Five architects behind the AI ​​economy explain where the wheels come off

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 to this editor about everything from chip shortages to orbital data centers to the possibility that the entire architecture underlying the technology is wrong.

On stage with TechCrunch: Christophe Fouquet, CEO of ASML, the Dutch company that has a monopoly on the extreme ultraviolet lithography machines without which modern chips would not exist; Francis deSouza, COO of Google Cloud, who is overseeing one of the largest infrastructure efforts in the company’s history; Qasar Younis, co-founder and CEO of Applied Intuition, a $15 billion physical AI company that started in simulation and has since moved into defense; Dimitry Shevelenko, Chief Business Officer of Perplexity, the AI-native search-to-agents company; and Eve Bodnia, a quantum physicist who left academia to challenge the fundamental architecture most of the AI ​​industry takes for granted at its inception, Logical Intelligence. (Meta’s former chief AI researcher, Yan LeCun, signed on as founding chairman of its technical research council 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 limitations begin further down the stack than many may realize. Fouquet was the first to say so, describing a “tremendous acceleration of chip manufacturing” while expressing his “strong belief” that despite all the effort, “in the next two, three, maybe five years, the market will be limited supply,” meaning the hyperscalers — Google, Microsoft, Amazon, Meta — aren’t going to get all the chips they’re paying for.

Highlighting just how big—and how fast growing—a problem this is, DeSouza reminded the audience that Google Cloud’s revenue passed $20 billion last quarter, growing 63%, while its backlog — the revenue committed but not yet delivered — nearly doubled in a single quarter, from $250 billion to $460 billion. “The demand is real,” he said with impressive composure.

For Younis, the limitation primarily comes from elsewhere. Applied Intuition builds autonomous systems for cars, trucks, drones, mining equipment and defense vehicles, and his bottleneck isn’t silicon — it’s data that can only be collected by sending machines out into the real world and seeing what happens. “You have to find it from the real world,” he said, and no amount of synthetic simulation will completely close that gap. “It will be a long time before you can fully train models that run on the physical world synthetically.”

The energy problem is also real

If chips are the first bottleneck, energy is the one looming behind. DeSouza confirmed that Google is exploring data centers in space as a serious response to energy constraints. “You get access to more abundant energy,” he noted. Of course, even in orbit it is not easy. The space DeSouza observed is a vacuum, so it eliminates convection, leaving radiation as the only way to shed heat into the surrounding environment (a much slower and more difficult process to engineer than the air and liquid cooling systems that data centers rely on today). But the company still treats it as a legitimate avenue.

The deeper argument de Souza made, somewhat surprisingly, was about efficiency through integration. Google’s strategy of developing its full AI stack—from custom TPU chips to models and agents—pays dividends in flops per watts (more calculation per unit of energy) that a company buying off-the-shelf components simply can’t replicate, he suggested. “Running Gemini on TPUs is much more energy efficient than any other configuration,” because chip designers know what’s going into the model before it ships, he said.

Fouquet made a similar point later in the discussion. “Nothing can be priceless,” he said. The industry is in a strange moment right now, investing extraordinary amounts of capital, driven by strategic necessity. But more computation means more energy, and more energy comes at a price.

Another form of intelligence

While the rest of the industry discusses scale, architecture and inference efficiency within the large language model paradigm, Bodnia is building something very different.

Her company, Logical Intelligence, is built on so-called energy-based models (EBMs), a class of AI that doesn’t predict the next token in a sequence, but instead tries to understand the rules behind data, in a way she claims is closer to how the human brain actually works. “Language is a user interface between my brain and yours,” she said. “Reasoning itself is not attached to any language.”

Her largest model runs for 200 million parameters—compared to hundreds of billions in leading LLMs—and she claims it runs thousands of times faster. More importantly, it is designed to update its knowledge as data changes, rather than requiring retraining from scratch.

For chip design, robotics, and other domains where a system needs to understand physical rules rather than linguistic patterns, she argues that EBMs are the more natural fit. “When you drive a car, you don’t look for patterns in any language. You look around, understand the rules of the world around you, and make a decision.” It’s an interesting argument, and one that will likely attract more attention in the coming months as the AI ​​field begins to question whether scale alone is sufficient.

Agents, railings and trust

Shevelenko spent much of the conversation explaining how Perplexity has evolved from a search product into what it now calls a “digital employee.” Perplexity Computer, its newest offering, is designed not as a tool a knowledge worker uses, but as a staff that a knowledge worker manages. “Every day you wake up and you have a hundred employees on your team,” he said of the opportunity. “What will you do to make the most of it?”

It’s a compelling pitch; it also raises obvious questions about control, so I asked them. His answer was: granularity. Enterprise administrators can specify not only which connectors and tools an agent can access, but whether those permissions are read-only or read-write—a distinction that matters enormously when agents act on enterprise systems. When Comet, Perplexity’s computing agent, takes action on a user’s behalf, it presents a plan and asks for approval first. Some users find the friction annoying, Shevelenko said, but he said he considers it important, especially after joining the board of Lazard, where he has found himself unexpectedly sympathetic to the conservative instincts of a CISO protecting a 180-year-old brand built entirely on client trust. “Granularity is the foundation of good security hygiene,” he said.

Sovereignty, not just security

Younis offered what may have been the panel’s most geopolitically charged observation, which is that physical AI and national sovereignty are entangled in ways that pure digital AI never was.

The Internet initially spread as an American technology and only faced pushback at the application layer—the Ubers and DoorDashes—when offline consequences became apparent. Physical AI is different. Autonomous vehicles, defense drones, mining equipment, agricultural machinery—these are manifesting in the real world in ways governments cannot ignore, raising questions about security, data collection, and who ultimately controls systems operating within a nation’s borders. “Almost consistently, all countries are saying: We don’t want this intelligence in physical form within our borders, controlled by another country.” Fewer nations, he told the crowd, can currently field a robot axis than possess nuclear weapons.

Fouquet framed it a little differently. China’s AI progress is real — DeepSeek’s release earlier this year sent something close to panic through parts of the industry — but that progress is limited below the model layer. Without access to EUV lithography, Chinese chip makers can’t produce the most advanced semiconductors, and models built on legacy hardware operate at a compound disadvantage, no matter how good the software gets. “Today, in the United States, you have the data, you have computer access, you have the chips, you have the talent. China is doing a very good job at the top of the stack, but is missing some elements below,” Fouquet said.

The generational question

Near the end of our panel, someone in the audience asked the obvious uncomfortable question: will all this affect the next generation’s ability to think critically?

The responses were upbeat, as you would expect from people who have staked their careers on this technology. DeSouza immediately pointed to the scale of problems that more powerful tools could finally let humanity solve. Think of neurological diseases whose biological mechanisms we do not yet understand, the removal of greenhouse gases and grid infrastructure that has been delayed for decades. “This should release us to the next level of creativity,” he said.

Shevelenko made a more pragmatic point: The startup task may be disappearing, but the ability to start something independently has never been more accessible. “[For] anyone who has Perplexity Computer. . . the limitation is your own curiosity and drive.”

Younis drew the sharpest distinction between knowledge work and physical work. He pointed to the fact that the average American farmer is 58 years old, and that labor shortages in mining, long-haul trucking and agriculture are chronic and growing — not because wages are too low, but because people don’t want those jobs. In these domains, physical AI does not displace willing workers. It fills a void that already exists and only seems to get deeper from here.