Reality Bites: The Chip and Energy Crunch That Nobody Wants to Talk About
The AI hype train is about to run out of track. At the Milken Institute Global Conference, five insiders from across the AI supply chain delivered a sobering dose of reality. ASML CEO Christophe Fouquet dropped the bluntest truth: despite a ‘huge acceleration’ in chip manufacturing, the market will be ‘supply limited’ for the next two to five years. That means Google, Microsoft, and Meta won’t get all the chips they’re paying for. Period. Google Cloud’s Francis deSouza tried to spin the $460 billion backlog as a success story, but it’s really a confession: demand is outpacing physical production capacity by a staggering margin. The industry is building castles on a foundation of sand.
Meanwhile, the energy problem is even more delusional. DeSouza confirmed Google is exploring orbital data centers as a serious solution. Space data centers. Because apparently building more terrestrial power plants is too hard, so let’s launch the problem into orbit where heat dissipation becomes a nightmare. The arrogance is breathtaking. DeSouza touted Google’s vertically integrated stack as an efficiency win, but even he admitted that more compute means more energy, and energy has a price. The industry is burning capital at an unprecedented rate, and nobody seems willing to admit that the laws of physics don’t care about your quarterly earnings report.
The Quiet Revolution They’re Ignoring
While the hyperscalers fight over the last scraps of silicon, a quantum physicist named Eve Bodnia is building something that actually makes sense. Her startup, Logical Intelligence, uses energy-based models (EBMs) that don’t predict the next token in a sentence. Instead, they try to understand the underlying rules of reality. Her largest model runs at just 200 million parameters, compared to the hundreds of billions in leading LLMs, and she claims it runs thousands of times faster. More importantly, it can update its knowledge without retraining from scratch. This is a direct challenge to the dominant paradigm, and the fact that Yann LeCun signed on as founding chair should make everyone in the LLM camp nervous. The AI establishment has bet everything on scale, and Bodnia is quietly suggesting that architecture, not size, is the real bottleneck. It’s an inconvenient truth for companies that have already spent billions on GPU clusters.
Sovereignty and the Physical AI Trap
Applied Intuition’s Qasar Younis dropped the most politically charged observation of the night: physical AI is fundamentally different from digital AI because it cannot escape national borders. Autonomous vehicles, defense drones, mining equipment, and agricultural machines operate in the real world, and every country is saying the same thing: ‘We don’t want this intelligence in our borders, controlled by another country.’ Younis pointed out that fewer nations can field a robotaxi than possess nuclear weapons. That’s not a flex. That’s a crisis. Meanwhile, ASML’s Fouquet reminded everyone that China, despite the DeepSeek panic earlier this year, still can’t make advanced chips without EUV lithography. The US has the data, the computing power, the chips, and the talent. China has a thriving model ecosystem built on constrained hardware. The geopolitical chess game is getting more complex, and the AI industry’s leadership seems utterly unprepared for the fragmentation that’s coming.
Source: Techcrunch
