There’s a lot of speculation and wishcasting here, but it overlooks a key point: when the AI hype cycle cools, we’ll be left with a huge oversupply of specialized hardware and production capacity that nobody needs, and that will permanently reshape the market.
To be clear, this isn’t an argument that AI itself is a bubble, it’s the centralized, rent-seeking interpretation of AI that’s inflating valuations and destroying real value. AI, like programming, is fundamentally decentralized. Picture how absurd it would sound to talk about “Java compute” or “Python compute” and call for trillion-dollar investments. Centralized compute for AI is that kind of low-value nonsense.
Most meaningful AI workloads belong at the edge, and that trend will only accelerate. Consumer-facing AI doesn’t require the same centralization that powers large corporate websites, which are essentially digital versions of a Walmart store or Disney theme park. AI is closer to a web browser than to a website: pretty much nobody (never say never) would willingly run their browser on someone else’s machine. That mismatch is why the centralized AI bubble will burst, and when it does, hardware supply and market dynamics will shift back toward consumers. It’s only a matter of time.