“Demis Hassabis identifies access to large-scale compute as the primary bottleneck for AGI development, required for both training larger models and conducting algorithmic experiments.”