Why AI failing could be good for society: The collapse of the generative AI bubble

In the frenzy to monetise the generative AI boom, one likely but underexplored scenario is being overlooked: what if AI doesn’t live up to expectations? What if it never works well enough to displace workers, companies fail to use it effectively, or most startups collapse?

Analysts already estimate that major AI players face a staggering $800 billion revenue shortfall. Despite huge investments and hype, productivity gains so far are modest and concentrated among programmers and copywriters. GenAI can do some useful things, but it has yet to transform the economy as promised.

Costly Dreams, Thin Returns

Running models like ChatGPT and Gemini requires enormous computing power and steady losses. OpenAI’s Sam Altman has joked that every polite answer costs millions, while acknowledging that even paid subscriptions lose money. Like many tech startups, AI firms are burning cash to attract users, but it’s unclear how they’ll make the business pay.

Advertising, the fallback of digital giants, may not be enough. Critics warn of the “enshittification” of AI platforms, where ads and restrictions replace free, helpful services. OpenAI is already exploring ads in ChatGPT, but whether it can cover costs is doubtful.

The Copyright Minefield

Compounding the problem, lawsuits over the use of copyrighted works threaten AI’s business model. Models have been trained using everything from novels to online posts without consent. In one example, an AI could reproduce nearly half of the first Harry Potter book from memory.

Settlements with creators could balloon. Anthropic proposed paying authors $3,000 per book to use their works, a deal valued at $1.5 billion, but courts rejected the plan. The risk of mounting legal costs could quickly erode lofty valuations.

Open-Source Shake-Up

Adding further pressure, open-source AI models are flourishing. Meta’s Llama and China’s DeepSeek are free to use and “good enough” alternatives. DeepSeek even caused AI stock prices to tumble with its release. These tools weaken the commercial case for high-priced models, raising doubts for investors.

If open models become widespread, firms like OpenAI and Google may struggle to justify their premium services. Even if lawsuits threaten them, open systems are hard to shut down once distributed.

Can AI Be Monetised at All?

The core challenge might be that generative AI’s foundation, global shared knowledge, is essentially unknowable. These systems rely on collective intellectual labour, making it difficult to attach a price tag to their outputs. That paradox undermines the very attempt to turn AI into a profitable commodity.

A Future That’s “Good Enough”

If generative AI can’t sustain profits, the outcomes may be mixed. Creators may lose lucrative contracts, and progress could plateau. Yet, users may gain access to free, useful tools while being spared from another hype-driven tech bubble.

In fact, the technology becoming less profitable for corporations could also mean less concentrated power in big tech’s hands. If the AI empire collapses under its own financial weight, the balance sheets, not regulators, may be where its downfall is decided.

With inputs from PTI