Five architects of the AI economy explain where the wheels are coming off

Earlier this week, five people involved with every layer of the AI ​​supply chain sat down at the Milken Global Conference in Beverly Hills, where they spoke to this editor about everything from chip shortages to orbital data centers to the possibility that the entire architecture underlying the technology is flawed.

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 investments in company 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, the 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 that most of the AI ​​industry takes for granted at her startup, Logical Intelligence. (Meta’s former chief AI scientist, Yan LeCun, signed on as the founding chairman of the technical research council earlier this year.)

Here’s what the five had to say:

The bottlenecks are real

The AI ​​boom is hitting hard physical limits, and the limitations start further down the stack than many may realize. Fouquet was the first to say this, describing a “tremendous acceleration in chip production,” while expressing his “strong belief” that despite all those efforts, “the market will have a limited supply for the next two, three, maybe five years,” meaning that the hyperscalers – Google, Microsoft, Amazon, Meta – won’t get all the chips they pay for, period.

DeSouza emphasized how big (and how quickly growing) this problem is, reminding the audience that Google Cloud’s revenue surpassed $20 billion last quarter, a growth of 63%, while the backlog (the committed but unrealized revenue) nearly doubled in one quarter, from $250 billion to $460 billion. “The demand is real,” he said with impressive calm.

For Younis, the limitation mainly comes from elsewhere. Applied Intuition is building autonomy systems for cars, trucks, drones, mining equipment and defense vehicles, and its bottleneck isn’t silicon; it’s the data you can only collect by sending machines into the real world and seeing what happens. “You have to find it in the real world,” he said, and no form of synthetic simulation can completely bridge that gap. “It will take a long time before you can fully train models that run synthetically on the physical world.”

The energy problem is also real

If chips are the first bottleneck, energy lurks behind them. 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. Even in orbit, it is of course not easy. DeSouza has observed that space is a vacuum, which eliminates convection, leaving radiation as the only way to release heat to the environment (a much slower and more difficult to develop process than the air and liquid cooling systems that data centers rely on today). But the company still considers it a legitimate path.

The deeper argument De Souza made was, somewhat surprisingly, about efficiency through integration. Google’s strategy to co-engineer its entire AI stack — from custom TPU chips to models and agents — is paying off in flops per watt (more computing power 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 power efficient than any other configuration,” because chip designers know what goes into the model before it ships, he said.

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

A different kind of intelligence

While the rest of the industry debates scale, architecture, and inference efficiency within the grand 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 underlying data, in a way she says is closer to how the human brain actually works. “Language is a user interface between my brain and yours,” she said. “The reasoning itself is not linked to any language.”

Her largest model includes 200 million parameters – compared to the 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 the data changes, rather than having to be retrained.

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 most natural solution. “When you drive, you’re not looking 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 that will likely attract more attention in the coming months as the AI ​​field begins to question whether scale alone is enough.

Agents, guardrails and trust

Shevelenko spent much of the conversation explaining how Perplexity evolved from a search product to something it now calls a “digital worker.” Perplexity Computer, its newest offering, is designed not as a tool that 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 are you going to do to make the best 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 has access to, but also whether these permissions are read-only or read-write – a distinction that is extremely important when agents act within enterprise systems. When Comet, Perplexity’s computer usage agent, takes action on a user’s behalf, it presents a plan and first asks for approval. Some users find the friction annoying, Shevelenko said, but he said he considers it essential, especially after joining Lazard’s board of directors, where he said he unexpectedly sympathized with the conservative instincts of a CISO protecting a 180-year-old brand built entirely on customer trust. “Granularity is the foundation of good safety hygiene,” he said.

Sovereignty, not just security

Younis raised perhaps the panel’s most geopolitically charged observation, namely that physical AI and national sovereignty are intertwined in a way that purely digital AI never has been.

The Internet initially spread as an American technology and only faced backlash at the application layer – the Ubers and DoorDashes – when the offline consequences became visible. Physical AI is different. Autonomous vehicles, defense drones, mining equipment, agricultural machinery – these are manifesting in the real world in ways that governments cannot ignore, raising questions about security, data collection and who ultimately controls the systems operating within a country’s borders. “Almost consistently, every country says: we don’t want this intelligence in physical form at our borders, controlled by another country.” There are currently fewer countries that could deploy a robotaxi, he told the crowd, than possess nuclear weapons.

Fouquet put it a little differently. The progress in AI in China is real – the release of DeepSeek earlier this year caused panic in parts of the industry – but that progress is limited below the model layer. Without access to EUV lithography, Chinese chipmakers cannot produce the most advanced semiconductors, and models built on older hardware operate at a greater disadvantage no matter how good the software becomes. “Today in the United States you have the data, the computer access, the chips and the talent. China is doing very well at the top of the pile, but is missing some elements below that,” Fouquet said.

The generation question

Towards the end of our panel, an audience member asked the obvious, uncomfortable question: Is this all going to impact the next generation’s ability to think critically?

The answers were optimistic, as you would expect from people who have staked their careers on this technology. DeSouza immediately pointed out the magnitude of the problems that humanity could ultimately tackle with more powerful tools. Think of neurological diseases whose biological mechanisms we do not yet understand, the removal of greenhouse gases and the network infrastructure that has been postponed for decades. “This should take us to the next level of creativity,” he said.

Shevelenko made a more pragmatic point: The entry-level job may be disappearing, but the ability to launch something on your own has never been more accessible. “[For] anyone who has Perplexity Computer. . . the limitation is your own curiosity and freedom of choice.”

Younis made the sharpest distinction between knowledge work and physical labor. 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 those domains, physical AI does not displace willing workers. It fills a void that already exists and that only seems to get bigger from here.