The Evolutionary Dynamics of the Artificial Intelligence Ecosystem
Card Grid View — Jacobides, Brusoni & Candelon (2021)
1. Three Stages of AI Development
- AI Enablement
- Foundational infrastructure: cloud computing, hardware, data storage
- Big Tech firms (AWS, Google, Microsoft) dominate this layer
- Includes data centers, chips, edge devices
- AI Production
- Libraries, frameworks, ML models, and tools
- TensorFlow, PyTorch as key examples
- Often open-source to drive ecosystem adoption
- AI Consumption
- End-user applications and integration into products
- AI takers use AI without building it themselves
2. Big Tech Dominance
- Key finding
- Small number of Big Tech firms dominate across all three AI layers
- Control infrastructure, production tools, and consumption channels
- Why Big Tech dominates
- Massive data advantages create feedback loops
- Cloud infrastructure requires enormous capital investment
- Network effects and economies of scale
- Open-source strategies lock in developers and users
- Implication
- Power and inequality concentrated in few firms
3. AI as a Unique General-Purpose Technology
- AI is not a standard GPT
- Unlike steam engine or electricity, AI improves itself through data
- Self-reinforcing feedback loops create winner-take-all dynamics
- No typical underinvestment problem — private incentives are massive
- Why this matters
- Public subsidies may not be needed for AI development
- Policy should focus on distribution, not just investment
- Governments should not treat AI as needing basic R&D support
4. China vs US: Different AI Ecosystems
- China's AI development
- "Sputnik moment" when AlphaGo defeated Ke Jie (2017)
- Massive national AI strategy launched
- Big Tech firms (Baidu, Alibaba, Tencent) control infrastructure
- Government plays active coordinating role
- US AI development
- Market-driven, private sector led
- Big Tech (Google, Amazon, Microsoft) dominates
- More fragmented government approach
- Key difference
- China: state-guided ecosystem; US: market-driven
5. AI Takers, Makers & Shapers
- AI Makers
- Build AI models, libraries, and infrastructure
- Includes Big Tech and specialized AI companies
- Control the core technology stack
- AI Takers
- Adopt and integrate AI into their products/services
- Only 11% see significant financial benefits
- Success requires data, talent, and organizational change
- AI Shapers / Traders
- Buy and resell AI solutions with added services
- Bundle, brand, and customize without building core AI
- Serve as intermediaries
6. Data as Strategic Advantage
- Why data is critical
- High-quality data is the fundamental bottleneck for AI
- Data enables model training, validation, and improvement
- Feedback loops: more users → more data → better models
- Who benefits from AI adoption
- Firms with large proprietary data sets
- Companies that integrate AI into core operations
- Organizations with strong data infrastructure
- Uneven adoption
- AI adoption is highly skewed: few firms capture most value
7. Schumpeterian Innovation & Policy
- Mark I vs Mark II patterns
- Mark I: creative destruction by entrepreneurs
- Mark II: creative accumulation by incumbents
- AI shows Mark II dynamics: incumbents accumulate advantage
- Policy implications
- Don't assume AI just needs general public support
- Focus on antitrust, data access, and competition policy
- Address power imbalances across the ecosystem
- Promote complementors and reduce barriers to entry