During the Milken Global Conference in Beverly Hills, five influential figures from the AI supply chain gathered to discuss critical challenges facing the industry. Their conversation covered a range of topics, including chip shortages, energy constraints, and the potential flaws in the foundational architecture of AI technology.
Key Participants:
- Christophe Fouquet, CEO of ASML
- Francis deSouza, COO of Google Cloud
- Qasar Younis, co-founder and CEO of Applied Intuition
- Dimitry Shevelenko, chief business officer of Perplexity
- Eve Bodnia, founder of Logical Intelligence
Real Bottlenecks in Production
The discussion revealed that the AI boom is encountering significant physical limitations. Fouquet emphasized that despite advancements in chip manufacturing, the supply will remain constrained for the next few years, impacting major tech companies like Google and Microsoft.
DeSouza added that Google Cloud is experiencing rapid growth, with its revenue surpassing $20 billion last quarter. However, the backlog of committed revenue has also surged, indicating a strong demand that is not being met.
Younis pointed out that for Applied Intuition, the primary bottleneck is not silicon but the real-world data required for training autonomy systems. He stressed the importance of gathering data from actual environments rather than relying solely on synthetic simulations.
Energy Constraints
Energy availability poses another significant challenge. DeSouza mentioned that Google is investigating the feasibility of space-based data centers to address energy limitations, although the engineering challenges in a vacuum environment are considerable.
Fouquet echoed this sentiment, noting that increased computation demands will inevitably lead to higher energy costs, which the industry must address.
Innovative Approaches to Intelligence
Bodnia is exploring alternative AI architectures through energy-based models (EBMs), which aim to mimic human cognitive processes more closely than traditional models. Her approach emphasizes understanding the underlying rules of data rather than merely predicting sequences, potentially offering advantages in fields requiring physical interaction.
Trust and Control in AI Systems
Shevelenko discussed the evolution of Perplexity's offerings, which now include a digital worker designed to assist knowledge workers. This raises important questions about control and security, which he addressed by advocating for granular permissions that allow enterprise administrators to manage agent capabilities effectively.
Geopolitical Implications
Younis highlighted the intertwining of physical AI and national sovereignty, noting that governments are increasingly wary of foreign-controlled AI systems operating within their borders. Fouquet added that while China is making strides in AI, its progress is limited by access to advanced semiconductor manufacturing technologies.
Future Generations and AI
As the conversation turned to the potential impact of AI on future generations, the panelists expressed optimism. DeSouza pointed out that advanced tools could help tackle complex global challenges, while Shevelenko noted the democratization of technology, enabling individuals to launch initiatives independently.
Younis emphasized that physical AI is not replacing jobs but rather filling existing labor shortages in sectors like agriculture and mining, where there is a growing reluctance to engage in physical labor.
