Five Architects of the AI Economy Explain Where the Wheels Are Coming Off

Summary

At the Milken Institute Global Conference in Beverly Hills, five pivotal figures spanning the AI supply chain—ASML CEO Christophe Fouquet, Google Cloud COO Francis deSouza, Applied Intuition CEO Qasar Younis, Perplexity CBO Dmitry Shevelenko, and Logical Intelligence CEO Eve Bodnia—revealed critical bottlenecks threatening the industry's trajectory. Despite massive infrastructure investments, the AI boom has hit hard physical limits that extend far deeper than most realize.

The semiconductor shortage is immediate and severe. ASML's Fouquet stated bluntly that despite accelerating chip manufacturing, "for the next two, three, maybe five years, the market will be supply limited." Google Cloud's scale amplifies the problem: $20 billion in quarterly revenue with 63% growth, yet a backlog that nearly doubled from $250 billion to $460 billion in a single quarter—unmet demand dwarfing actual shipments. For physical AI companies like Applied Intuition, the constraint isn't silicon but real-world data; synthetic simulation cannot fully replicate the data needed to train autonomous systems for real-world deployment.

Energy has emerged as the even more pressing long-term challenge. Google is exploring orbital data centers as a serious infrastructure response, with deSouza confirming this as a legitimate strategy despite unique engineering obstacles (space has no convection, requiring radiation cooling rather than air/liquid systems). Google's competitive advantage lies in its integrated approach—co-engineering custom TPU chips with Gemini models and agents delivers efficiency (flops per watt) that commodity-component buyers cannot match, suggesting vertical integration may become essential for operating at hyperscaler scale.

• Vertical integration (custom chips, proprietary models, and agents) provides decisive efficiency advantages over commodity-based competitors.
• Some founders (e.g., Logical Intelligence) are challenging whether the foundational AI architecture itself may be architecturally flawed.

Editorial Opinion

The candid admissions from infrastructure leaders that physical bottlenecks—not algorithmic innovation—will define AI's next decade is a watershed moment. The era of capital-driven scaling has ended; the industry is now bound by geology, manufacturing capacity, and thermodynamics. Most tellingly, the asymmetry between integrated vertically-engineered stacks (Google's TPU-Gemini approach) and off-the-shelf buyers hints at a coming consolidation: only a handful of deep-pocketed incumbents with full-stack engineering capabilities may be able to compete at scale, potentially accelerating AI's concentration even as the technology appears to democratize.