What the AI Bubble Discourse Gets Wrong
Everyone has a hot take on whether AI is a bubble. Most of them are wrong in the same way. Here's a more honest read on what's inflated, what's real, and what matters.
The AI bubble debate is almost entirely conducted by people who've never actually deployed an AI system in production. The critics are mostly right about the valuations. They're mostly wrong about the technology.
Here's the answer up front: there is a real bubble in AI company valuations. There is also real, durable technology change happening. These two things are not contradictory, and most of the discourse treats them as if they are. "AI is a bubble" and "AI is transformative" can both be true at the same time — they're describing different things.
what's actually inflated
The numbers are real, and they're large. In Q1 2026, investors poured $300 billion into startups globally — with AI capturing 80% of it ($242B). Four of the five largest venture rounds in history were closed in Q1 alone: OpenAI raised $122B, Anthropic $30B (reaching a $965B valuation by mid-year, overtaking OpenAI in private market value), xAI $20B, Waymo $16B. Seed-stage AI companies are valued 42% higher than non-AI peers at the same stage. North American deal count fell 26% year-over-year while dollars invested surged 190% — fewer companies, much larger checks. That concentration pattern is what bubbles look like from inside them.
The assumption that AI productivity gains will translate directly into corporate profit at the rate being modeled is optimistic. Productivity gains often get competed away rather than captured as profit — they lower prices, improve quality, or expand labor force output without a corresponding revenue windfall.
The expectation that every category of knowledge work will be transformed on the same timeline is probably wrong. Some sectors will move fast; others will move very slowly, for regulatory, cultural, or structural reasons that the technology doesn't solve.
what's not a bubble
The underlying capability advances are real. The models are genuinely more capable than they were two years ago, across a broad range of tasks. This is not marketing. The improvements are measurable and the trajectory is consistent.
The economic value of specific AI applications is real and compounding. Businesses using AI well for the right applications are seeing genuine efficiency gains that aren't going away when the hype cycle moderates.
The infrastructure buildout — compute, data centers, power — is massive capital expenditure by sophisticated actors with long time horizons. These are not dot-com-era speculation bets. They are real assets being built for anticipated demand. Whether the demand materializes on schedule is uncertain; the infrastructure will remain.
the mistake both sides make
AI pessimists point at inflated valuations and conclude the technology is overhyped. This conflates the financial market's pricing of AI with the technology itself. Markets overprice and underprice things; that doesn't change what the technology does.
AI optimists point at genuine capability progress and conclude the market valuations are justified. This conflates what the technology can do with what the business models built on it can capture. Most of the value AI creates will not be captured by AI companies — it will be captured by the industries that deploy AI effectively.
The honest read: we are in a period of genuine technological transition with an overlaid financial bubble. The bubble will correct. The technology won't undo itself.
what this means for builders
If you're building on AI infrastructure, the bubble matters for funding and valuation but less for whether the technology works. The products that solve real problems will still be valuable after the correction. The products that rode the hype without solving real problems will be in trouble.
The builders I'd want to be are the ones who: are not dependent on the bubble continuing (i.e., don't need constant re-up on venture capital to survive), are building on capabilities that were real before the hype and will be real after, and are solving problems that exist independent of whether anyone thinks AI is transformative.
From my own bench
I'm building with AI tools every day. The capabilities are real. My costs are real. My infrastructure needs are real. None of that changes based on what the stock market does with AI company valuations.
The bubble affects the landscape I build in — plenty of well-funded competitors — and the noise-to-signal ratio in the space. It doesn't affect whether the underlying tools work.
The bottom line
"AI bubble" and "AI is changing everything" are not mutually exclusive. Some of the valuations will correct. The technology will continue to advance. The businesses that solve real problems will survive the correction; the ones that were riding narrative won't.
This is what every major technology transition looks like from the inside. The hype is real, the crash is coming in some sectors, and the technology isn't going back in the box.
— Dru Edwards