Will AI inbreeding pop the everything bubble?

Entire new economies and digital infrastructure are being built on the promise of artificial intelligence, but even Elon Musk’s Grok has begun warning that firms like OpenAI are running out of high-quality human data, leaving only AI-generated material to feed back into future models.

Some researchers have dubbed this phenomenon “AI inbreeding,” formally known as model collapse. As Oxford and Cambridge scientists describe it, model collapse occurs when repeated training on AI-generated data causes continuous degradation in performance, with models losing diversity, accuracy and realism over time.

Analysts now warn that the supply of usable human text could be exhausted by 2026–2032, forcing model developers to rely increasingly on synthetic data.

The estimate is based on the available stock of human-generated public text of roughly 300 trillion tokens, where a token represents a fragment of language used by AI models to process and generate text.

Although there are ways to mitigate the problem of so-called AI inbreeding, “The risk is real and growing,” Grok responded when asked about the phenomenon.

While the technical details may sound abstract, the implications for investors are real: if newer models deliver diminishing returns, the AI growth narrative could falter.

If newer models deliver diminishing returns, the AI growth story could lose momentum. Valuations assume constant progress, so if utility declines, corporate adoption and revenue expectations across the sector could follow.

AI bubble fears are real

While many analysts have warned that the AI boom may be overinflating Big Tech valuations, especially since the scale of infrastructure build-out has yet to match real demand, a decline in the utility of AI applications could spell trouble for investors who have priced AI stocks for perfection.

This could leave the Magnificent Seven stocks, which together have pledged hundreds of billions of dollars in AI-related capital expenditures this year alone, at significant risk.

However, the potential fallout extends well beyond the mega-caps. According to CB Insights, more than 1,300 AI startups are now valued at $100 million or more, including nearly 500 so-called “unicorns” worth at least $1 billion.

The surge has drawn comparisons to both the dot-com bubble and the 2008 financial crisis.

“We point out that the share of the economy devoted to AI investment is nearly a third greater than the share of the economy devoted to internet related investments back during the dotcom bubble,” said Jared Bernstein, former chair of the Council of Economic Advisers under President Biden.

Meanwhile, a recent Bank of America Global Fund Manager Survey found that an “AI equity bubble” was cited as the top global risk for the first time on record.

Amid mounting market concentration and a frenzy of AI-driven dealmaking, Michael O’Rourke, chief market strategist at JonesTrading, said he “absolutely” believes the market is already in an AI bubble.