Five architects of the AI ​​economy explain where the wheels are coming from

Earlier this week, five people involved in every aspect of AI sat down at the Milken Global Conference in Beverly Hills, where they talked to this editor about everything from the lack of chips to data centers to the possibility that the entire infrastructure that supports the technology is flawed.

On stage with TechCrunch: Christophe Fouquet, CEO of ASML, a Dutch company that manages ultraviolet lithography without which modern chips would not exist; Francis deSouza, COO of Google Cloud, who is overseeing one of the largest corporate initiatives in the company’s history; Qasar Younis, co-founder and CEO of Applied Intuition, a $15 billion AI company that started in simulation and moved into security; Dimitry Shevelenko, chief business officer of Perplexity, an AI-native search-to-agents company; and Eve Bodnia, a physicist who left academia to challenge the construction many AI companies take for granted in her startup, Logical Intelligence. (Meta AI’s former chief scientist, Yan LeCun, signed on as founding chair of its technology research group earlier this year.)

Here is what the five had to say:

Limitations are real

The development of AI is moving at a rapid pace, and the barriers are starting to get lower than many realize. Fouquet was the first to say this, describing the “great acceleration of the production of chips,” while expressing his “firm belief” that despite his efforts, “for two, three, maybe five years, the market will be limited,” meaning that the hyperscalers – Google, Microsoft, Amazon, Meta – will not get all the chips they are paying for, no.

DeSouza showed how big – and fast growing – this story is, reminding the audience that Google Cloud’s revenue exceeded $ 20 billion last quarter, a growth of 63%, while its balance – which was paid but not paid – almost doubled in one quarter, from $ 250 billion to $ 460 billion. “The requirements are real,” he said with impressive calm.

For Younis, the obstacle comes from elsewhere. Applied Intuition develops autonomous systems for cars, trucks, drones, mining equipment and defense vehicles, and its bottle is not silicon – it’s what one can collect by sending a machine into the real world and seeing what happens. “You have to get it from the real world,” he said, and no artificial simulation can bridge that gap. “It’s going to be a long time before you train all the species that go around.”

The power problem is also true

If the chips are the first obstacle, the power is the one that comes behind it. DeSouza confirmed that Google is looking at data centers in space as a major solution to the energy crisis. “You get a lot of energy,” he said. Of course, even in the process, it is not easy. DeSouza saw space as empty, thus eliminating motion, leaving radiation as the only way to heat the environment (a much cheaper and more complex method than the air and water cooling systems that data centers rely on today). But the company is still acting as an official channel.

The most profound argument de Souza made was, unsurprisingly, about efficiency through integration. Google’s way of making all of its AI hardware — from TPU chips to models and assistants — pays for units of flops per watt (multiple units of power) that an off-the-shelf hardware company can’t replicate, he said. “The Gemini run on TPUs is much easier than any other evolution,” because chipmakers know what’s coming to the prototype before it ships, he said.

Fouquet made a similar point later in the discussion. “Nothing is too precious,” he said. The industry is in a strange time right now, investing heavily, driven by the need for technology. But more compute means more power, and more power comes with a price.

Another kind of genius

While all the other companies argue about size, design, and technical expertise within the main language, Bodnia is creating something completely different.

His company, Logical Intelligence, is built on so-called dynamic models (EBMs), a group of AI that does not predict the next signal in a systematic way but instead tries to understand the rules underlying the data, in a way that he argues is close to how the human brain works. “Language is the communication channel between my brain and yours,” he said. “Imagination itself is not connected with any language.”

Its largest version runs 200 million sessions – compared to hundreds of billions for leading LLMs – and is said to run thousands of times faster. Most importantly, it is designed to update its knowledge as the data changes, rather than needing to be retrained from scratch.

In chip design, robotics and other areas where the system must understand physical laws rather than language, EBMs are said to be the natural fit. “When you’re driving a car, you’re not looking for the alphabet in any language. It’s an interesting debate and one that will likely attract a lot of attention in the coming months, as the field of AI begins to question whether scale alone is enough.”

Agents, security, and trust

Shevelenko spent much of the discussion explaining how Perplexity has evolved from a search engine into a so-called “digital worker.” The Perplexity Computer, his newest offering, is not designed as a tool that a connoisseur uses, but as a stick that a connoisseur controls. “Every day you wake up and you have a hundred employees on your team,” he said of the opportunity. “What are you doing to make the most of it?”

It is a compelling word; it also raises obvious questions about control, so I asked them. The answer was: granularity. Business managers can not only specify which connections and tools the agent can access, but whether the permissions are read-only or written – a distinction that becomes even more important when agents are working within companies. When Comet, Perplexity’s computer operator, acts on the user’s behalf, it issues a plan and asks for initial approval. Some users find the conflict irritating, Shevelenko said, but he said he feels it’s important, especially after joining the Lazard team, where he said he found himself unexpectedly empathizing with the CISO’s attitude to protect a 180-year-old brand built entirely on customer trust. “Granularity is the foundation of good hygiene,” he said.

Governance, not just security

Younis offered what was perhaps the group’s biggest observation, namely that physical AI and world governance are drawn in ways that digital AI has never been.

The Internet spread like an American technology and faced only a push in the utility sector – Ubers and DoorDashes – the results of the Internet appeared. Physical AI is different. Autonomous vehicles, defense drones, mining equipment, agricultural machines – these appear in the real world in ways that governments cannot ignore, raising questions about security, data collection, and those who control systems that operate across national borders. Almost always, every country says: we don’t want this intelligence in our borders, controlled by another country. Fewer countries, he told the crowd, are currently capable of developing nuclear weapons than possessing nuclear weapons.

Fouquet put it differently. China’s AI progress is real — the release of DeepSeek earlier this year sent something close to panic through some parts of the industry — but that progress is limited under the model sector. Without access to EUV labels, Chinese chipmakers can’t produce high-quality devices, and models built on old devices struggle to perform well no matter how good the software is. “Today, in the United States, you have data, you have access to computers, you have chips, you have skills. China does a very good job at the top of the stack, but it lacks some things below,” said Fouquet.

A generational question

Towards the end of our panel, one of the audience members asked an unsettlingly obvious question: will all of this affect the next generation’s ability to think critically?

The responses were positive, as you would expect from people who have dedicated their careers to this technology. DeSouza immediately showed the magnitude of the problems that superpowers could allow people to solve. Think neurological diseases we don’t understand, greenhouse gas emissions, and infrastructure that has been stalled for decades. “This should bring us to a new level of creativity,” he said.

Shevelenko made a very important point: entry work may be finished, but the ability to start something on their own has never been found. “(To) anyone with Computer Perplexity . . . the only limitation is your interest and organization.”

Younis made a sharp distinction between intellectual work and physical work. He said the average American farmer is 58 years old and that unemployment in mining, trucking, and agriculture is difficult and growing — not because wages are low, but because people don’t want the jobs. In those fields, physical AI does not replace dedicated workers. It fills the existing space and seems to get deeper from here.