The Subprime AI Crisis Is Here — And It Looks a Lot Like 2007

In a piece that has been circulating widely among AI practitioners and investors, writer Ed Zitron lays out a stark thesis: the AI industry has structurally reproduced the subprime mortgage crisis, and the adjustment is already underway. The essay, published on his newsletter Where's Your Ed At, draws a pointed parallel between the teaser-rate mortgages that preceded the 2008 financial collapse and the below-cost AI pricing strategies that have defined the industry's growth phase.

The Teaser Rate Parallel

The subprime mortgage crisis was not primarily a story about poor borrowers. It was a story about financial products designed to look affordable at first and expensive later — adjustable rate mortgages that locked borrowers into a pattern before repricing upward. Zitron argues that OpenAI, Anthropic, Google, and the wave of AI startups built on their APIs are doing the same thing: pricing access below cost to build user dependency, while accumulating losses that eventually require a pricing correction.

The indicators are no longer theoretical. OpenAI's reported unit economics — where a $20/month subscriber can cost $65 or more in compute — are the AI-era equivalent of 0% introductory APR on a credit card that resets to 28%. Sora's video generation unit economics are, by multiple accounts, catastrophic. Claude Pro's recurring usage limits and the sudden third-party tool restrictions are, in Zitron's framework, the adjustable rate resetting.

The Lock-in Mechanism

What makes this crisis structurally similar to the mortgage crisis is the lock-in mechanism. Businesses that have built workflows, products, and team processes around specific AI tools face meaningful switching costs — not the kind that required moving a family, but the kind that requires rebuilding integrations, retraining staff, and absorbing a period of reduced productivity. The teaser pricing worked. The dependency is real.

The question Zitron's framing raises is not whether AI pricing will increase — it will — but whether the value at the higher price point will be sufficient to retain the users and businesses that got on board at the introductory rate. For AI tools that have genuinely transformed workflows, the answer may be yes. For tools where AI was the default because it was cheap rather than because it was transformative, the answer is less clear.

What Comes Next

The subprime crisis ended badly for lenders but eventually resulted in a housing market that, while scarred, continued to function. Zitron's scenario is similarly not apocalyptic — but it is a significant repricing event that will favor AI products with genuine, irreplaceable utility over those that succeeded primarily on the basis of subsidized access. Which products fall into which category is the question that will define AI valuations over the next 24 months.