

Competition problems arise when control over this semantic layer becomes a source of market power. A dominant firm may be able to determine not merely what information exists, but also how markets, products and competitors are digitally understood and presented.

Existing competition law can address many of these problems through rules concerning abuse of dominance, discriminatory access, interoperability, tying, self-preferencing, refusal to supply and exclusionary conduct. The Google Shopping litigation confirms, for example, that the treatment and presentation of information by a dominant search service can fall within Article 102 TFEU where the conduct is capable of restricting competition.

2. Why Semantic Infrastructure Has Economic Importance

Traditional infrastructure connects physical actors. Roads connect locations, telecommunications networks connect users, and electricity networks connect producers and consumers.

Semantic infrastructure performs a similar function for information.

Suppose thousands of sellers offer products online. A semantic system may determine:

Raw data → classification → indexing → interpretation → ranking → recommendation → consumer discovery

A business excluded at the classification or indexing stage may technically remain online while becoming commercially difficult to find.

Consequently, control over semantic infrastructure can create a form of gateway power.

The competition concern becomes especially significant when businesses cannot economically avoid the infrastructure. The Google Shopping judgment concerned circumstances in which a dominant general search provider treated its own specialised comparison-shopping results more favourably than competing services. The Court of Justice ultimately upheld the finding that the conduct could constitute an abuse under Article 102 TFEU.

3. Relevant Markets

Competition authorities first have to identify the relevant market.

Semantic infrastructures can create several overlapping markets rather than one single “semantic market”, such as:

Upstream markets

• data collection;
• databases;
• knowledge graphs;
• ontology services;
• semantic-processing technologies.

Infrastructure markets

• search engines;
• operating systems;
• cloud platforms;
• semantic APIs;
• AI platforms;
• interoperability services.

Downstream markets

• online shopping;
• advertising;
• recommendation services;
• specialised search;
• digital publishing;
• enterprise software.

This distinction matters because a company might possess market power in an infrastructure market and use it to obtain advantages in another market.

This is commonly analysed as leveraging.

4. Main Competition Concerns

4.1 Semantic Gatekeeping

A platform controlling an important semantic layer may effectively determine which businesses, products or information are recognised by downstream systems.

The infrastructure operator might determine:

• which entities are indexed;
• which product categories exist;
• how businesses are classified;
• which attributes receive importance;
• which sources are treated as authoritative;
• how relationships between entities are represented.

When access to such infrastructure becomes commercially indispensable, control can become a substantial competitive advantage.

The important competition-law question is therefore not simply whether the firm owns the technology. Authorities must determine whether the firm's conduct restricts competition through its control of that infrastructure.

4.2 Self-Preferencing

A vertically integrated semantic infrastructure provider may operate both:

1. the infrastructure used to organise information; and
2. services competing through that infrastructure.

This produces a potential conflict.

For example:

Platform controls classification → platform operates competing service → classification/ranking favours platform service → rival visibility declines.

The Google Shopping litigation is particularly important here.

In Google and Alphabet v Commission (Google Shopping), Case T-612/17, the General Court examined Google's more favourable positioning and display of its own comparison-shopping service while competing comparison-shopping services were subject to Google's general ranking mechanisms.

On appeal, the Court of Justice in Case C-48/22 P upheld the central abuse finding and addressed leveraging, potential foreclosure and competition on the merits.

This reasoning is highly relevant to semantic infrastructure because classification and ranking architecture can determine downstream visibility.

4.3 Interoperability Restrictions

Semantic systems become particularly valuable when different databases and services can understand one another.

A dominant provider could potentially restrict interoperability by withholding:

• schemas;
• interface specifications;
• protocol information;
• semantic mappings;
• compatibility documentation;
• APIs.

This can increase rivals' costs.

A major precedent is:

Microsoft Corp. v Commission — Case T-201/04

Microsoft was found to have abused its dominant position in circumstances including its refusal to provide and authorise the use of certain interoperability information concerning Windows work-group server environments.

The General Court upheld the Commission's core interoperability findings.

The case is especially relevant because the Court recognised the competitive significance of interoperability in software markets. The judgment records the Commission's finding that the interoperability obtainable by Microsoft's competitors was insufficient for them to remain viably competitive.

The principle can potentially extend conceptually to semantic infrastructures where meaningful interoperability depends upon access to technical specifications.

4.4 Semantic Lock-In

Semantic infrastructure can also generate switching costs.

Imagine an organisation has spent years structuring millions of records according to one provider's:

• taxonomy;
• identifiers;
• metadata;
• ontology;
• APIs;
• knowledge graph.

Changing provider may require extensive conversion.

This can produce:

Semantic standard → adoption → accumulated structured data → switching costs → lock-in → stronger market power

The concern is not that lock-in itself is automatically unlawful. Competition law becomes relevant where market power associated with lock-in is accompanied by exclusionary or exploitative conduct prohibited by the applicable competition regime.

4.5 Network Effects

Semantic systems often improve as participation increases.

More users can produce:

• more searches;
• more behavioural information;
• better entity matching;
• improved classifications;
• richer knowledge graphs;
• better recommendations.

This can create feedback effects:

More users → more data → improved semantic system → better service → more users

Such effects can make entry difficult for new competitors.

Again, network effects are not themselves an infringement. They become important when assessing market power, entry barriers and the effects of potentially exclusionary behaviour.

4.6 Control of Semantic Data

Semantic infrastructure often depends on large amounts of structured information.

A company possessing uniquely valuable data may enjoy advantages that rivals cannot easily reproduce.

Possible competition concerns include:

• exclusive access to important datasets;
• restrictions on data portability;
• discriminatory API access;
• contractual restrictions preventing data sharing;
• combining datasets in ways competitors cannot replicate;
• withholding information required for interoperability.

Competition authorities therefore have to distinguish legitimate investment advantages from conduct capable of foreclosing competition.

4.7 Semantic Standards and Standardisation

Common semantic standards can produce enormous efficiencies.

For example, standard identifiers may allow different software systems to recognise that:

“Company A Ltd”, “Company A Limited” and a particular corporate identifier refer to the same entity.

But standardisation can also create competition concerns if control over a standard is used to exclude competitors.

Problems can arise where:

• competitors cannot obtain necessary specifications;
• standards discriminate between participants;
• licensing conditions favour particular firms;
• dominant participants control standard-setting;
• proprietary extensions make switching difficult.

Competition law therefore seeks to preserve the benefits of standardisation without allowing standards to become exclusionary bottlenecks.

5. Relevant Case Laws

Because courts have not yet developed a standalone doctrine called “competition law of semantic infrastructures,” the strongest precedents come from adjacent areas such as search, interoperability, platforms, essential inputs and digital ecosystems.

Case 1 — Google and Alphabet v Commission (Google Shopping), T-612/17

This is one of the most directly relevant cases.

Google operated general search while also offering its own comparison-shopping service.

The Commission concluded that Google gave its own comparison-shopping service favourable treatment relative to competing comparison-shopping services. The General Court substantially upheld the Commission's infringement finding.

Relevance to Semantic Infrastructure

Search engines do more than retrieve information. Their systems classify and rank information.

Therefore, control over the information-discovery architecture can affect competition between downstream services.

The case demonstrates that the architecture through which information is presented can itself become relevant to competition analysis.

Case 2 — Google and Alphabet v Commission, C-48/22 P

The Court of Justice delivered its appeal judgment on 10 September 2024.

It upheld the central finding concerning Google's conduct and discussed whether the behaviour represented competition on the merits or an exclusionary leveraging practice capable of restricting competition.

Semantic Significance

The case illustrates that dominance combined with control over information ranking does not automatically produce liability.

The legal analysis focuses on the specific characteristics of the conduct and its competitive effects.

This distinction is important for future semantic-infrastructure cases.

Case 3 — Microsoft v Commission, T-201/04

This landmark case concerned Microsoft's dominant position in PC operating systems and conduct involving work-group server operating systems and media players.

A major issue was Microsoft's refusal to provide certain interoperability information.

The General Court substantially upheld the Commission's decision concerning interoperability.

Semantic Significance

Modern semantic ecosystems depend heavily on interoperability.

If competitors require specifications to make their systems communicate meaningfully with a dominant infrastructure, Microsoft's interoperability principles may become relevant.

The case therefore provides an important conceptual foundation for competition disputes involving closed semantic ecosystems.

Case 4 — Commercial Solvents v Commission, Joined Cases 6/73 and 7/73

Commercial Solvents concerned a dominant undertaking's refusal to continue supplying an input to a customer that competed with it downstream.

The Court accepted that a dominant undertaking could infringe competition rules by using control over an upstream input to eliminate competition in a downstream market.

Semantic Significance

A comparable structure could arise where a company controls an important semantic input—for example:

• a unique structured dataset;
• entity identifiers;
• classification information;
• semantic API access.

If the infrastructure owner also competes downstream, withholding the input could potentially raise refusal-to-supply or leveraging questions.

The precise legal conditions would still have to be established.

Case 5 — United Brands v Commission, Case 27/76

United Brands remains a foundational authority concerning dominance under European competition law.

It helped establish the concept of dominance as economic strength enabling an undertaking to behave to an appreciable extent independently of competitors, customers and ultimately consumers.

Semantic Significance

This principle matters when assessing whether a semantic infrastructure provider possesses market power.

Relevant factors could include:

• market share;
• switching costs;
• network effects;
• data advantages;
• interoperability barriers;
• economies of scale;
• dependence of downstream firms.

Possessing a popular semantic system alone does not establish unlawful dominance.

Case 6 — Hoffmann-La Roche v Commission, Case 85/76

Hoffmann-La Roche is another foundational Article 102 judgment.

The case dealt importantly with dominance and exclusionary loyalty-inducing arrangements.

Semantic Significance

The broader lesson is that a dominant firm has special competition-law responsibilities and cannot use mechanisms that improperly reinforce customer dependence and exclude rivals.

In semantic markets, comparable concerns might arise from contractual or technical arrangements that make customers effectively dependent upon one ontology, API or semantic ecosystem.

Case 7 — Bronner, Case C-7/97

Oscar Bronner sought access to a nationwide newspaper home-delivery system operated by a competitor.

The Court adopted a demanding standard for compelling a dominant undertaking to provide access to infrastructure.

Among other matters, indispensability and the possibility of alternatives were central considerations.

Semantic Significance

Bronner is extremely important because not every valuable semantic infrastructure must automatically be opened to competitors.

A rival generally cannot establish an abuse merely by showing:

“Access to this system would make competing easier.”

Questions such as indispensability, realistic alternatives and the likely elimination of effective competition can become crucial.

Thus Bronner limits overly broad applications of mandatory-access theories to semantic infrastructure.

Case 8 — IMS Health, Case C-418/01

IMS Health concerned a copyrighted geographical structure used for pharmaceutical sales information.

The dispute raised the circumstances in which refusal to license intellectual property by a dominant undertaking could constitute abuse.

Semantic Significance

This case has particularly strong conceptual relevance because semantic infrastructures may themselves contain protected intellectual assets such as:

• database structures;
• classification systems;
• proprietary ontologies;
• software;
• schemas.

Competition law must therefore balance intellectual-property protection against exceptional circumstances in which refusal of access may harm competition.

6. Essential-Facility Questions

A semantic infrastructure might sometimes be argued to function as an essential facility.

But this doctrine must be applied cautiously.

An authority would normally need to investigate questions such as:

Is access genuinely indispensable?

Can competitors realistically construct alternatives?

Would denial eliminate effective competition?

Is there an objective justification for the refusal?

Would mandatory access undermine legitimate investment incentives?

Bronner demonstrates why the threshold for compulsory access is intentionally demanding.

7. Semantic Self-Preferencing

One particularly important future issue is semantic self-preferencing.

Suppose a digital assistant receives the query:

“Find a nearby product suitable for X.”

Before answering, its semantic system may:

1. interpret the request;
2. identify relevant entities;
3. classify products;
4. determine relationships;
5. rank candidates;
6. generate an answer.

If the platform operates competing products, it could theoretically influence competition at any one of those layers.

Traditional ranking discrimination may therefore evolve into something deeper:

semantic discrimination.

A competitor might be classified less accurately, connected to fewer relevant concepts, or omitted from a knowledge structure altogether.

Competition authorities would need evidence demonstrating that such treatment constitutes anticompetitive conduct rather than merely technical differences or legitimate product design.

8. AI and Semantic Infrastructure

Generative AI makes this subject increasingly significant.

AI systems rely on semantic relationships between:

• people;
• companies;
• products;
• locations;
• concepts;
• documents;
• services.

AI assistants can consequently become important gateways between businesses and consumers.

Potential competition questions include whether dominant providers could:

• favour affiliated services in generated answers;
• restrict competing AI systems' access to important semantic APIs;
• combine proprietary data and distribution advantages;
• impose restrictive interoperability conditions;
• make competing services difficult to discover.

Competition analysis would still require evidence of dominance, relevant conduct and competitive harm under the governing legal test.

9. Semantic Infrastructure and Market Entry

New competitors may face several cumulative barriers.

A new entrant may need:

Data + ontology + computing resources + distribution + users + interoperability + continuous learning

Existing firms may already possess all seven.

This creates a potential semantic entry barrier.

The strongest incumbent advantages may therefore come not from a single database but from the combination of:

data + relationships + classifications + user feedback + infrastructure.

Competition authorities increasingly need to examine ecosystems rather than isolated products when these components reinforce one another.

10. Potential Forms of Abuse

Depending upon jurisdiction and factual circumstances, potentially problematic conduct could include:

None of these practices is automatically unlawful. The authority must apply the applicable legal requirements and examine context and effects.

11. Consumer Harm

Competition problems involving semantic infrastructure may harm consumers even where the service appears free.

Possible harms include:

• reduced choice;
• lower-quality search results;
• reduced innovation;
• distorted recommendations;
• increased downstream prices;
• less interoperability;
• greater switching costs;
• reduced diversity of information sources.

Therefore, price is only one dimension of competition.

Quality, innovation, choice and access to information may be equally important.

12. Competition-Law Assessment Framework

A useful framework is:

Stage 1 — Identify the Semantic Infrastructure

Determine what actually performs the semantic function:

ontology, knowledge graph, ranking architecture, metadata standard, search index, API or AI model.

Stage 2 — Define Relevant Markets

Identify upstream infrastructure markets and downstream markets affected by the infrastructure.

Stage 3 — Determine Market Power

Examine:

• market shares;
• network effects;
• switching costs;
• data advantages;
• interoperability;
• entry barriers;
• multi-homing.

Stage 4 — Identify Conduct

Determine whether there is:

• refusal of access;
• discrimination;
• tying;
• self-preferencing;
• exclusive dealing;
• interoperability restriction.

Stage 5 — Examine Foreclosure

Ask whether competitors are actually or potentially prevented from competing effectively.

Stage 6 — Examine Justifications

Restrictions may sometimes be justified by matters such as:

• security;
• privacy;
• system integrity;
• intellectual property;
• technical efficiency;
• quality control.

Stage 7 — Consider Remedies

Where an infringement is established, depending on jurisdiction and proportionality, remedies might involve:

• interoperability requirements;
• non-discriminatory access;
• changes to ranking treatment;
• data portability;
• API access;
• licensing obligations;
• removal of exclusionary contractual restrictions.

Microsoft demonstrates the importance of interoperability remedies in digital markets, while Google Shopping illustrates the competition significance of discriminatory treatment within information-discovery infrastructure.

13. Key Case-Law Principles

The cases can be brought together as follows:

United Brands
→ establishes foundational principles for assessing dominance.

Hoffmann-La Roche
→ explains dominant-firm responsibilities and exclusionary conduct.

Commercial Solvents
→ demonstrates how control over an upstream input may be used against downstream competition.

Bronner
→ places strict limits on compulsory infrastructure access.

IMS Health
→ addresses exceptional compulsory-access questions involving intellectual property.

Microsoft v Commission
→ demonstrates the competition significance of interoperability in digital ecosystems.

Google Shopping (T-612/17)
→ addresses discriminatory treatment and leveraging through search infrastructure.

Google Shopping (C-48/22 P)
→ confirms the central Article 102 analysis at Court of Justice level and clarifies the assessment of competition on the merits and potential foreclosure.

Conclusion

Competition law concerning semantic infrastructures can be understood as the application of established competition principles to the digital systems that determine how information is classified, connected, interpreted, discovered and ranked.

The central concern is not ownership of semantic technology itself. Firms remain free to develop better ontologies, search systems, knowledge graphs and AI technologies through competition on the merits.

The legal concern becomes stronger when control over a semantic layer combines with market power and exclusionary conduct—for example, discriminatory ranking, self-preferencing, restrictive interoperability, exclusionary access conditions, tying or strategically withholding indispensable inputs.

The case law already provides a substantial doctrinal foundation. Commercial Solvents, United Brands, Hoffmann-La Roche, Bronner and IMS Health supply the traditional principles concerning dominance and access, while Microsoft provides an important interoperability precedent and Google Shopping shows how the architecture governing digital discovery and visibility can itself become central to an Article 102 investigation.

As AI assistants, knowledge graphs and machine-readable ecosystems become more important, competition analysis is therefore likely to focus increasingly not merely on who controls data, but on who controls the infrastructure that determines what that data means and how it reaches the market.