国际清算银行-AI供应链（英）-2025.3_19页_408kb

AI Supply Chain Summary

Core Content

This paper explores the structure and dynamics of the artificial intelligence (AI) supply chain, which consists of five key layers: hardware, cloud infrastructure, training data, foundation models, and AI applications. It examines the market structure of each layer, the economic forces that shape them, and the growing influence of big tech companies across the entire supply chain. The analysis also highlights the challenges posed by a concentrated AI supply chain and the policy implications for regulators.

Main Layers of the AI Supply Chain

• Hardware: Specialised microprocessors, particularly GPUs, are critical for AI computations. Nvidia dominates the market with over 90% market share and high gross margins.
• Cloud Infrastructure: The cloud computing market is dominated by AWS (31%), Microsoft Azure (24%), and Google Cloud Platform (11%) globally, with even higher concentrations in regions like the EU and Asia-Pacific.
• Training Data: Training data are vast and include text, images, and audio. Large firms can leverage proprietary data and strategic acquisitions to secure new data sources.
• Foundation Models: These are large, pre-trained models that can be adapted for various uses. The market is dominated by a few firms, including OpenAI, Google DeepMind, Anthropic, and Meta.
• AI Applications: These are user-facing tools such as ChatGPT, Claude, Gemini, DALL-E, and GitHub Copilot, which rely on foundation models and cloud infrastructure.

Key Economic Forces

• High fixed costs and economies of scale/scope are common in the hardware and cloud layers, creating barriers to entry and favoring large firms.
• Network effects and consumer inertia contribute to market concentration, especially in cloud and hardware markets.
• Switching costs are significant, both exogenous (due to technical differences) and endogenous (due to software lock-in).
• Egress fees and exclusive partnerships further entrench the dominance of big tech firms in cloud infrastructure.
• Data feedback loops can enhance the competitive advantage of large firms, but may also lead to diminishing returns in some cases.

Role of Big Tech Companies

• Big tech firms are vertically integrating across all layers of the AI supply chain, leveraging their control over data, cloud infrastructure, and computational resources.
• They are investing heavily in AI, with big techs accounting for 33% of total capital raised by AI firms in 2023 and 67% of capital raised by generative AI firms.
• These firms are also forming strategic partnerships and investments with AI startups, often imposing exclusive conditions to ensure continued dominance.
• By controlling training data and AI chips, big techs can reduce costs and improve efficiency in developing and deploying AI models.
• The data gravity effect suggests that big techs can continuously improve their AI models by using their own data and computational resources, creating a self-reinforcing loop.

Challenges of a Concentrated AI Supply Chain

• Consumer choice and innovation may be limited due to the dominance of a few firms.
• Operational resilience and cybersecurity could be compromised if the supply chain becomes too concentrated.
• Financial stability is at risk due to the potential for systemic risk and concentration in key technologies.
• Market tipping is a concern, as early entrants may gain significant advantages, reducing competition and increasing the likelihood of a "winner takes all" scenario.

Policy Considerations

• Regulatory oversight is essential to address the concentration in the AI supply chain, especially given the cross-border nature of the AI ecosystem.
• International and domestic cooperation among regulatory bodies will be key to ensuring fair competition and mitigating risks to financial stability and cybersecurity.
• There is a need for policy frameworks that address the unique characteristics of each layer of the AI supply chain, including data privacy, market access, and innovation incentives.

Conclusion

The AI supply chain is becoming increasingly complex and concentrated, with big tech firms playing a central role. The dominance of these firms across multiple layers poses challenges to competition, innovation, and financial stability. Effective regulatory responses are necessary to ensure a balanced and resilient AI ecosystem.