Five architects of the AI economy explain where the wheels are coming off
Source: TechCrunch
TechCrunch Live at Milken Global – Beverly Hills
Earlier this week, five people who touch every layer of the AI supply chain sat down at the Milken Global Conference in Beverly Hills, where they talked with this editor about everything from chip shortages to orbital data centers to the possibility that the whole architecture that undergirds the tech is wrong.
On stage with TechCrunch
| Speaker | Role |
|---|---|
| Christophe Fouquet | CEO, ASML – the Dutch company that holds a monopoly on extreme‑ultraviolet lithography machines (the machines without which modern chips would not exist) |
| Francis deSouza | COO, Google Cloud – overseeing one of the biggest infrastructure bets in corporate history |
| Qasar Younis | Co‑founder & CEO, Applied Intuition – a $15 B physical‑AI company that started in simulation and has since moved into defense |
| Dimitry Shevelenko | Chief Business Officer, Perplexity – the AI‑native “search‑to‑agents” company |
| Eve Bodnia | Quantum physicist‑turned‑entrepreneur, Logical Intelligence – challenging the foundational architecture most of the AI industry takes for granted (Meta’s former chief AI scientist, Yan LeCun, signed on as founding chair of its technical research board earlier this year) |
Here’s what the five had to say
The bottlenecks are real
The AI boom is running into hard physical limits, and the constraints begin further down the stack than many may realize.
- Fouquet described a “huge acceleration of chips manufacturing,” but expressed his “strong belief” that for the next two, three, maybe five years, the market will be supply‑limited. In other words, the hyperscalers — Google, Microsoft, Amazon, Meta — aren’t going to get all the chips they’re paying for, full stop.
- deSouza highlighted the scale of the issue, noting that Google Cloud’s revenue crossed $20 B last quarter, up 63 %, while its backlog nearly doubled in a single quarter, from $250 B to $460 B. “The demand is real,” he said.
- For Younis, the constraint is primarily data. Applied Intuition builds autonomy systems for cars, trucks, drones, mining equipment and defense vehicles, and his bottleneck isn’t silicon—it’s the real‑world data needed to train models. “You have to find it from the real world,” he said, adding that synthetic simulation can’t fully close that gap.
The energy problem is also real
If chips are the first bottleneck, energy is the one looming behind them.
- deSouza confirmed that Google is exploring data centers in space as a response to energy constraints. He noted that in orbit, convection is absent, leaving radiation as the only way to shed heat—a slower, harder‑to‑engineer process than today’s air‑ and liquid‑cooling systems.
- He also emphasized efficiency through integration: Google’s co‑engineered AI stack (custom TPU chips → models → agents) yields higher flops per watt than off‑the‑shelf solutions. “Running Gemini on TPUs is much more energy‑efficient than any other configuration,” he said.
- Fouquet echoed this, warning that “nothing can be priceless.” More compute means more energy, and energy has a price.
A different kind of intelligence
While most of the industry focuses on scaling large‑language models, Bodnia is building something very different.
- Logical Intelligence uses energy‑based models (EBMs), which aim to understand the rules underlying data rather than predicting the next token.
“Language is a user interface between my brain and yours,” she said. “The reasoning itself is not attached to any language.”
- Her largest model has 200 M parameters (versus hundreds of billions in leading LLMs) and runs thousands of times faster. It can update its knowledge as data changes, avoiding full retraining.
- EBMs are suited for domains like chip design and robotics, where grasping physical rules is more important than linguistic patterns.
Agents, guardrails, and trust
Shevelenko explained how Perplexity has evolved from a search product into a “digital worker.”
- Perplexity Computer, the newest offering, is designed not as a tool a knowledge worker uses, but as a staff that a knowledge worker directs.
“Every day you wake up and you have a hundred staff on your team,” he said.
He highlighted three pillars for responsible agent deployment:
- Guardrails – built‑in safety constraints.
- Transparency – clear provenance of data and reasoning steps.
- Trust – continuous monitoring and feedback loops for correction or override.
Event details
| Location | Dates |
|---|---|
| San Francisco, CA | October 13‑15, 2026 |
Granular Control
“What are you going to do to make the most of it?”
Enterprises can specify not just which connectors and tools an agent can access, but also whether those permissions are read‑only or read‑write. When Comet, Perplexity’s computer‑use agent, takes actions on a user’s behalf, it presents a plan and asks for approval first. Shevelenko admitted some users find the friction annoying, but he considers it essential—especially after joining the board of Lazard, where he sympathizes with the conservative instincts of a CISO protecting a 180‑year‑old brand.
“Granularity is the bedrock of good security hygiene,” he said.
Sovereignty, not just safety
Younis warned that physical AI and national sovereignty are now entangled in ways digital AI never was.
“Almost consistently, every country is saying: we don’t want this intelligence in a physical form in our borders, controlled by another country.”
—Younis
Physical AI manifests in autonomous vehicles, defense drones, mining equipment, and agricultural machines—systems governments can’t ignore. Fewer nations can field a robotaxi than possess nuclear weapons.
Fouquet added that China’s AI progress is constrained by lack of EUV lithography. Without access to the most advanced chips, Chinese models run on older hardware, putting them at a compounding disadvantage.
“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.”
—Fouquet
The generation question
An audience member asked whether all this would impact the next generation’s capacity for critical thinking.
- deSouza pointed to the scale of problems more powerful tools could address—neurological diseases, greenhouse‑gas removal, and aging grid infrastructure.
“This should unleash us to the next level of creativity,” he said.
- Shevelenko noted that while entry‑level jobs may disappear, the ability to launch something independently has never been more accessible.
“For anybody who has Perplexity Computer… the constraint is your own curiosity and agency.”
- Younis highlighted labor shortages in physical sectors (farming, mining, trucking). Physical AI isn’t displacing willing workers; it’s filling a void that already exists and is likely to deepen.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.