Geopolitical Fragmentation of Compute Supply Chains and Competition Effects

Introduction

Geopolitical fragmentation of compute supply chains refers to the division of the global supply chain for computational resources—such as GPUs, CPUs, advanced semiconductors, semiconductor-manufacturing equipment, cloud infrastructure, AI accelerators, memory, networking equipment and data centres—into politically aligned or strategically controlled blocs.

Compute has become a critical input for artificial intelligence, cloud computing, autonomous systems, cybersecurity, scientific research and digital services. Consequently, export controls, sanctions, investment screening, subsidies, localisation requirements and national-security policies can alter the competitive structure of compute markets.

From a competition-law perspective, geopolitical fragmentation creates a difficult intersection between national security and market competition. Measures intended to protect technological sovereignty can simultaneously:

• restrict competitors' access to critical inputs;
• increase barriers to entry;
• reinforce incumbent semiconductor or cloud firms;
• fragment previously global markets;
• create regional compute monopolies or oligopolies;
• encourage vertical integration;
• produce discriminatory access to GPUs and cloud capacity; and
• generate divergent regulatory standards.

The central competition question is therefore:

When does legitimate geopolitical risk management become a mechanism that materially distorts competition in compute markets?

1. Meaning and Structure of Compute Supply-Chain Fragmentation

The compute ecosystem is not a single market. It consists of interconnected layers:

A. Semiconductor design

Examples include:

• CPUs;
• GPUs;
• AI accelerators;
• NPUs;
• specialised ASICs.

B. Semiconductor manufacturing

Advanced chips depend on a relatively small number of sophisticated fabrication facilities.

C. Semiconductor manufacturing equipment

Advanced fabrication requires highly specialised equipment, including lithography and other semiconductor-production technologies.

D. Advanced packaging and memory

AI accelerators increasingly depend upon high-bandwidth memory and sophisticated packaging.

E. Networking

Large AI clusters require:

• high-speed interconnects;
• switches;
• optical networking;
• specialised networking processors.

F. Data centres

Compute ultimately requires:

• electricity;
• cooling;
• land;
• networking;
• servers;
• storage.

G. Cloud platforms

Cloud providers aggregate compute and distribute it through:

• infrastructure-as-a-service;
• AI-as-a-service;
• GPU rental;
• inference APIs;
• managed AI platforms.

Thus, geopolitical intervention at one layer can create competition effects throughout the entire chain.

2. Sources of Geopolitical Fragmentation

2.1 Export controls

Governments may restrict exports of advanced:

• GPUs;
• AI accelerators;
• semiconductor manufacturing equipment;
• design software;
• high-performance computing technology.

Such restrictions can divide the market into jurisdictions with access to frontier technology and jurisdictions without it.

Competition effect

A restricted jurisdiction may experience:

1. reduced access to frontier chips;
2. higher compute prices;
3. reduced innovation;
4. greater dependence on domestic suppliers; and
5. accelerated consolidation among those suppliers.

3. Technology Sovereignty and Competition

Governments increasingly view compute as strategic infrastructure.

A state may therefore encourage domestic production through:

• subsidies;
• tax incentives;
• government procurement;
• domestic-content requirements;
• state-backed financing;
• restrictions on foreign acquisitions.

These measures may strengthen domestic competitors.

However, they can also produce protected national champions.

The competition-law concern is particularly strong where:

government protection + scarce compute + dominant infrastructure + regulatory barriers

combine to create durable market power.

4. Creation of Regional Compute Oligopolies

Fragmentation can reduce the number of effective competitors.

Suppose ten global suppliers previously served a worldwide market.

If geopolitical restrictions divide the market into three blocs, each bloc may contain only two or three viable suppliers.

The result can be:

Global competition → regional competition → concentrated oligopoly

This can increase:

• prices;
• margins;
• switching costs;
• capacity constraints;
• bargaining power of infrastructure suppliers.

5. Vertical Foreclosure

Geopolitical fragmentation can encourage vertical integration.

For example:

Chip designer → cloud provider → AI model provider

may become integrated.

A vertically integrated company controlling scarce GPUs could potentially disadvantage independent AI developers by:

• reserving capacity for its own models;
• charging competitors higher prices;
• limiting API access;
• imposing discriminatory contractual conditions;
• refusing interoperability.

The competition issue resembles traditional essential-input and vertical-foreclosure theories, but with compute replacing conventional physical infrastructure.

6. Compute as an Essential Facility

In extreme circumstances, frontier compute infrastructure may acquire characteristics similar to an essential facility.

A competition authority could ask:

1. Is the input indispensable?
2. Is there a realistic substitute?
3. Can competitors economically reproduce the infrastructure?
4. Does the owner control access?
5. Does denial eliminate effective competition?

This is particularly important for AI markets because training frontier models can require enormous computational resources.

However, scarcity alone does not automatically establish an essential-facilities obligation.

7. Competition Effects of Export Controls

Export controls can produce an unusual competition effect.

They may reduce foreign competition while strengthening domestic firms.

For example:

Export restriction

↓

Foreign competitors lose access to advanced technology

↓

Domestic suppliers face less competitive pressure

↓

Domestic suppliers obtain greater market share

↓

Potentially greater concentration

This creates a tension between national-security objectives and competitive neutrality.

Competition law generally cannot simply invalidate a national-security measure, but authorities may examine its consequences where domestic firms subsequently engage in anticompetitive conduct.

8. Subsidies and Competitive Neutrality

Government subsidies for domestic semiconductor and compute industries can produce significant structural effects.

Subsidies may:

• reduce capital costs;
• enable excess capacity;
• facilitate market entry;
• accelerate technological development.

But they may also allow inefficient firms to survive.

The resulting market can therefore experience:

Short-term effect

More supply and potentially lower prices.

Long-term effect

Potential concentration if subsidies disproportionately strengthen a small number of firms.

This is particularly important where subsidy recipients also control critical infrastructure.

9. Geopolitical Fragmentation and Cloud Competition

Cloud computing is particularly vulnerable because cloud providers can simultaneously operate as:

• infrastructure suppliers;
• GPU purchasers;
• AI platform operators;
• AI model distributors;
• application competitors.

A geopolitically fragmented world may increase the strategic value of cloud infrastructure.

A dominant cloud provider could potentially use scarce compute access to favour its own downstream products.

Competition concerns may therefore include:

• discriminatory GPU allocation;
• tying;
• bundling;
• preferential capacity allocation;
• interoperability restrictions;
• data portability barriers;
• cloud switching costs.

10. Geographical Discrimination

Fragmentation may produce geographically differentiated access.

For example:

This creates a phenomenon that can be described as geopolitical compute discrimination.

11. Competition Between Domestic and Foreign Standards

Geopolitical fragmentation may also produce incompatible technical standards.

Examples include differences concerning:

• chip architectures;
• AI accelerators;
• cloud APIs;
• cybersecurity requirements;
• data localisation;
• AI safety certification;
• encryption;
• semiconductor design standards.

If a dominant firm controls a regional standard, competitors may face substantial switching costs.

This resembles standard-setting and interoperability competition.

12. Strategic Stockpiling and Capacity Hoarding

Governments and large corporations may stockpile advanced chips.

This can generate competition effects even without conventional exclusionary conduct.

If scarce GPU capacity is purchased in advance:

Stockpiling → reduced available supply → higher prices → reduced access for smaller competitors

The competitive harm may disproportionately affect:

• startups;
• universities;
• independent AI developers;
• smaller cloud providers.

Thus, capacity allocation itself can become a competition issue.

13. Case Law

The following cases provide useful doctrinal foundations for analysing geopolitical fragmentation even though most arose in telecommunications, infrastructure, technology, essential facilities, State measures or international economic regulation rather than modern GPU markets specifically.

1. United Brands v Commission — C-27/76

The Court of Justice developed important principles concerning dominant market power and access to indispensable commercial relationships.

Relevance

A dominant compute-infrastructure provider controlling a critical input could potentially attract Article 102 TFEU scrutiny where its conduct amounts to exclusionary abuse.

The case helps establish the broader principle that dominant firms have special responsibilities not to distort competitive conditions.

14. Bronner v Mediaprint — C-7/97

This is one of the most important cases for the essential-facilities doctrine.

The Court imposed a demanding threshold before requiring a dominant undertaking to provide access to infrastructure.

Relevance to compute

A GPU cloud, semiconductor platform or compute network should not automatically become an essential facility merely because competitors depend upon it.

The Bronner principles suggest asking whether:

• the infrastructure is indispensable;
• duplication is realistically possible;
• access refusal eliminates effective competition; and
• there is no reasonable substitute.

This is particularly important for frontier AI compute.

15. IMS Health v Commission — C-418/01

The case concerned access to protected information infrastructure and refined the circumstances under which refusal to license/access infrastructure may constitute abuse.

Competition significance

The Court identified stringent conditions surrounding:

• indispensability;
• elimination of competition;
• prevention of a new product;
• absence of objective justification.

Compute relevance

A dominant AI infrastructure provider controlling a unique computational ecosystem might face analogous questions where competitors cannot realistically reproduce the necessary infrastructure.

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

The Microsoft case is particularly important for technology ecosystems.

The General Court considered Microsoft's refusal to provide interoperability information and the resulting competitive effects.

Relevance to geopolitical compute fragmentation

Compute ecosystems increasingly depend upon interoperability between:

• chips;
• operating systems;
• cloud platforms;
• AI frameworks;
• APIs;
• networking systems.

A geopolitical bloc that develops a proprietary technological ecosystem could create technical lock-in.

If a dominant firm uses interoperability restrictions to protect its downstream market, Microsoft provides a useful analytical framework.

17. Slovak Telekom v Commission — Joined Cases C-165/19 P and C-166/19 P

This case concerned access to telecommunications infrastructure and exclusionary conduct.

Relevance

Telecommunications infrastructure and compute infrastructure share an important characteristic:

competitors may require access to infrastructure controlled by another undertaking to compete downstream.

The case therefore helps analyse:

• margin squeeze;
• access conditions;
• infrastructure dependence;
• downstream foreclosure.

For compute markets, analogous issues may arise when a dominant cloud provider controls both infrastructure and downstream AI services.

18. Deutsche Telekom v Commission — C-280/08 P

The case concerned a dominant telecommunications operator and margin squeeze.

Compute application

Imagine a cloud provider:

• charges independent AI firms very high GPU-access prices;
• simultaneously provides cheaper compute to its own downstream AI products.

This could potentially generate a margin-squeeze theory where the downstream competitor cannot compete effectively against the integrated provider.

The case therefore provides an important conceptual bridge between telecommunications infrastructure and cloud-compute infrastructure.

19. Huawei Technologies v ZTE — Joined Cases C-170/13 and C-47/13

Huawei v ZTE concerned standard-essential patents and the interaction between intellectual property rights and competition law.

Relevance to fragmented compute ecosystems

Future compute ecosystems may depend upon:

• standard-essential technologies;
• interoperability protocols;
• proprietary interfaces;
• AI accelerator standards.

Geopolitical fragmentation could result in competing technological standards.

Huawei v ZTE demonstrates why access to standardised technology can become a competition-law issue when control over a technological standard gives a firm substantial bargaining power.

20. Intel v Commission — C-413/14 P

Intel concerned exclusionary rebates and the competitive effects of practices by a dominant supplier.

Relevance

Suppose a dominant semiconductor supplier provides preferential pricing or rebates to:

• major cloud providers;
• data-centre operators;
• large AI companies,

in return for purchasing commitments that make it difficult for rival chip suppliers to obtain sufficient scale.

The Intel judgment is important for analysing whether such arrangements are capable of foreclosure and how their economic effects should be assessed.

21. European Commission v Broadcom / VMware — Modern Technology-Platform Context

The Broadcom-related enforcement environment illustrates the increasing importance of competition scrutiny in technology ecosystems involving infrastructure and interoperability.

Its broader relevance lies in examining how control over strategically important technological components can affect downstream competition.

The case illustrates the modern movement away from analysing technology markets solely through consumer prices and toward considering:

• interoperability;
• access;
• ecosystem control;
• switching costs;
• foreclosure.

22. Lessons From the Case Law

The cases collectively establish several useful principles.

Principle 1 — Infrastructure control can generate market power

Control over infrastructure can create substantial competitive leverage.

Principle 2 — Indispensability matters

Not every scarce resource constitutes an essential facility.

Principle 3 — Vertical integration can create foreclosure risks

Where an infrastructure owner competes downstream, discriminatory access can become problematic.

Principle 4 — Interoperability matters

Technological incompatibility can reinforce market power.

Principle 5 — Exclusionary rebates can reinforce concentration

Long-term purchasing arrangements can potentially prevent rival technologies from achieving scale.

Principle 6 — Competition analysis remains distinct from national-security policy

A State may legitimately restrict technology transfers for security reasons. But private firms cannot necessarily invoke geopolitical considerations as a blanket justification for otherwise unlawful exclusionary conduct.

23. Geopolitical Fragmentation and Merger Control

Fragmentation may substantially change merger analysis.

A semiconductor merger that previously appeared relatively modest could become strategically significant if geopolitical restrictions reduce the number of alternative suppliers.

For example:

Before fragmentation

10 suppliers → relatively competitive market

After fragmentation

3 permitted suppliers → concentrated market

A merger between two of those three firms could create a substantially greater competition concern.

Authorities may therefore need to consider:

• geopolitical supply restrictions;
• effective global supply;
• regional substitutes;
• technological substitutability;
• future innovation;
• access to manufacturing capacity.

24. Innovation Competition

Compute fragmentation can affect innovation competition, not merely prices.

AI startups may require access to frontier compute to develop:

• new models;
• scientific applications;
• autonomous systems;
• specialised AI products.

If compute is concentrated among incumbent firms, innovative entrants may never reach sufficient scale.

This creates a dynamic competition problem.

The relevant question becomes:

Does restricted access to compute prevent future competitors from emerging?

25. The "Compute Divide"

A significant consequence may be the emergence of a compute divide.

Large firms:

• obtain priority GPU allocations;
• negotiate long-term contracts;
• build private data centres;
• secure energy;
• acquire specialised chips.

Smaller firms:

• purchase compute at spot prices;
• experience shortages;
• face higher financing costs;
• lack bargaining power.

The result can be:

Compute scarcity → unequal access → unequal innovation capacity → increased concentration

This may be more significant than conventional price effects.

26. National Champions and Competition Neutrality

Governments may deliberately create national champions.

A national champion can receive:

• subsidies;
• preferential procurement;
• tax incentives;
• regulatory exemptions;
• government-backed financing.

The competition concern arises when the champion subsequently acquires private competitors or controls essential infrastructure.

Competition authorities may therefore need to distinguish between:

Legitimate industrial policy

and

State-enabled market foreclosure.

27. Data Centres, Energy and Compute Concentration

Geopolitical fragmentation also affects data-centre markets.

Advanced AI data centres require enormous amounts of electricity.

Consequently, competition may extend beyond chips into:

• electricity procurement;
• transmission infrastructure;
• renewable-energy contracts;
• cooling infrastructure;
• land;
• grid connections.

A firm controlling scarce energy-plus-compute capacity could possess a substantial competitive advantage.

Thus:

Compute competition increasingly becomes infrastructure competition.

28. Possible Competition-Law Remedies

Authorities may consider several remedies.

Structural remedies

• divestiture;
• separation of infrastructure and downstream operations;
• limits on exclusive procurement.

Behavioural remedies

• non-discriminatory access;
• interoperability obligations;
• transparent GPU allocation;
• prohibition of retaliatory pricing;
• fair licensing.

Data and portability remedies

• cloud portability;
• model portability;
• API interoperability;
• data-transfer rights.

Merger remedies

• capacity commitments;
• licensing;
• technology access;
• supply commitments.

29. Regulatory Coordination

Because compute supply chains are inherently international, unilateral competition enforcement may be insufficient.

Competition authorities may need coordination concerning:

• semiconductor mergers;
• export-control effects;
• cloud markets;
• AI infrastructure;
• cross-border subsidies;
• supply agreements.

However, national-security policies may legitimately differ.

The resulting challenge is:

How can competition authorities preserve competitive markets without undermining legitimate national-security controls?

30. A Competition-Law Analytical Framework

A useful framework is:

Step 1 — Identify the relevant compute market

Is the market:

• GPUs?
• AI accelerators?
• cloud GPU rental?
• frontier-model training?
• inference?
• semiconductor equipment?

Step 2 — Identify geopolitical restrictions

Examine:

• export controls;
• sanctions;
• investment restrictions;
• localisation;
• subsidies.

Step 3 — Measure concentration

Consider:

• market shares;
• HHI;
• capacity;
• technological leadership;
• access to manufacturing.

Step 4 — Analyse vertical relationships

Determine whether firms operate across:

chips → cloud → models → applications

Step 5 — Assess exclusion

Look for:

• refusal to supply;
• discriminatory pricing;
• exclusive contracts;
• capacity reservation;
• interoperability restrictions.

Step 6 — Consider innovation effects

Examine whether smaller competitors can realistically obtain sufficient compute.

Step 7 — Consider objective justification

Separate:

legitimate national-security measures

from

private commercial exclusion justified merely by reference to geopolitical conditions.

Step 8 — Design remedies

Possible remedies include:

• interoperability;
• non-discrimination;
• access obligations;
• divestiture;
• supply commitments.

31. Hypothetical Example

Assume three countries form separate technology blocs.

Country A controls advanced GPU design.

Country B controls advanced semiconductor manufacturing.

Country C controls a major cloud-compute ecosystem.

Export controls prevent firms in Country C from purchasing the newest GPUs.

Country C's domestic cloud providers therefore acquire older chips and raise prices.

One provider simultaneously develops its own AI model.

It gives its AI subsidiary preferential access to scarce GPU capacity while independent AI developers receive access only at substantially higher prices.

There are now two separate competition problems:

Geopolitical problem

The export restrictions fragment the global compute market.

Competition-law problem

The cloud provider potentially exploits scarcity to disadvantage downstream competitors.

The second problem does not automatically disappear merely because the original scarcity arose from geopolitics.

32. Key Competition Effects

The principal effects can be summarised as follows:

33. Emerging Legal Concept: Compute Sovereignty

Compute sovereignty describes a State's ability to maintain domestic access to strategically important computational resources.

It may involve:

• domestic semiconductor production;
• sovereign cloud infrastructure;
• national AI supercomputers;
• strategic GPU reserves;
• domestic data centres;
• government-controlled compute capacity.

Compute sovereignty may be justified on national-security grounds.

But from a competition perspective, excessive sovereignty policies can create closed national markets.

The long-term danger is therefore:

security-driven fragmentation → protected national champions → reduced competition → technological concentration.

Conclusion

Geopolitical fragmentation of compute supply chains represents a new frontier of competition law because compute is becoming a strategic economic input rather than merely another technological product.

The principal competition risks arise when geopolitical fragmentation combines with:

• concentrated chip production;
• scarce GPU capacity;
• cloud-platform dominance;
• vertical integration;
• exclusive procurement;
• interoperability restrictions;
• national champions;
• subsidies;
• strategic stockpiling.

The jurisprudence of United Brands, Bronner, IMS Health, Microsoft, Slovak Telekom, Deutsche Telekom, Huawei v ZTE and Intel provides useful doctrinal tools for analysing these problems.

The central legal distinction should remain:

States may legitimately protect national security and technological sovereignty, but private undertakings should not be permitted to convert geopolitical scarcity into unlawful market foreclosure.

Ultimately, competition authorities will increasingly have to evaluate access to compute, capacity, interoperability, technological substitution and innovation opportunities, rather than relying solely upon conventional price-based market analysis.