The Hardware Wall: Why AI's Ambition is Outstripping Its Infrastructure
AI-generated image · Bay Street Wire
Bruce Flatt warns that the AI race is hitting a physical limit, suggesting that the industry's inability to build enough compute is forcing a necessary correction.
In the world of consumer gadgets and high-end hardware, we often talk about software optimization and algorithmic breakthroughs. But as I see it, we've hit a fundamental hardware wall. The software hype is officially outstripping the physical capacity of our world to support it.
As the Financial Post first reported, Bruce Flatt, the chief executive of Brookfield Corp., has made it clear that the artificial-intelligence race is already decelerating. The reason isn't a lack of imagination or code; it's a lack of concrete and silicon. Flatt noted that developers simply cannot build the infrastructure quickly enough to keep pace with the demands of AI companies.
During Brookfield's annual investor day on Thursday, Flatt was blunt about the gap between expectation and reality. He stated that the industry cannot build enough infrastructure, adding, “We can’t even build a fraction of what everyone thinks they need.”
This creates a fascinating tension with the public narrative coming from the top of the AI food chain. The Financial Post reports that leaders like OpenAI CEO Sam Altman and Anthropic PBC CEO Dario Amodei have recently suggested that AI labs should intentionally slow the pace of development to manage unpredictable risks and ensure models remain under human control.
However, from my perspective, Flatt is right on the money: the slowdown is already happening regardless of these ethical concerns. As Flatt put it, even if these leaders didn't choose to pivot, the sector is slowing down anyway because there is not enough compute to deliver the requirements of the industry.
The scale of the deficit is staggering. According to Brookfield's estimates, over the next decade, capital investment for AI infrastructure will need to exceed US$7 trillion.
While a forced slowdown might seem like a failure of scaling, Flatt argues that this friction is actually a positive development. In his view, the inability to build at an infinite pace will bring more discipline into the system. For those of us tracking the hardware, it's a reminder that no matter how advanced the AI becomes, it still needs a place to plug in.