The AI economy is facing a multitude of challenges, and five experts from the industry shed light on the critical issues that could impact its future. The panel discussion at the Milken Global Conference in Beverly Hills revealed a complex landscape where physical limitations, energy constraints, and geopolitical considerations are shaping the trajectory of AI development.
One of the key bottlenecks, as highlighted by Christophe Fouquet, CEO of ASML, is the supply of chips. Despite significant efforts, the market is expected to be supply-limited for the next few years, impacting hyperscalers like Google, Microsoft, Amazon, and Meta. This shortage is not just about the number of chips but also the quality and performance required for advanced AI applications.
Francis deSouza, COO of Google Cloud, emphasized the growing demand for AI infrastructure. Google Cloud's revenue crossed $20 billion in the last quarter, with a 63% growth rate and a nearly doubled backlog. This surge in demand underscores the industry's need for more efficient and scalable solutions.
Qasar Younis, CEO of Applied Intuition, drew attention to the data collection challenge. While silicon is not the primary bottleneck, the data required to train AI models is. Applied Intuition focuses on autonomy systems for various industries, and Younis emphasizes the importance of gathering real-world data, which cannot be fully replicated through synthetic simulations.
The energy problem is another critical issue. DeSouza mentioned Google's exploration of data centers in space to address energy constraints. However, the challenges of cooling in a vacuum and the need for efficient integration of AI components are significant. Fouquet and deSouza both stressed the importance of energy efficiency, with deSouza citing Google's strategy of co-engineering its AI stack for optimal performance.
Eve Bodnia, a quantum physicist and founder of Logical Intelligence, introduced a different approach with energy-based models (EBMs). These models aim to understand the underlying rules of data, mimicking the human brain's process. Bodnia's models, with 200 million parameters, run thousands of times faster than leading LLMs and can update knowledge as data changes, making them suitable for chip design and robotics.
Dimitry Shevelenko, Chief Business Officer of Perplexity, discussed the evolution of their search product into a 'digital worker.' Perplexity Computer is designed to assist knowledge workers, raising questions about control and security. Shevelenko emphasized the importance of granularity in permissions, ensuring that agents can be directed while maintaining security and trust.
The geopolitical implications of physical AI were highlighted by Younis. He noted that national sovereignty is at stake with physical AI, unlike in the digital realm. Governments are increasingly concerned about the control and safety of systems operating within their borders, especially in industries like autonomous vehicles and defense. Fouquet added that China's AI progress is constrained below the model layer due to a lack of access to EUV lithography.
The discussion also addressed the impact of AI on the next generation's critical thinking. While deSouza and Shevelenko offered optimistic perspectives, emphasizing the potential for addressing complex global challenges, Younis drew a distinction between knowledge work and physical labor. He argued that physical AI is filling labor gaps in industries like farming and mining, rather than displacing workers.
In conclusion, the AI economy is navigating a complex landscape of physical limitations, energy constraints, and geopolitical considerations. The experts' insights highlight the need for innovative solutions, efficient integration, and a nuanced understanding of the industry's impact on society and the future workforce.
