The rise of artificial intelligence (AI) hinges on a multifaceted supply chain, according to a Bank for International Settlements (BIS) report.
Spanning five critical layers—hardware, cloud infrastructure, training data, foundation models, and AI applications—this ecosystem is shaped by rapid technological change, high fixed costs, economies of scale, network effects, and strategic maneuvers by dominant firms.
The report, authored by Fernando Restoy and Raihan Zamil, explores the market dynamics of each layer, the growing dominance of big tech, and the implications for competition, innovation, and systemic stability.
The hardware layer, encompassing chips like GPUs and TPUs, is the foundation of AI’s computational power.
Companies like Nvidia dominate due to high fixed costs and economies of scale, with the top three firms controlling over 80% of the GPU market.
This concentration raises concerns about supply chain resilience, especially amid geopolitical tensions affecting semiconductor production in East Asia.
Strategic behavior, such as exclusive supplier agreements, further entrenches these players.
Cloud infrastructure, the second layer, provides the computational backbone for AI development.
Hyperscalers—Amazon Web Services (AWS), Microsoft Azure, and Google Cloud—command 65% of the global market, leveraging vast data centers and network effects.
High upfront costs deter new entrants, while vertical integration with their own AI services amplifies their influence.
The report warns that this oligopoly could stifle competition and innovation if smaller firms struggle to access affordable computing resources.
Training data, the third layer, fuels AI model accuracy.
Big tech firms like Alphabet and Meta hoard proprietary datasets from their platforms, creating a “data moat” that disadvantages startups.
The market for third-party data providers exists but is fragmented, with quality and ethical sourcing (e.g., privacy concerns) posing challenges. Network effects amplify the advantage of firms with large, real-time data flows, reinforcing market power.
Foundation models—large-scale AI systems like GPT or BERT—form the fourth layer.
Developing these requires immense resources: OpenAI’s GPT-3, for instance, cost $10 million to train.
Firms like OpenAI (backed by Microsoft), Google, and Anthropic dominate, benefiting from economies of scale and first-mover advantages.
The report notes a trend toward closed-source models, limiting access for smaller innovators and raising barriers to entry.
The final layer, AI applications, integrates models into consumer-facing tools (for instance, chatbots or recommendation engines).
Here, big tech’s reach extends through acquisitions (e.g., Microsoft’s GitHub purchase) and partnerships, though niche startups persist in specialized sectors like healthcare.
Competition is fierce, but network effects favor incumbents with established user bases.
Big tech’s expanding footprint across all layers—via vertical integration and strategic investments—raises red flags.
The report highlights risks to consumer choice, as bundled services lock users into ecosystems, and to innovation, as startups face resource squeezes.
Operational resilience is threatened by reliance on concentrated suppliers, while cybersecurity and financial stability face heightened risks from interconnected failures—imagine a cloud outage crippling AI-driven banking.
The BIS urges policymakers to address these dynamics.
As noted in the update from BIS, options include antitrust scrutiny, open data initiatives, and resilience standards.
As AI reshapes economies, its supply chain’s structure will determine whether it fosters broad progress or entrenches a tech oligarchy.