Back to Blog
May 7, 2026 • Connie Loizos

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

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

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

At the Milken Global Conference in Beverly Hills, five leaders spanning the entire AI supply chain—from chip manufacturing to autonomous systems to AI-native search—sat down with TechCrunch to discuss the critical bottlenecks, architectural debates, and geopolitical tensions shaping the AI economy.

The Panel

  • Christophe Fouquet, CEO of ASML (monopoly holder of extreme ultraviolet lithography machines)
  • Francis deSouza, COO of Google Cloud
  • Qasar Younis, Co-founder and CEO of Applied Intuition ($15B physical AI company)
  • Dimitry Shevelenko, Chief Business Officer of Perplexity
  • Eve Bodnia, Co-founder of Logical Intelligence (quantum physicist challenging LLM architecture)

Key Insights

The Bottlenecks Are Real

Chip Supply Constraints:

  • ASML's Fouquet predicts the market will remain "supply limited" for the next 2-5 years
  • Hyperscalers (Google, Microsoft, Amazon, Meta) won't get all the chips they're paying for
  • Google Cloud's backlog nearly doubled in a single quarter: from $250B to $460B

Physical AI Data Gap:

  • Applied Intuition's Younis highlighted that autonomous systems require real-world data that can't be fully replicated synthetically
  • "You have to find it from the real world"—no amount of simulation closes the gap completely

The Energy Problem

Google Exploring Space Data Centers:

  • Google is seriously exploring orbital data centers as a response to energy constraints
  • Space offers more abundant energy but presents cooling challenges (only radiation works in vacuum)
  • DeSouza emphasized efficiency through vertical integration: co-engineering the full AI stack (custom TPU chips through models) delivers superior flops per watt

Cost Reality:

  • "Nothing can be priceless," warned Fouquet
  • More compute = more energy = unavoidable costs

A Different Kind of Intelligence

Energy-Based Models (EBMs) vs. LLMs:

  • Logical Intelligence's Bodnia is building AI using energy-based models instead of large language models
  • EBMs understand underlying rules rather than predict next tokens
  • Her largest model: 200 million parameters (vs. hundreds of billions in leading LLMs)
  • Claims to run thousands of times faster
  • Designed to update knowledge as data changes without retraining from scratch

Key Insight:

  • "Language is a user interface between my brain and yours. The reasoning itself is not attached to any language."
  • Better suited for physical AI domains (chip design, robotics) where systems need to grasp physical rules, not linguistic patterns

Agents, Guardrails, and Trust

Perplexity's Evolution:

  • Shifted from search product to "digital worker"
  • Perplexity Computer designed as staff that knowledge workers direct
  • "Every day you wake up and you have a hundred staff on your team"

Security by Design:

  • Granular permission controls: read-only vs. read-write access
  • Comet (computer-use agent) presents plans and asks for approval before taking actions
  • "Granularity is the bedrock of good security hygiene," says Shevelenko

Sovereignty, Not Just Safety

Physical AI = Geopolitical Issue:

  • Younis: "Almost consistently, every country is saying: we don't want this intelligence in a physical form in our borders, controlled by another country."
  • Fewer nations can field a robotaxi than possess nuclear weapons
  • Physical AI (autonomous vehicles, defense drones, mining equipment) raises sovereignty concerns in ways purely digital AI never did

China's AI Constraint:

  • Fouquet: China excels at the software layer but lacks access to EUV lithography
  • Cannot manufacture the most advanced semiconductors
  • "Today, in the United States, you have the data, you have the computing access, you have the chips, you have the talent. China does a very good job on the top of the stack, but is lacking some elements below."

The Generation Question

Optimistic Perspectives:

  • DeSouza: More powerful tools will let humanity address neurological diseases, greenhouse gas removal, grid infrastructure
  • Shevelenko: Entry-level jobs may disappear, but ability to launch something independently has never been more accessible
  • Younis: Physical AI fills labor voids in agriculture (average farmer is 58), mining, and trucking—jobs people don't want, not jobs being displaced

Takeaways

  • Supply-side constraints will define AI progress for the next 2-5 years
  • Energy is the next major bottleneck after chips
  • Alternative architectures (EBMs) may challenge the LLM paradigm for physical AI
  • Geopolitical sovereignty concerns are reshaping the physical AI landscape
  • Granular security controls are essential for enterprise AI agent deployment