Artificial Intelligence Markets: Competition Risks and Strategic Developments

 The sources indicate that while artificial intelligence (AI) has the potential to deliver substantial economic and social gains, these benefits are contingent upon maintaining competitive and contestable markets across the entire AI value chain.

Drivers of Economic Impact

• Productivity Growth: AI is widely expected to increase productivity across the global economy, though estimates of this impact vary.
• Declining Costs and Innovation: In the foundation model segment, competition has led to a rapid increase in performance and a sharp decline in prices. For instance, the quality-adjusted price index for text-to-text models fell by nearly 80% between January 2024 and April 2026.
• The AI Economic Frontier: This concept describes the trade-off where users can access increasingly better model performance at lower prices. The number of developers reaching this frontier has grown from two to eight in three years, indicating high pressure on leading firms to innovate.

Barriers to Realizing Economic Gains

The sources highlight that realizing these gains assumes broad access to affordable AI tools, which is threatened by several structural risks:

• Market Concentration: Key inputs like specialized chips (NVIDIA holds 90% of the GPU market) and cloud infrastructure (three providers control 74% of the global market) are highly concentrated.
• Rising Effective Costs: While price-per-token is falling, the "effective cost" of using AI may rise as AI agents consume significantly more tokens to execute complex tasks.
• Systemic Adoption Challenges: Broad productivity gains require "systemic adoption"—deep integration into business processes—which necessitates slow-moving organizational changes and complementary investments in data and skills.

Competition and Structural Risks

The economic impact is closely tied to whether markets remain contestable over time. Several factors threaten this:

• Vertical Integration: Large tech firms often operate across multiple layers of the value chain (hardware, cloud, and applications), allowing them to bundle products, favor their own offerings, and raise switching costs for users.
• First-Mover Advantages: Early leaders benefit from data feedback loops and self-improving AI capabilities, which can create "market tipping" where a few dominant players become entrenched.
• Input Bottlenecks: Barriers to entry are high due to the concentration of critical inputs like compute, proprietary datasets, and specialized skills.

Macroeconomic and Global Asymmetries

The sources warn that AI may widen the gap between leading and lagging economies:

• Investment Concentration: AI investment is heavily concentrated in the United States and China. In 2025, total estimated AI investments in the U.S. were more than double those of all other OECD countries combined.
• Economic Dependence: Countries without domestic AI ecosystems may reap some productivity benefits but will likely capture only a limited share of market rents. Furthermore, they may become increasingly reliant on imported AI services, exposing them to service disruptions and the pricing power of foreign oligopolies.

The sources present a "mixed picture" of AI markets, where intense market dynamism in specific segments—particularly foundation models—coexists with deep-seated structural risks that threaten long-term competition.

Evidence of Market Dynamism

Despite high entry costs and the dominance of a few tech giants, the AI model development segment has shown significant signs of competitive pressure and innovation over the past three years:

• Rapid Entry and Model Proliferation: The number of developers focusing on language models for cognitive tasks grew from 9 in January 2024 to 47 in April 2026. During this same period, the number of active text-to-text models surged from 22 to 453.
• Declining Quality-Adjusted Prices: The aggregate OECD price index for text-to-text models fell by nearly 80% between January 2024 and April 2026. This decline is observed across all segments, including coding and agentic capabilities.
• Contested Leadership: Technological leadership is frequently challenged by "dynamic, specialised providers" who compete directly with tech giants.

The AI Economic Frontier

A key indicator of market dynamism is the AI Economic Frontier, which tracks the best available model performance for a given price.

• Frontier Movement: The frontier has moved steadily, offering better models at lower prices.
• Increasing Participants: The number of developers reaching this frontier increased from two to eight over three years.
• Transience of Leadership: Presence at the "technological frontier"—defined by model quality and price—often lasts only a few months, indicating that even leading firms face continuous pressure from newcomers.

Dynamism vs. Structural Risks

The sources emphasize that this dynamism is currently concentrated in the model development layer and may be fragile due to issues in other parts of the AI value chain:

• Input Bottlenecks: Upstream segments like high-end chips (where one firm holds 90% market share) and cloud infrastructure (where three providers control 74% of the market) remain structurally concentrated.
• Vertical Integration: Large incumbents often operate across multiple layers (model development, cloud, and downstream applications), allowing them to use preferential partnerships and product bundling to entrench their positions.
• Data Feedback Loops: First-movers benefit from self-improving AI capabilities and exclusive data access, which can create "market tipping" where a few players become dominant despite initial dynamism.

The Role of Open Source

Open-source development is cited as a significant factor in mitigating concentration by lowering entry costs and putting price pressure on incumbents. However, the sources warn that open-source ecosystems can also be used by large firms to create exclusive bundles that reinforce their own first-mover advantages.

Policy Implications for Maintaining Dynamism

To ensure these dynamic outcomes last, the sources suggest that policymakers must remain vigilant. They recommend promoting market contestability by facilitating access to critical inputs (data, compute, and skills) and scrutinizing vertical strategies that could lock in users or foreclose rivals. Coordination across borders is highlighted as essential to handle the global nature of these fast-evolving markets.

While the AI model development segment currently shows signs of dynamism, the sources emphasize that broader structural risks to competition are "real, persistent and expected to intensify" across the entire AI value chain. These risks threaten to entrench incumbent positions and limit the long-term contestability of the market.

Concentration of Critical Inputs

A primary structural risk is the high concentration of the essential inputs required to develop and deploy AI:

• Hardware: The market for specialized chips is extremely concentrated; for example, NVIDIA accounted for 90% of the GPU market in 2023. High fixed costs and long development times create significant barriers to entry for new designers.
• Cloud Infrastructure: Large-scale AI requires massive compute power, which is dominated by a few providers. In 2023, the three major cloud providers controlled 74% of the global market.
• Data: Incumbents with large user bases benefit from exclusive data partnerships and proprietary datasets, which can lead to data concentration and gatekeeping.
• Specialized Skills: The scarcity of talent needed to build frontier models further restricts the ability of new entrants to compete effectively.

First-Mover Advantages and Feedback Loops

The sources highlight that early leaders benefit from self-reinforcing cycles that can lead to "market tipping," where one or two firms dominate a segment:

• Data Feedback Loops: More users generate more data, which is used to improve models, attracting even more users.
• Cumulative Learning: AI models can exhibit self-improving capabilities, allowing first movers to maintain durable leadership through continuous refinement.
• Reputational Advantages: Early success in the market creates brand loyalty and trust that newcomers find difficult to overcome.

Vertical Integration and Ecosystem Foreclosure

Many leading digital firms operate across multiple layers of the value chain—from chips and cloud to models and downstream applications. This vertical integration presents several competition risks:

• Preferential Access: Vertically integrated firms may provide their own downstream services with better or cheaper access to upstream compute and models.
• Product Bundling: Firms can leverage their dominance in one area (like cloud services or operating systems) to bundle AI tools, effectively raising switching costs and locking in users.
• Strategic Partnerships: Large incumbents often form exclusive or preferential partnerships with leading AI labs, potentially foreclosing rivals' access to the most capable "frontier" models.

Barriers to Contestability

The structure of AI markets creates high barriers that prevent new firms from challenging incumbents:

• High Sunk Costs: The massive investment required for R&D and compute infrastructure—with the five largest U.S. tech firms forecast to spend $660 billion in capital expenditure in 2026—makes entry prohibitively expensive for most.
• User Lock-in: Integration of AI into core business processes and IT systems creates high operational and cybersecurity risks if a user tries to switch providers.
• Regulatory Influence: Dominant, vertically integrated firms may use their market power to shape regulations in ways that favor their business models and disadvantage smaller competitors.

Cross-Country Asymmetries

Finally, these structural risks have a geopolitical dimension. AI investment and capacity are heavily concentrated in a few countries, primarily the United States and China. Economies with weaker domestic AI ecosystems risk becoming increasingly reliant on imported services, exposing them to the pricing power and potential service disruptions of foreign oligopolies.

The sources emphasize that while some segments of AI development show dynamism, the broader AI value chain is subject to structural risks that are "real, persistent and expected to intensify". These risks are categorized across various layers, from upstream infrastructure to downstream applications.

Upstream Risks: Input Bottlenecks and Concentration

The primary risks in the upstream segments involve the extreme concentration of the physical and digital resources required to build AI:

• Hardware (Chips): This segment is characterized by high fixed costs, long development times, and a "dependency on a small number of suppliers". For example, NVIDIA accounted for 90% of the GPU market in 2023. The risk here is the foreclosure of access or preferential access for large incumbents.
• Cloud Infrastructure: Dominated by a few providers (three firms controlled 74% of the global market in 2023), this layer faces risks of "entrenchment" and the "leveraging of cloud market power into adjacent AI segments" through product bundling and vertical foreclosure.
• Data: Incumbents benefit from exclusive data partnerships and "data feedback loops," where existing large user bases provide a continuous stream of data to improve models. This creates gatekeeping effects that reduce the ability of new entrants to train competitive models.

Midstream Risks: Foundation Model Oligopolies

The foundation model layer, which serves as the backbone for generative AI, faces "oligopolistic market structures".

• Durable Leadership: Despite recent price declines, leadership may become entrenched through cumulative learning and self-improving AI capabilities.
• Selective Access: There is a risk of "preferential or selective access to frontier models," where providers favor their own downstream services or those of strategic partners.

Downstream Risks: Market Tipping and Lock-in

As AI moves into consumer and business services, the risks shift toward user control and market dominance:

• User Lock-in: Integration into existing digital ecosystems and high switching costs can prevent users from moving to better or cheaper alternatives.
• Market Tipping: Risks include "market tipping in key use cases" where a single provider becomes the default due to network effects and exclusive distribution.

Integrative Risks Across the Chain

Two overarching risks threaten the contestability of the entire value chain:

• Vertical Integration: Leading digital firms often operate across multiple layers (e.g., owning the cloud infrastructure, developing the model, and providing the downstream app). This allows them to "favour their own offerings" and bundle AI with existing products, effectively raising barriers for specialized rivals.
• First-Mover Advantages: Concentration is compounded by positive feedback loops where early leaders gain reputational advantages and superior data, making it difficult for newcomers to compete even if they have innovative technology.

Geopolitical and Macroeconomic Dimensions

The sources also highlight that these value chain risks have a global impact. AI investment and capacity are heavily concentrated in the United States and China. Countries with weaker domestic ecosystems may become increasingly reliant on imported AI services, exposing them to the pricing power, service disruptions, and regulatory conflicts of foreign oligopolies.

The sources indicate that while AI has the potential for significant economic growth, several barriers to adoption prevent users and firms from fully realizing these benefits. These barriers are closely linked to market concentration and the strategic behavior of dominant firms.

Economic and Financial Barriers

• Rising Effective Costs: While the price per million tokens has declined by nearly 80% since 2024, the "effective cost" of using AI may still rise. This is because AI agents—which execute complex, multi-step tasks—consume tokens at a much higher intensity than simple chatbots.
• Need for Complementary Investments: Broad productivity gains require "systemic adoption," which means deep integration into core business processes rather than occasional use. This necessitates slow-moving organizational changes and significant investments in intangibles like data and specialized skills.
• High Entry and Sunk Costs: For smaller firms and SMEs, the high investment barriers for R&D and compute infrastructure can be prohibitively expensive.

Technical and Operational Barriers

• Cybersecurity and Safety Risks: As AI agents become more deeply integrated into IT systems, concerns over safety and operational reliability are becoming critical barriers.
• Interoperability Issues: A lack of system interoperability can prevent firms from adopting diverse AI solutions, forcing them to rely on a single provider.
• Energy and Supply Chain Disruptions: Recent energy shocks and supply-chain issues can reinforce concentration, as only vertically integrated giants may have the infrastructure to cope with rising costs and tightening financing conditions.

Competition-Related Barriers

• Switching Costs and User Lock-in: High operational risks and the deep integration of AI into business processes make it difficult for users to switch providers. Incumbents often leverage this by bundling AI tools with existing digital ecosystems or cybersecurity solutions, effectively locking in users.
• Preferential Access: Large incumbents may use exclusive partnerships or vertical integration to provide themselves with better or cheaper access to "frontier" models, making it harder for rivals to offer competitive downstream services.

Geographic and Cultural Barriers

• Language Disparities: AI adoption is hampered in many countries because performance in non-English languages remains significantly lower.
• Cross-Country Asymmetries: Countries with weak domestic AI ecosystems face a "productivity divide," becoming increasingly reliant on imported AI services from a few global giants in the U.S. and China. This dependence exposes them to potential service disruptions and the pricing power of foreign oligopolies.

To address these barriers, the sources suggest that policymakers should focus on facilitating access to critical inputs (data, compute, and skills) and promoting market transparency to ensure the market remains contestable for new entrants.

The source provides a clear roadmap for policymakers, emphasizing that they must adopt a balanced, proactive, and forward-looking approach to monitoring and promoting competition in AI markets. Because AI has the potential for significant economic gains, policy efforts should focus on ensuring the market remains open and contestable across the entire value chain.

Direct Actions to Enhance Contestability

Policymakers are encouraged to foster competition and innovation simultaneously through several specific measures:

• Facilitating Access to Inputs: Policies should aim to improve access to critical AI inputs, specifically data, compute capacity, and specialized skills.
• Supporting SMEs: Efforts should be made to remove investment barriers that prevent small and medium-sized enterprises (SMEs) from entering or competing in the market.
• Improving Transparency and Experimentation: Increasing market transparency and utilizing regulatory sandboxes can encourage experimentation without stifling innovation.
• Promoting Interoperability: Ensuring system interoperability is vital to allow users to switch between different AI solutions at a low cost, preventing provider lock-in.

Scrutiny of Market Strategies

The source highlights the need for continuous vigilance regarding the strategic behavior of dominant firms:

• Real-Time Monitoring: Authorities must monitor fast-evolving market developments in real time to identify emerging risks.
• Scrutinizing Vertical Integration: Policymakers should closely examine vertical strategies, such as acquisitions and partnerships, that may allow firms to favor their own products or foreclose rivals from accessing "frontier" models.
• Addressing Regulatory Influence: There is a need to protect against undue influence or lobbying by dominant, vertically integrated firms that might try to shape regulations to their advantage.

International and Domestic Coordination

Given that AI markets are global and involve complex infrastructure, coordination is essential:

• International Cooperation: Knowledge sharing and cooperation among international competition authorities can help build expertise and ensure a coordinated, timely response to market shifts.
• Domestic Regulatory Alignment: Competition authorities should coordinate with other domestic bodies, such as energy and digital regulators. This ensures that bottlenecks in one part of the value chain—such as access to electricity or specialized chips—do not undermine competition in other segments.

The Challenge of Timing

The source warns that policymakers face a critical timing dilemma: while premature action might inadvertently hamper innovation, late action risks allowing market concentration to become entrenched, which would make remedial steps significantly harder to implement in the future. Consequently, sustained vigilance and a proactive stance are required to secure long-term competitive outcomes.