Title page
Contents
Executive summary 1
1. Artificial intelligence competition concerns 2
2. Market entry barriers and competition along the AI model value chain 3
2.1. Competition between AI models 3
2.2. Fixed computing costs as a barrier to market entry 4
2.3. Access to model training data 6
2.4. Access to intermediate model deployers and end users 7
3. Policy responses to bottlenecks in the AI value chain 8
3.1. Bottlenecks in upstream AI model inputs markets 8
3.2. Bottlenecks in downstream markets for model deployers and users 9
4. Conclusions 10
References 12
Figure 1. Foundation models by access type 3
Figure 2. Estimate training costs and computing requirements for selected AI models 5
Figure 3. Collaboration agreements between big tech and smaller AI start-ups 6
Figure 4. The collaboration-competition model between AI start-ups and GAMMANs 8