ASML CEO Christophe Fouquet told the Milken Global Conference in Beverly Hills this week that the AI chip market will remain supply-limited for the next two to five years, even with what he called a 'huge acceleration of chips manufacturing.' Sharing the stage, Google Cloud COO Francis deSouza said the unit's revenue crossed $20 billion last quarter at 63% growth, while its committed-but-undelivered backlog nearly doubled in a single quarter, from $250 billion to $460 billion. The implication is that Google, Microsoft, Amazon and Meta are not going to receive all the chips they have already paid for.
ASML holds a monopoly on the extreme ultraviolet lithography machines required to manufacture leading-edge chips, which makes Fouquet's read on capacity the most authoritative one in the room. 'The demand is real,' deSouza said, calmly, while sitting on the largest backlog disclosure in cloud history.
The other panelists were Qasar Younis, co-founder and CEO of Applied Intuition, a $15 billion physical-AI company that has expanded from automotive simulation into defense; Dimitry Shevelenko, chief business officer of Perplexity; and Eve Bodnia, a quantum physicist whose startup Logical Intelligence is trying to replace the transformer paradigm. Yan LeCun, Meta's former chief AI scientist, signed on as Logical Intelligence's founding technical board chair earlier this year.
Key facts
- 01ASML CEO Christophe Fouquet predicts the chip market will stay supply-limited for two to five years despite a 'huge acceleration' in manufacturing.
- 02Google Cloud revenue crossed $20B last quarter at 63% growth; backlog nearly doubled from $250B to $460B in one quarter.
- 03Applied Intuition, valued at $15B, says physical-AI training is bottlenecked by real-world data, not silicon.
- 04Logical Intelligence's largest energy-based model runs at 200M parameters versus hundreds of billions for leading LLMs, with claimed thousands-of-times-faster inference.
- 05Perplexity's Comet agent requires plan approval before acting; CBO Dimitry Shevelenko also sits on the board of 180-year-old Lazard.
Younis described a different bottleneck. Applied Intuition builds autonomy systems for cars, trucks, drones, mining equipment and defense vehicles, and the constraint is data, not silicon. 'You have to find it from the real world,' he said. 'There will be a long time before you can fully train models that run on the physical world synthetically.'
“Google Cloud's backlog nearly doubled in a single quarter, from $250 billion to $460 billion, while revenue crossed $20 billion last quarter at 63% growth.”— Jaeden Schafer
Energy is the constraint sitting behind the chip constraint. DeSouza confirmed that Google is treating orbital data centers as a serious option because, as he put it, 'you get access to more abundant energy.' He conceded the engineering is brutal — vacuum eliminates convection, leaving radiation as the only way to shed heat — but the company is studying it anyway.
DeSouza's broader pitch was vertical integration. Google co-engineers its TPUs, models and agents, and he argued the resulting flops-per-watt advantage cannot be matched by buyers of off-the-shelf parts. 'Running Gemini on TPUs is much more energy efficient than any other configuration,' he said. Fouquet's framing was blunter: 'Nothing can be priceless.' More compute means more energy, and energy has a price that eventually shows up on a balance sheet.
Bodnia is betting the entire stack is wrong. Logical Intelligence builds energy-based models, or EBMs, which try to learn the rules underlying data rather than predict the next token in a sequence. Her largest model runs at 200 million parameters versus the hundreds of billions in frontier LLMs, and she claims it runs thousands of times faster while updating its knowledge incrementally rather than requiring retraining from scratch.
'Language is a user interface between my brain and yours,' Bodnia said. 'The reasoning itself is not attached to any language.' She argued EBMs are a more natural fit for chip design and robotics, where systems need to grasp physical rules rather than linguistic patterns. The argument lands at a moment when the industry is openly questioning whether scale alone is enough.
Shevelenko spent his time reframing Perplexity from a search product into a 'digital worker.' Its newest offering, Perplexity Computer, is pitched as staff a knowledge worker directs rather than a tool the worker uses. 'Every day you wake up and you have a hundred staff on your team,' he said. Comet, Perplexity's computer-use agent, presents a plan and asks for approval before acting — friction that some users find annoying and that Shevelenko, who joined the board of 180-year-old Lazard, defended as basic security hygiene.
Younis offered the panel's sharpest geopolitical point: physical AI and national sovereignty are entangled in a way 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,' he said, noting that fewer nations can field a robotaxi than possess nuclear weapons. Fouquet added that China's progress, including DeepSeek's release earlier this year, is real but constrained below the model layer because Chinese fabs lack EUV access.
The optimism on display has limits the speakers acknowledged on stage. A $460 billion backlog is also a $460 billion delivery risk if Fouquet's two-to-five-year supply ceiling holds. Bodnia's EBM claims — 200 million parameters, thousands of times faster — remain unverified by independent benchmarks, and physical-AI export controls of the sort Younis described are easier to announce than to enforce.
What this panel actually documents is the moment the AI economy stopped being a software story. The bottlenecks are now lithography machines, gigawatts, real-world driving miles, and which government will let which robot cross which border. The hyperscalers can keep signing $20-billion-a-quarter contracts, but the wheels Fouquet was politely warning about are mechanical, electrical, and political — and none of them respond to a higher capex number on the next earnings call.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
