Five architects of the AI economy explain where the wheels are coming off

At the Milken Global Conference, five key architects of the modern AI stack convened to deliver a dose of reality: the industry's exponential growth is colliding with hard physical limits. Leaders from ASML, Google Cloud, Applied Intuition, Perplexity, and Logical Intelligence warned that critical bottlenecks in chip supply, energy, and real-world data are already defining the next several years of development. The consensus suggests that despite unprecedented investment, the ability to build and deploy advanced AI is fundamentally constrained not by software, but by tangible, real-world resources.

The Hard Constraints on AI Scale

The operational details reveal a supply chain under immense pressure. Christophe Fouquet, CEO of chip-machine monopoly ASML, stated he has a "strong belief" that the market will remain "supply limited" for the next two to five years, meaning hyperscalers will not get all the silicon they order. Underscoring this demand, Google Cloud COO Francis deSouza noted his division's backlog nearly doubled in a single quarter to $460 billion. The bottlenecks extend beyond hardware:

• Silicon Scarcity: The manufacturing capacity for cutting-edge chips cannot keep pace with surging demand from hyperscalers like Google, Microsoft, and Amazon.
• Energy Demands: Power constraints are so significant that Google is actively exploring orbital data centers to access space-based solar energy, despite major thermal engineering hurdles.
• Data Bottlenecks: For physical AI systems, Applied Intuition CEO Qasar Younis explained that the primary constraint is gathering real-world interaction data, which he says cannot be fully replicated with synthetic simulation.

Sovereignty, Security, and New Architectures

These physical limitations are forcing a strategic re-evaluation across the ecosystem. On a geopolitical level, Younis observed that nations are increasingly viewing physical AI as an issue of national sovereignty, reluctant to allow foreign-controlled autonomous systems to operate within their borders. This is compounded by hardware access; ASML's Fouquet noted that China's inability to acquire EUV lithography machines creates a durable disadvantage below the model layer. In response to scaling costs, startups like Eve Bodnia's Logical Intelligence are pioneering energy-based models (EBMs) that are thousands of times faster and smaller than LLMs. Meanwhile, companies like Perplexity are focused on the practical application layer, building agent-based "digital workers" with granular, CISO-friendly security controls to build trust within the enterprise.

While the AI industry races to scale large models, the emerging strategic imperatives are solving fundamental constraints through full-stack efficiency, pioneering less resource-intensive architectures, and navigating the complex geopolitics of physical AI deployment.